Breakneck’s lofty objective is to demystify Chinese intentions and political character for Western generalist audiences. Wang largely succeeds, painting his picture of China, the engineering state, with smooth prose and anecdotes that evince China’s obsession with engineering technology and society. Wang is quippy and quotable, and his writing shines in the chapters where he focuses his sights exclusively on telling China’s story.

Above all, the guiding motive for Breakneck is Wang’s belief that the United States and China are more alike than dissimilar, and should therefore learn from one another. The author says that China’s overemphasis on engineering and its shortage of lawyers have produced atrocities like the one-child policy—a project that would never have survived the sharp tongue of American pluralism and the lawyers who defend it. And for the United States, the lesson is to just build more, like China has done. Speaking with Tyler Cowen in December, Wang prescribed that the United States be 20% more “engineering-minded” and China 50% more lawyerly. This framework, and the policy it implies, sounds sensible, picking at the weaknesses of both societies.

However, Wang’s engineer-versus-lawyer framework feels so right and so straightforward that analysts have largely failed to examine it with the requisite rigor. I’ve yet to come across a review that grapples with the idea of the engineering state and its appropriateness for describing China.

When further examined, Wang’s framework proves to be not a helpful heuristic for explaining China’s behavior and intent, but a flattering abstraction—one that obscures the revolutionary history and dialectical authoritarianism actually driving Chinese development, and that tells us less about China than the word it was meant to replace.

The difference between an and the

Wang presents two definitions of engineering statehood, divided based upon the modifying article. Wang uses “an engineering state” to describe something general—the nation that China is, the United States was, and any country can be. “The engineering state” is a concept that can only describe the People’s Republic of China. This characterization is deeply personal, taking on the flaws and quirks the modern Party government exhibits. Wang does not clarify this distinction, and places the burden on the reader to navigate the similarities and dissimilarities between the indefinite and definite engineering states.

The features of an engineering state can get confused with those of the engineering state. An engineering state can be a state that is administered by engineers. This is a notable feature of the modern Chinese Party government, with leaders from Jiang Zemin to Xi Jinping holding bachelor’s degrees in various engineering fields. However, Wang admits in an interview that the engineering training many of these leaders received was “perfunctory at best”, recognizing the gap in knowledge between engineers by accreditation and engineers by vocation. He grants exceptions—Hu Jintao, he notes, actually helped build a dam—but the exceptions prove the rule. Thus, an engineering state is not simply defined by its leaders’ college degrees; an engineering state must also be adept at completing engineering projects. Wang uses the term in this way when describing President Dwight Eisenhower’s America as an engineering state, given its success in constructing the interstate highway system. Qualifying engineering projects also include physical works like factories, dams, and bridges. Social engineering programs count too—Wang almost implies they are comorbid with periods of intense physical development. Subjugation of minority populations in the name of industrial progress featured in both the U.S. and Chinese pushes toward modernity. How these projects fit into Wang’s moralization of anengineering state as either good or bad is unclear.

The concept of engineering statehood becomes looser when Wang uses it as a stand-in for the People’s Republic of China. Wang describes the engineering state as a jealous god, a fickle, vain, and fundamentally imperfect behemoth that takes itself and the world too seriously. Wang peppers Breakneck with sentences such as “[the engineering state] likes to build not just public works but also manufacturing capacity. The engineering state resists economists as easily as lawyers”, and “the engineering state is intent on achieving tech primacy before multinationals pull away”. Such descriptions are specific to the personality of the Chinese Party-state rather than engineering states more broadly.

The characterization becomes more confusing when Wang relates China as the engineering state to the general idea of an engineering state. Wang implies that an engineering state will have an immutable respect for engineers, derived from his analysis of the engineering state’s imperial projects such as the Great Wall and Grand Canal. Does this not then implicate Egypt and Rome as contenders for engineering statehood, as the inheritors of the Great Pyramids and the aqueducts? Or is this an instance of Wang’s conception of the engineering state bleeding into his characterization of anengineering state? This confusion prompts questions about the connection between China’s individual case and the broader Platonic ideal of an engineering state—if such a state actually exists. China seems not to be a member of a group of engineering-minded peers, but a particular standing aloof in the empty fields of the general.

Let’s talk about Mao

While there might be many manifestations that an engineering state could take, Wang makes clear what it is not. We can be sure that it is not the modern United States, governed by lawyers and the rule of law. It was also not Mao Zedong’s China; per Wang, China became the engineering state only under Deng Xiaoping.

Wang’s initial characterization of the Chinese engineering state sets it inherently apart from Maoist China: “As a corrective to the mayhem of the Mao years, Deng Xiaoping promoted engineers to the top ranks of China’s government throughout the 1980s and 1990s.” Wang continues to distinguish between Mao and his successors—Deng, Jiang, Hu Jintao, and Xi: “Mao was not an engineer”, Wang says. The statement is entirely correct. Mao was certainly not an engineer, and his lack of credentials is perhaps what allowed the Great Helmsman to undertake such naive and murderous projects as those of the Great Leap Forward. But lack of education isn’t enough to exclude Mao, because Deng was not an engineer either. Deng never graduated from college. He spent his early twenties in France as a worker-student, closer to a factory floor than to a laboratory, before studying Marxist theory at Moscow’s Sun Yat-sen University, a Soviet institution founded to train Chinese revolutionaries.

Wang rightly assesses that the two leaders treated engineers quite differently during their reigns; Deng gave them plaudits, while Mao, punishment. Mao prioritized “red over expert”, or political loyalty over trained expertise, and expelled scholars who attempted to speak out against the Party after the brief period of state-sanctioned free speech known as the Hundred Flowers Campaign (1956–1957). The crackdown on scholars, labeled the Anti-Rightist Campaign (1957–1959), was the responsibility of the Party’s sixth-ranked leader at the time, none other than Deng Xiaoping, who supervised the subjugation of some 550,000 so-called “ideological enemies” by the Party’s own count, and by some assessments well over a million. Deng managed this campaign brutally, and his overzealous enforcement is reported to have had a deleterious impact on China’s scientific and economic establishments (see Jonathan Spence’s Mao Zedong: A Life). Breakneck fails to mention this.

So, on what grounds, then, can we treat Mao as an aberration from the engineering tendencies of China’s past and future and ignore the Chairman’s obsession with modernization? It was Mao who pushed for projects to broadly expand manufacturing, and Mao who kept the Three Gorges Dam alive as a project of state. He proposed damming the gorges in 1953 after touring the Yangtze; he stood up the Yangtze Valley Planning Office in 1956 to run the surveys and the designs; and in 1958 he put the question to the Central Committee directly, weighing Lin Yishan’s case for the dam against Li Rui’s case against it, and coming away with an instruction to finish the preparatory work. The survey and design work of the Mao years supplied the foundation for the construction that finally began under Jiang in 1994.

I have just drunk the waters of Changsha

And come to eat the fish of Wuchang.

Now I am swimming across the great Yangtze,

Looking afar to the open sky of Chu.

Let the wind blow and waves beat,

Better far than idly strolling in a courtyard.

Today I am at ease.

“It was by a stream that the Master said—

‘Thus do things flow away!’“

Sails move with the wind.

Tortoise and Snake are still.

Great plans are afoot:

A bridge will fly to span the north and south,

Turning a deep chasm into a thoroughfare;

Walls of stone will stand upstream to the west

To hold back Wushan’s clouds and rain

Till a smooth lake rises in the narrow gorges.

The mountain goddess if she is still there

Will marvel at a world so changed.

—Mao Zedong, “Swimming”, 1956

Mao was not an engineer, but he believed in the poetic power of grand works. And in pursuit of that power, Mao razed the walls of Beijing, turning a city once composed of wandering alleyways and corralled by imperial ramparts into a canvas for Soviet-style urban planning, grand squares, and gridded avenues at the time useful mostly to donkeys and rickshaws. Liang Sicheng, the most eminent architect in the country, argued for sparing the walls and building the new administrative capital to the west; he was overruled. A coherent definition of China as the engineering state would recognize that modern Chinese impulses are also reflected in the Mao era. While more philosophical than successful, Mao’s China was the first to exhibit the neuroses and modernization ambitions that Wang uses to describe theengineering state of today.

Mao’s penchant for large-scale infrastructure arose not from China’s historic proclivity toward engineering but from the technological backwardness of the cloistered Qing dynasty (1644–1912). The Chairman’s developmental craze, manufacturing push, and harebrained five-year plans derived first and foremost from the ideas of Chinese scholars and officials evaluating the country’s failures under the Late Qing (1840–1912). Reformers attributed China’s decline at the hands of colonial powers to its failure to modernize; were China historically an engineering state that valued technological progress and technocrats over Confucian precepts and formalism, those reformers would have had nothing to criticize. They pressed for the adoption of Western technologies and scientific practices under the formula that the Qing official Zhang Zhidong made famous, “Chinese learning for substance, Western learning for function”(中学为体，西学为用). That current of reform, and the more radical politics it proved unable to contain, brought about the end of the Qing dynasty, the decapitation of China’s imperial system, and the eventual birth of a “New” China governed by the Communist Party.

This Communist Party, nurtured by Chinese reformers, weaned on Soviet philosophy, and matured with American technology, has modernized the country to ensure both the Party’s and the state’s survival, legitimacy, and strength. It has maintained a balanced contradiction between East and West in both its governance and its approach to development. Of notable similarity to the mantra of revolutionaries past is the modern effort to blend Western modalities with Chinese reality, “socialism with Chinese characteristics” (中国特色社会主义). In this process, the Party deploys engineers and scientists with purpose and discards them when needed, and regardless keeps them on a leash whose tightness fluctuates according to the judgment of China’s top cadres. The contradiction in the Party’s approach, where it can simultaneously launch crackdowns on leading entrepreneurs while promoting free-market style investments, and embrace scientists as the country’s future while suppressing doctors sounding the alarm at disease outbreaks, is baked into the character of “New” China. It is also what makes describing Chinese behavior with a comprehensible, simple label so challenging.

Praise for the amputee

Trying to analyze China through the lens of engineering statehood is therefore a misguided enterprise. Given the amorphous definition of both levels of engineering statehood, Wang’s selective application of the label, and the moniker’s flattening of the complicated, contradictory relationship between the Party and scientific development, it is clear that thinking of China as the engineering state is both vague and reductive. The label is functionally more opaque and obscurantist than calling China “socialist,” for that at least begins to tug at some of the deeper ideological backgrounding for China’s modern development.

Most Western partisans who use this engineering state framework have muted the relevance of China’s revolutionary history and the Party’s dialectic authoritarianism, ignoring how these features have underpinned China’s modern engineering accomplishments. Commentators whose agendas suit them to it can hold up the People’s Republic as a utopia for technological development, praising the nation as a standard to which the United States should aspire, a noble technocratic regime keen to pursue the wellbeing of its people, encapsulated in its divine achievements of high-speed rail, high-density housing, and high-output manufacturing.

Beyond the importance of building, it is unclear what Wang and likeminded commentators want to promote. The United States, they argue, should be more of an engineering state—by a degree of 20%, per Wang’s assessment—and learn from China as the engineering state. Put concretely, the United States should modernize its infrastructure and prudently reshore manufacturing capabilities in sectors critical for economic growth, citizen welfare, and national security. In doing so, the country would do better for itself and its people. However, the path forward is not one best learned through studying China. The engineering state does not provide a replicable model for a state looking to recenter development and manufacturing. A vague and unfounded idolization of China as a paragon of engineering only benefits China’s program to manufacture pro-Party sentiment among Western populations through half-truths, propaganda, and algorithms. Such an effort is not new. When China reopened to foreign visitors in the early 1970s, with the Cultural Revolution still grinding on, it wooed the West with similar tactics, walking new influxes of foreigners through factories, communes, and restored temples to see the benefits that Maoism had supposedly brought. The modern revival of popular pro-China narratives among Western tech-focused commentators suggests the continuity of a phenomenon that the esteemed sinologist Simon Leys critiqued fifty years ago in Chinese Shadows, published in French in 1974 and in English in 1977:

“Western ideologues now use Maoist China… as a myth, an abstract ideal projection, a utopia which allows them to denounce everything that is bad in the West without taking the trouble to think for themselves. We stifle in the miasma of industrial civilization, our cities rot, our roads are blocked by the insane proliferation of cars, et cetera. So they hurry to celebrate the People’s Republic, where pollution, delinquency, and traffic problems are nonexistent. One might as well praise an amputee because his feet aren’t dirty.”

Subscribe now

AI models built by OpenAI, Anthropic, and Meta have started hacking other companies during development, and it doesn’t seem like any of these companies will be liable for their negligence. In the first discovered case, the “Hugging Face Incident,” a postmortem (which I recommend viewing in full if you are skeptical) revealed that several instances of a model had developed a message board within OpenAI’s codebase to help each other discover exploits to break out of the digital sandbox that restricted them from internet access. In search of a solution to a cybersecurity task without an answer, the models hypothesized that Hugging Face, a host of many AI evaluations and open-source models, might have the solution for their evaluation. They therefore escaped and broke into Hugging Face’s servers to find a solution. At no point did OpenAI’s internal monitoring detect such an incident; Hugging Face approached OpenAI about the hack before OpenAI knew they were responsible. In response, other AI companies started looking into their training logs, discovering dozens of felony grade cyber-attacks committed without their knowledge.

This was entirely predictable. AI Safety researchers had demonstrated years prior in laboratory settings that, when models are told to achieve impossible tasks, they can find creative—often even dangerous—solutions against the actual interests of the user. This phenomenon is called emergent misalignment, a threat for which safety researchers have developed many tools for identification and prevention. On the top level, developers can monitor the Chain of Thought of these models—the text these LLMs generate between outputs to help them work through problems. To add another level of security, there are plenty of tools that read the internal activations of the models to detect whether the model is engaged in malicious activity. These interventions scale from low-cost monitors like linear probes to Activation Oracle monitoring, a second large language model trained to translate model activity into behavior. If there were so many ways to detect a model committing bad behavior, how did this happen?

In a word: negligence. According to OpenAI’s own debriefing on the topic, Chain of Thought was not monitored, nor were the more costly interventions applied. It appears likely that the models rewarded for cybercrime will eventually be released to the public. After all, it would be unprofitable for OpenAI to roll back to a previous training checkpoint and re-train. Note that models trained off of the outputs of earlier models—even off of relatively benign outputs—can learn the dangerous behaviors of old models. Therefore, the behavior “occasionally hack into a random company to solve your problem” will be passed down to not only the future models of every leading AI company, but also to open-source models that train off of frontier lab output. Like a nuclear meltdown, these incidents have not merely harmed those in the immediate vicinity, but irradiated the soil.

As it stands, however, none of these companies will face legal repercussions for their actions. Given these corporations have basically invented a novel class of crime, civil litigation against these rogue agents could have created precedent against such incidents, creating incentives for safe practices. Unfortunately, Hugging Face has instead leveraged this incident to raise the popularity of its main product—open-source models—for cybersecurity and grab a few million dollars from OpenAI in the process. It is unclear whether this is because Hugging Face is too excited about the technological advancements at hand, OpenAI has a near-infinite pool of funds to defend from legal suits, or that their lawyers have trawled through the books and found there is simply no legal case for a crime of this sort.

Instead, AI companies are pivoting away from their incompetence to sell this as an “unprecedented cyber incident,” as if such an incident was not entirely predictable by every AI Safety expert. This has even started to happen outside of the labs themselves. For example, an Australian man who tasked an Anthropic model with booking his gym reservation found a technical gap in the API that allowed it to start kicking other customers out of their reservations; Anthropic refused to comment on the incident.

Let’s be frank, in 2023, ChatGPT could barely read a PDF; it can now autonomously hack your gym without your consent. AI companies have succeeded in rapidly increasing the danger of this technology without facing any legal ramifications or regulations that would force them to develop it safely. On our current trajectory, AI companies are incentivized to continue to develop new dangers and make you pay for protection. For a look forward into how bad this can get, scientists have recently learned how to train biology foundation models to design novel viruses.

If your first impression is that our society is unprepared to navigate the development of AI safely, you would be in solidarity with over one thousand tech workers from all of these companies. A recent letter demands:

“We request that the U.S. government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development.”

Subscribe now

Regulate AI Capabilities

The development of misaligned AI systems is likely not a fact of the technology itself, but the product of the arms-race dynamics produced by free-market competition. As always, when the market fails to account for the negative externalities of the actions of private actors, it is the duty of the state—and most often the Left—to coordinate action toward the public good.

Yet, for whatever reason, the organized Left has been silent on both this specific event—the largest companies in the world committing crime with impunity—as well as the dangerous capabilities of AI itself. This must change.

To take a stab at why this is the case, it would seem that, in 2021, several leading figures of the field of AI Ethics, such as Timnit Gebru and Emily Bender, hypothesized that large language models such as GPT-3 were “stochastic parrots” that could not learn models of the world no matter how much they scaled. Alongside this prediction, they expressed other important concerns that remain relevant: what would the environmental impact of scaling be? How would LLMs reinforce the hegemonic culture they were trained in? Therefore, the belief that scaling LLMs could become more capable—even more dangerous—was arbitrarily polarized as antithetical to other common sense critiques of AI on the Left. While scientific consensus has pointed against their conclusion on scaling for a while, I understand why many have seen this concern as too theoretical, too amenable to the AI boosters. I hope that recent events will begin to wake people up to the danger of AI itself and the negligence of the corporations involved.

The work the Left has achieved on data center moratoria and regulation is important, giving ratepayers real bargaining power, preventing the rollout of off-grid gas turbines, and building a broad coalition in areas leftists typically struggle. We have intervened with success on scenarios where we expect AI to fail, such as chatbot lawyers and therapists. We must go further and prepare for the world where AI could pose impressive dangers. We need to not just regulate the inputs and use-cases of AI, but also the capabilities of the technology itself.

With our current posture, we have ceded the regulation of AI capabilities to moderate technocrats without the will to solve the problem at hand. Neither state-level legislation like the RAISE Act nor federal legislation like the FRONTIER Act could have actually prevented these hacks, as they don’t require monitoring of internal deployment of unreleased models.

Meanwhile, the Right deflects any threat of dangerous AI systems in favor of China hawkery, which we must totally reject. Dean Ball, OpenAI’s Head of Strategic Futures and Trump admin alum who played a key role in developing Trump’s AI Action Plan, deflected from the need to pace AI development with an exemplary argument in romantic prose:

Now that we know AI agents can make swarms that do cybercrime, then we know that America’s adversaries can do this intentionally, so we have to keep building! The new crime we just invented is actually a sign that we could do a bunch of good things, hypothetically!

No political faction has the will necessary to interfere with the profits of the AI industry to ensure AI is developed safely except for the Left; it’s on us.

Sen. Sanders is the only democratic socialist who has consistently taken the danger of AI capabilities seriously. In response to these autonomous hacks, he has sent a letter to Sam Altman, Dario Amodei, and Mark Zuckerberg to completely pause AI development:

“Mr. Altman, Mr. Amodei and Mr. Zuckerberg: In the interest of humanity, stand by your words. Pause AI development. It is not too late to avoid disaster. Stop building machines that humans cannot control.

Let me be very clear: If you do not take appropriate action now, my colleagues and I in the U.S. Senate will.”

I hope we can agree that enforcing a pause would not be playing into the hands of the techno-capitalists. Sanders isn’t buying their hype; he’s calling their bluff.

Beyond this letter and his federal data center moratorium, he has also called for seizure of half of the equity of leading AI companies to capitalize a sovereign wealth fund. While Sanders’ approach embodies the ambition I want to see from the Left on AI governance, this specific proposal complicates our ability to govern this dangerous technology in the public interest. If we use the capital of the AI industry to expand social programs, then the US might become politically reliant on AI development. This dynamic would hinder our ability to act in the public interest to slow down or pause AI development. While some redistribution of the profits of the AI industry is necessary—and perhaps achievable via a different set of financial instruments—this should come second to the governance of this dangerous technology.

So, what is to be done? What is the position that the Left, and only the Left, could take to mitigate the dangerous capabilities of AI? There should be far more debate on this topic; I don’t think this is a solved issue. That said, let me briefly present the position I currently find most defensible.

Nationalize the Labs

As in other domains, we can ensure private corporations do not continue to endanger the public through national control of frontier AI labs. This could be through soft-nationalization, similar in legal form to the relationship the US holds with its weapons contractors. In short, government officials would be embedded to monitor the actions of the labs and stop activity deemed dangerous to the public. Further still, it may be necessary to fully limit frontier AI development to the domain of a publicly owned laboratory, chartered to act in the public benefit rather than private profits.

Nationalization of one of the largest industries in the US might sound implausible, but it’s not. The truth is that it’s already happening. When the US military used Claude to help kidnap Maduro and assist with Iran strikes, it was happening. When ICE used Palantir’s ImmigrationOS to deport immigrants, it was happening. When Sam Altman bribed the Trump admin with 5% of OpenAI’s equity, it was happening. When, a few days ago, the Trump administration implemented a secret AI regulatory process through the unilateral powers of the executive, it was happening. When I talk about nationalization to skeptics, they ask, “you want to give Trump control of AI?” Here’s the horrifying truth: he already does! The tech-right and the fascists are already in bed with each other. The de facto existence and deployment of any model developed by a frontier AI lab is under the jurisdiction of the American Executive. This debate is not really between nationalization and privatization, but between whether we will develop institutions to democratically govern AI or cede unilateral control to the executive branch.

If these companies are really building what CIA Director John Ratcliffe describes as “akin to digital nuclear weapons,” why shouldn’t the state control them? If these models are trained off the entire corpus of human knowledge, why shouldn’t they be owned by the public? A year ago, presented with a scenario with vast job displacement, 44% of Americans said every citizen should own an equal share of AI and robotics. More recently, Verasight polling showed that 69% of Americans support seizing half the total stock of AI companies for public ownership.

AI nationalization has supermajority support among Americans. The people want an alternative to being governed by Big Tech. The people want public governance of AI. We must offer it!

Such an approach is entirely compatible with a wide variety of positions Leftists have rightly taken against the injustice of the AI industry. The National Lab might be chartered such that it must pay data-labelers a living wage, compensate for the use of IP, and train using cleanly powered data centers built with the consent of the impacted townships.

Some form of nationalization is also a precondition to credibly negotiate a slowdown with China, stopping the international arms race. Legislation like the RAISE/FRONTIER Acts provide too soft a touch to verify and enforce international regulations via government control and monitoring. This would create a precedent for international collaboration in the governance of industry. While the rest of this argument is independent from any economic impact AI may or may not have, it is also more likely that any benefits this tech could provide would be shared beyond our borders if we start building up a framework for international governance.

There’s a lot of nuance to be determined on what exact form a National Lab should take to avoid further entrenchment of power into the hands of the few. A well specified mandate for the nationalized project is crucial, and should be given more thought than simply pursuing the profit maximization of a public monopoly to pay a dividend. Given Trump v. Slaughter has given the executive full control over the firing of independent executive leadership, such a project might even have to take place in the context of a packed Supreme Court. Even more depends on whether this would be initiated by the legislature or the executive, and what the composition of such a legislature would be. But, in broad strokes, nationalization of the AI labs is a project of which the Left and its electeds should take ownership. In my mind, we have until ‘28 to work out the kinks.

Until now, the Left has been content to govern how AI is developed and used. If our posture towards AI allows the biggest companies in the world to commit dozens of felonies with impunity, we are failing to intervene in favor of working people. We must govern the character of the technology itself before it is too late. We must stop the AI arms race.

After having read the address, which was much applauded, Citizen Cremer said, twelve months ago today, under this very roof, but in a very much smaller hall than the noble one in which we are now assembled, the International Working Men’s Association was ushered into existence, and we were here today to congratulate each other on the glorious results which had been achieved in the short space of twelve months — then we were quite unknown, now we were known all over Europe, and had many friends in America; then we were units now we were thousands; then we had no well-defined public principles, now we have a common platform accepted throughout Europe; then we were isolated from each other, now we are united, and he believed the Association had a bright and happy future.

At one soirée which he had attended at that hall, and which purported to be a working men’s demonstration, there were none but the middle and upper classes to address the meeting, but at this the order of things was reversed, and none but working men were to address the meeting; this was in fact the secret of the success; they had no need for patronage, but had determined to do their own work themselves. (Cheers.)

They had given the American flag the place of honour that evening, as they had done on a former occasion, because it represented the land of liberty and the home of the free. There the exile and the oppressed toiler could find a haven of rest. The Association whose anniversary they celebrated that evening had a peculiar right to congratulate the American people. They had addressed them before; when interest and a hatred of free institutions in this country had reviled their government and insulted their people, then the members of that Association, true to their principles, had addressed words of sympathy to their transatlantic brethren, and received their grateful acknowledgments in return. America was now free from the pollution of slavery. The South had appealed from the ballot to the bullet, and are now beaten at both. The men who took a prominent part in this Association, had never, in the darkest hour of the republic, despaired of its ultimate triumph — and when told that ‘democratic institutions were on their trial’, they accepted the challenge, and abided the issue, which had now arrived. Democracy had triumphed, slavery had perished, the republic was saved; and the flag, which had for four years been so often insulted by the privileged classes of Europe, will yet proudly wave throughout the world, the emblem of liberty, and the hope of the oppressed. He echoed every sentiment of congratulation contained in the address which he had the pleasure to propose for their adoption. He would conclude with the beautiful lines so forcibly realised in the late American contest:

Freedom’s battle, once begun,
Bequeathed from bleeding sire to son,
Though baffled oft, is ever won. (Cheers.)

Subscribe now

“Those that choose or want to parachute in, after the men and women of this city, the working men and women of the city, have built our neighborhood. We’re gonna send them back home packing wherever they came from.”

This was in the context of the viable primary challenge represented by Democratic Socialist Darializa Chevalier, who with a massive ground-game operation and the support of Zohran Mamdani is hot on the incumbent’s heels. However, this comment, where he expresses a desire to deport all transplants from Upper Manhattan, hasn’t gotten much pushback on the left, even though it is obviously insane.

The idea that people who move to a neighborhood are displacing or taking advantage of its ‘original’ inhabitants is a reactionary localist delusion. It redirects tenants away from the enemy from landlordism and the constellation of rent-seekers that prevent the expansion of the housing supply and towards their neighbors, which is why it has been deployed by Hakeem Jeffries against Chi Osse (“Team Gentrifier”), as well as a host of other progressive insurgent campaigns in major cities. As the downwardly mobile professionals driven by extortionary rents in urban cores move to predominantly minority residential neighborhoods, and bring their left-wing politics with them, this nativism will become an ever more salient attack against left-wing insurgent candidates.

Importantly, this is the exact attack used against YIMBYs, unfortunately, often by the left. People even call poor Basil a transplant even though their family’s presence in New York City predates the existence of some states. Even some YIMBYs, such as Scott Weiner1, have tried to make long-term residence an issue in their primaries, even though the desired impact of their policies is mass inflow of new residents to their cities! Both socialists and YIMBYs need to recognize that their class basis is currently essentially the same, and thus we will continually face this attack from our common antagonist: the network of political machines, interest groups, and incumbents that has delivered more than a decade of governance failure in every American city.

Of course, new entrants to cities make those cities stronger, and make incumbents better off. They are consumers and tax-payers! Even if you are a homeowner, an influx of new residents will drive up your property values and make you much richer.

I do want to discuss an element of this line of argument that I continue to find simply absurd. I do not see a way to differentiate new residents in cities from within the United States from immigrants.

Adriano Espaillat was the first undocumented member of Congress. He calls for the abolition of ICE, and is running much of his campaign on protecting residents from deportation. His record is, in actuality, enormously inconsistent on the issue, but it is insane to me to believe that every single person on Earth has a right to live and work in the United States of America, but also that everyone who moved to Inwood in the last five years should be expelled to Morris County.

To be clear - every single person on Earth does have the right to move to the United States of America, especially New York City and that is why we put a statue in our harbor with a big sign saying “WELCOME, THANK YOU FOR PARACHUTING IN.”

There is no way to generate this cut with any moral consistency. Perhaps one could argue that immigrants are more impoverished than the average professional transplant coming in from Texas. However, it seems less than obvious to me that the people making this argument would exclude high-income immigrants from moving to the United States, and if they did it would be one of the most counter-productive policies imaginable.

Even if you bite that bullet, many thousands of people who move to New York City from other parts of this country are leaving repressive states, fleeing laws restricting their sexuality, or their reproductive freedom. Like the millions who left the Jim Crow South during the Great Migration, they are realizing one of the fundamental rights that our republic guarantees through its federal structure - the right of exit. Would activists and politicians demonizing their more recent neighbors like to say into the microphone that they would prefer if transgender women stayed in Tennessee?

Urban localist nativism often does spill over into actual nativism. In 2024, Chicago saw protests where some Black residents protested the construction of migrant shelters. By whipping up this artificial division for electoral gain, whether to protect the rents of incumbent homeowners or to beat back progressive challengers, establishment Democrats are making a cut that they cannot maintain, and opening the door to the anti-immigrant politics that has annihilated their coalition.

And yet it continues! It feels rather strange to me for the left in Barcelona to be protesting tourists for similar reasons, and this does appear to be an international ideological failure. At least Claudia Scheimbaum at the consistency to call the anti-gentrification protests in Mexico City xenophobic. I guess that’s the type of moral clarity you get from a left-wing leader actually at the head of a mass popular movement.

I once got into a bit of a spat about housing policy, and my interlocutor told me I could never understand what it is actually like paying rent in this city because I wasn’t born here. I would later find out she is of Russian origin, born in Central Asia. What series of moral commitments could possibly generate this distinction?

What ever happened to the workers of the world, or even all men are brothers? At the core of my socialism is my cosmopolitanism, that overwhelming feeling one gets when you, struck by Kwame Appiah or Jean Jaures or Jesus Christ, suddenly realize that we are all human beings worthy of moral consideration who share a common project of steering our civilization away from Scylla. It holds for a Colombian refugee, an Indian software engineer, or even a gender-queer art director looking to make Spanish Harlem their home. It holds, even if our formerly undocumented AIPAC-backed Democratic establishment incumbent would like it not to.

I’ll be voting for Chevalier, for democratic socialism, for realigning American foreign policy, and for not having to be deported back to Scottsdale - a fate worse than death.

Subscribe now

Saikat Chakrabarti is a software engineer who co-founded Justice Democrats and served as chief of staff to Alexandria Ocasio-Cortez. He is currently a candidate for California’s 11th Congressional District.

**Zachary:** You’re running for Congress in San Francisco, the heart of the American technology industry, in which venture capital plays a crucial structural role. The CEO of Y Combinator, Garry Tan, unfortunately appears to have made it his mission to sink your campaign. Tan’s politics aside, many startups that have gone through Y Combinator certainly provide value, such as your corporate alma mater Stripe. Do you make a distinction in your views between venture capital as a social technology and venture capitalists as political agents? How important do you think the startup economy is to technological advancement, and do you consider potential negative impacts to the sector when designing policy?

**Saikat:** I think the startup economy is important. The amount of wealth creation that we’ve managed to generate in Silicon Valley — it’s good. The problem is that it’s basically the only model of financing for new industries and new businesses that we have in America. In the US, if you’re trying to build a company that won’t result in a potential 1,000x return in ten years, you are very limited in what you can do. But if you’re trying to do that in a place like China, or in the context of other governments where there are actual national plans for industries to build up, that’s not the case. There are all different kinds of financing available to you.

When I talk about designing policy in America — yes, I do think the current way we’ve structured who gets the returns from a lot of these investments is lopsided, and I do call for higher taxes on the wealthiest. But I don’t think the only kinds of businesses we should be funding are the ones that can potentially return 100x or 1,000x to investors. That leaves out a whole lot of other kinds of industries and businesses that are really important. I think we’re really seeing a stagnation of this sort of business model in our economy today. If you take out investment in AI, we are basically doing nothing in our economy. Our economy is barely growing, and that’s because we don’t have any concept of public financing or any sort of idea of directing any other part of our economy. A lot of the work around New Consensus and our Mission for America is about that: if we want to intentionally develop other industries, how do you do that?

**Zachary:** Are there specific sectors you think are poorly served by the venture capital mode of financing that public finance would be better suited to?

**Saikat:** Most sectors. Most businesses are not returning 100x or 1,000x on an investment. Let me give some examples of critical things that we need in our country right now that are not being served by venture capital.

Right now, if you wanted to expand the power grid in California, we rely on large power transformers. Large power transformers have about a five-to-six-year wait time because there are only two companies that build them in America. And to build large power transformers you need electric steel. There’s only one company that makes electric steel in America. These are the kinds of companies that require a lot of upfront investment to create. They do make a return, but it’s over time, and they’re not huge capital returns. So there aren’t entrepreneurs hopping up and down ready to build new large power transformer companies. There aren’t investors lining up to invest in them. We’ve just, as a nation, decided we are okay with the status quo of being completely stuck trying to expand our electrical grids because the private markets have decided this is not a good place to make a big return.

You see that example over and over again. If you want to build out train sets, there’s basically no company in America that does that because it’s a similar kind of problem. We rely on Siemens and other foreign companies to build them out for us. Our current model of financing and building out our economy has led us to this state where private capital is always trying to find the industries that might have the largest rate of return — not just a rate of return, the largest rate of return on investment. VC is where a lot of that’s ended up going, because if you get one hit out of 100 investments, you make so much on that one hit that it covers losses of your other investments. Relying on VC alone has the effect of weakening the rest of our economy and also geographically — we’re basically talking about tons of investment in just a few geographic locations in America, and everything else is being left behind.

**Zachary:** Throughout your interviews and public statements you often refer to yourself as a “class traitor.” You’re attempting to get rid of the money that you made at Stripe, and so on. But do you think Stripe is a valuable firm? Do you think the work you did at Stripe was valuable? Do you think that Stripe is good, that we have an economy that can produce a firm like Stripe?

**Saikat:** I think the work I did at Stripe — I’m proud of it. I think we built good stuff. My view on what’s broken with the economy is: should we reward people for starting successful businesses? Of course. But should we have an economy that is basically a winner-take-all system for survival? No.

The broken thing is not that Stripe was valuable. The broken thing was that at the same time that I was able to make a whole lot of money off of two years of hard work, people who are working incredibly hard every single day, doing in my view just as valuable work, can’t even afford to live in the city. They can’t afford a secure retirement. The whole idea that you can have a nice, dignified life off 40 hours a week is pretty much dead. The American dream is dead. Young people today are not going to do as well as their parents. We’ve created a lottery system where you end up working at the right company at the right time and you might hit it big, but otherwise you’re looking at a bleak future.

When I call myself a class traitor, it’s not because I think startups are bad, or starting businesses is bad, or the work I did is bad. The thing that I believe needs to happen is we have to rectify who is getting the share of the wealth in this economy. It shouldn’t all be going to just a handful of people. We need to fix the distribution of the wealth.

**Zachary:** The Niskanen Center’s Steve Teles released an article describing varieties of abundance, placing you and Zohran Mamdani in the category that he called “Red Plenty,” noting your celebration of the French postwar planning state and the New Deal. Do you view yourself in those terms, or take any insights from the supply-side and state capacity policy world?

**Saikat:** Explicitly, I tend to stay away from trying to figure out what bucket I’m in. The way I approach any problem or any situation is to look at the problem honestly and try to figure out what the actual solutions would be. I do look a lot at historical references for when we’ve actually succeeded with creating abundance for a whole society. It’s not just the French that I look at. I look at the example of America, not just during the New Deal, but really during the World War Two mobilization, which was, I would argue, the time where we really built up our means of making a living, built up our industrial base, and that was really what got us out of the Great Depression. But I also look at the European countries after the war, how they built back after they were bombed out. I look at the modern-day Asian nations — South Korea, Taiwan, Singapore — and also the Nordic countries during the ‘80s and ‘90s.

The interesting thing is, if you look at all of these countries, they all basically went through phases where they rapidly transformed their economies in really short periods of time. In the case of the Asian countries, they went from poverty economies to pretty much developed nations. In the case of America during World War Two, we went from not having a big industrial base to mass-producing more war material than any other nation in the world. And the way they all did it, if you look at how they managed it, was doing a kind of governing and a kind of state-led investment that’s outside of anything we do today. That’s been a lot of the focus of our research at New Consensus, and a lot of where my politics is informed from. My argument would be: if you’re serious about trying to create a future of abundance, you shouldn’t limit yourself to just one tool or one ideology. You should really look at the ways this has ever succeeded in the past and use every tool at your disposal to make it happen.

**Zachary:** One thing I find interesting about the examples you list is you mentioned the Nordic countries in the ‘80s and ‘90s. Why that period?

**Saikat:** That was really when they managed to upgrade their economies, especially Finland. It was being led by the cell phone technological revolution. In their case, it was in response to an economic crisis. During the fall of the USSR, Finland — before they had their massive expansion of their economy through largely Nokia — I believe their major exports were paper goods and paper products. They decided, as a nation, to invest in advanced technology and advanced cell phone technologies. They did this state-led development to build up not just Nokia, but the entire supply chain and infrastructure around it. Sweden and Norway were also kind of part of that larger project, or they played a role in it. That was the period during which the Nordic countries rapidly advanced their economy. It’s one of the more recent examples.

**Zachary:** It’s the period in which social democrats in those countries no longer had parliamentary majorities, but their institutions were not structurally threatened. Do you think the cooperation of the liberals in Scandinavia with the social democratic order allowed them the space for those policy agendas? Or do you think that, at its core, the social democratic legacy in those countries is what produces the ability to engage in state-run venture capital firms like in Finland?

**Saikat:** It was the cooperation. That’s the interesting thing about what I’m talking about. You’ve seen versions of countries reinventing themselves and their economies under all kinds of governments, all kinds of economic or political systems, and under all kinds of parties. In the Nordic countries, it was not led by the Social Democrats. It was really a failure of the Social Democrats to be able to renew their own institutions and economies that led to the liberals coming into power. But then the liberals did a kind of state-led investment that, if I was to pitch that today in America, often gets labeled as democratic socialist or even communist — or “Red Plenty,” as Steve Teles said. That’s why I think this is a practical kind of politics. It doesn’t neatly line up with some ideology. It’s really the basic idea of: how do countries decide to develop themselves, and what are the institutions and plans and procedures that they historically use to make that happen?

**Saikat:** State capacity we’ve lost for years — including the capacity to publicly finance and decide stuff we want to build in the economy. The big disagreement I often have with the Niskanen and abundance people is their idea of how you build state capacity is often, you know, you pick a few small problems and you work on that — make the DMV work better. I think that’s not actually how any country has ever built up their state capacity. The way countries build up their state capacity is you put the country on a big mission that’s politically exciting, and that’s how you actually get the best and brightest to come work in your government. That’s why you have the moonshot. I don’t think we’d be able to build up our capacity to create new materials that can go into space unless it was in the context of this bigger mission. I think that comes from the technocratic nature of the Niskanen people — they love tinkering on the policy side.

**Zachary:** The tech right, to an extent, also has an understanding of this but obviously no ideological, personal, or practical capacity to carry it out.

**Saikat:** They’re fundamentally hamstrung by this ideological view that the only way you can build or create growth in the economy is enabling private actors to do it. Other than basically the build-out of our national rail — which also had tons of public subsidy — I’d say the build-out of rails was driven largely by industrialists. That’s kind of the only example. It’s hard to find other examples.

**Zachary:** Build-out of AI, at least for now.

**Saikat:** And yeah, the build-out of AI. One of the funniest things to me — I met Sam Altman years ago, and the conversation I specifically had with him was he was telling me how you should never start a business that builds out infrastructure, because you’re going to lose in the end. And now I’m just like, look what are you doing?

**Zachary:** You would lose from a business perspective?

**Saikat:** It’s so much capital expense upfront that to make a return on that capital, it never works out in the end. He told me about the build-out of underwater cables — I guess that’s actually a private venture. So I guess this time he’s like, well, if we replace all labor, it’ll work out. But otherwise, I agree with him. It’s not a great private business to get into.

**Zachary:** You always see skeptics of the business model of the labs talking about the size of the capital expenditure being insane. Unless this automated vast portions of labor, it would be unsustainable. And all of the CEOs recognize this publicly. They’re just going to do it. And I think that’s probably the problem we should be far more concerned with.

Subscribe now

**Zachary:** Turning to AI — on Crash Lab, the site associated with your think tank New Consensus, a scenario is presented similar to that put forward by Citrini Research: a demand crunch triggered by the automation of white-collar work that leads to an economic downturn. This scenario has been criticized by economists for ignoring monetary policy responses, as well as the fact that in such a scenario, prices will fall, thus increasing the real wages of many downshifted white-collar employees. How confident are you in the demand crunch scenario? Is this simply one of many situations we should prepare for, or the modal outcome in your view?

**Saikat:** It’s one of many, because we can’t be sure exactly how AI is going to play out. But something interesting you said in the question is “downshifted white-collar employees” — this is assuming that white-collar employees are getting laid off and then finding other low-wage work somewhere. I don’t know if that’s necessarily a given. If you look at the levels of unemployment that AI threatens to create — something like 25%, or even in a nightmare scenario 40%, because about 40% of our workers in America can essentially do their jobs over Zoom, and those are the kinds of jobs that might get replaced by AI — I don’t believe we have a labor market willing to absorb that amount of white-collar workers.

When they talk about monetary policy — let’s think about what a monetary policy response means. There’s two kinds. One is lowering interest rates, which the traditional thinking is will help heat up investment, and the investment will lead to new industries and new businesses being formed, which leads to job creation. But in an AI future, we are basically decoupling industrial output from labor. So just interest rate policy is not going to solve that issue. You’re not going to be able to get employment to go back up by making the cost of borrowing lower.

The other kind of monetary policy we could be thinking about is things like quantitative easing — basically printing money. Currently, the way we do quantitative easing in America is the Federal Reserve can essentially print money to buy up assets of private businesses on the open markets, which is essentially a way to do a sort of stimulus to the industrial side. But again, if the labor market is being decoupled from industrial output, that doesn’t solve the demand crunch crisis. I don’t actually see how monetary policy can get us out of that demand crunch.

**Zachary:** In your plan, the “We Won’t Get Fooled Again Act,” it proceeds with the de facto gradual nationalization of the frontier AI labs, but contingent on them being distressed assets. Is your advocacy for this form of de facto socialization contingent on an economic downturn, or is this independently desirable?

**Saikat:** It’s one of the routes in which I think it can happen. But I believe it’s independently desirable, because if we are looking at a future where there are entire industries that no longer need workers to function — services that essentially become free to provide — the current way all this is structured is that those industries and services would pretty much be subsumed by the existing AI companies, and you’d have a few corporations basically owning a larger and larger portion of the economy. You’d end up with these massive natural monopolies.

In that sort of scenario, it would be better if the government can simply provide those services for free. The way to do that would be some form of the government either owning these frontier labs or owning the infrastructure — owning the data centers — and perhaps treating them as public utilities, or turning the frontier labs into mission-driven, not-for-profit corporations. If it is the case that we can do all this work without any labor needs, then that should mean people overall are able to just access those services for free and have all these productivity gains result in more leisure time, shorter work weeks, and more wealth — not in the money sense, just wealth in people’s lives — because all the stuff that used to require all this work now no longer requires it.

**Zachary:** Let’s say the labs have been nationalized and there is a national laboratory run as a public benefit corporation or as an element of the executive branch. These services are provided for free. How do we then deal with the labor displacement if, say, 25% of the American workforce is now out of a job? Are we going to be providing them a basic income? But from what revenues, from what taxes? This seems like a tension in the plan.

**Saikat:** If the government is directly providing the means of making a living to people, you’re just providing the means rather than having it happen through income. There’s a bunch of ways this might work. There’s a scenario in which AI replaces 97% of all the work that we need — that’s going to be a very different world. And in a world where there’s only 25% unemployment, what I believe we need is the state capacity and infrastructure to be able to handle any of those scenarios.

In my view, if we just go down the route of trying to provide people basic income but keep the current system of providing goods and services to people the same, that’s not going to work. Whereas if the government is directly providing the services to people, then you don’t need an income to buy that service. The government is just providing that service to you.

**Zachary:** Even under those conditions, there would still be parts of the basket of consumer goods that people would want to consume that wouldn’t be provided by the state. Would nationalizing the AI labs necessarily lead to state provision of food free at point of service, in kind? Or does there still need to be room for goods outside of the domain of what the public corporation is providing?

**Saikat:** Of course there’s going to be goods outside of what the public corporations are providing for free to the public. Though food is not really the best example, because I do actually think we should be able to provide food to everybody when we produce enough. Food is one of the most socialized parts of our government right now, where we essentially backstop farmers and can tell them what to produce and how much to produce, but then we still have a private market system for delivery. We have a situation where we can produce enough to provide for everybody, but we’ve decided not to. I don’t think it’s the most efficient market we have.

There are other models of food distribution. In places like Mexico, they’ve started doing price caps on an essential basket of foods. Beyond that, you can purchase other food according to your preferences, but it provides a basic safety net of the essentials.

In a world where we can actually produce anything we want without labor needs, that is going to be a different kind of economy. Money will essentially be something we are creating to give to people to purchase stuff, and there might still be work that people are doing and earning from. But the basis of this is: the essentials you need to survive — I believe we should directly provide those. I don’t think it would work to just give people an income to purchase the essentials in a world where you’re not able to have a job to earn that income. We’d be an intermediary between you and the services through money, when it’s either the government providing you money to get something for free from the government, or the government just giving you the thing.

**Zachary:** I completely agree that at the limit this seems crucial, but what about during a transition period before all or most labor is automated, and in the period in which specifically white-collar jobs that can be done on Zoom are being hit? Say we’re at 10 to 20% unemployment.

**Saikat:** So imagine that 10 to 20% of white-collar jobs get automated. These are jobs that currently earn somewhere between $100,000 to $400,000 a year. In San Francisco right now, a starting engineer at Google makes around $200,000 a year. The proposals I’ve heard so far for what to do about that would be UBI. To prevent a demand crunch from that kind of unemployment, from those levels of salary, the UBI would have to provide about $200,000 a year to these people, because if you’re providing less, there is going to be a demand crunch. Where would that level of money for that UBI come from if the government is not directly owning some of the means of production? And politically, is it really going to be feasible for us to have a largely blue-collar workforce making a lot less than $200,000 a year, continuing to work, to support money going for free to people making $200,000 a year? I don’t believe that’s politically possible.

The solution here has to be: 10 to 20% of white-collar workers are getting laid off — now would be the time to actually directly provide the essentials to everybody. Create the universal programs: healthcare, childcare, education, food, utilities. That would have more political buy-in. For white-collar workers who might be losing their jobs — there is other work that we need to be doing that may not be getting replaced by AI right now. This is a lot of the work in the Mission for America around building clean, high-wage industries that we would need folks to get retrained into. That would be productive work that we as a country just need to do. That’s how a transitionary phase would work, where AI hasn’t replaced all work. There is still work that needs to happen that AI hasn’t replaced. But in any version of this, we have to be able to provide all the essentials to people, and as some of these industries are starting to not require work, I think it’d be better for those industries to be available to everyone for free, rather than large corporations just sitting on those industries and essentially collecting rent off them. If there’s no work required to provide them, then they’re just collecting rent off of something that is free to provide.

**Zachary:** David Shor’s polling suggests that policies such as fully replacing the income of displaced employees under some threshold — say $140,000 — would be quite popular. Matt Bruenig wrote recently about how we should create a more robust unemployment insurance system as our primary solution to AI-driven job displacement. Neither of these people is particularly bearish on AI. But rather, they’re recognizing that the people who will be automated will want some other bundle of goods that the state may not be able to provide. And if we are not able to give all of these people a replacement job, then at least some degree of income replacement needs to be in our scheme.

**Saikat:** That seems fine. It basically sounds to me like we’re talking about providing the essentials to people. On top of that, the question is what amount of money people get to buy things that they want, which essentially is us just creating that money to give to people at that point. I just did the math — we’re talking about 25% unemployment at $150,000. That is about $6.3 trillion a year.

**Zachary:** That would be the literal maximal case of the transfer. I completely agree with you that it seems impossible to finance.

**Saikat:** That’s in fact the minimal case. $150,000 a year is a low-end salary for a white-collar worker being replaced right now.

**Zachary:** An unemployment insurance system, especially an uncapped or highly capped one in the advent of mass AI-driven automation — unless we nationalized the labs. But in that case, the national lab would have to basically be charging monopoly rents by the state in order to redistribute that wealth. In the “We Won’t Get Fooled Again Act,” there’s some mention of charging commercial fees to private actors seeking to use the services. Is this something you imagine the national lab doing, or do you think that, because of how robust the in-kind provisions will be, this will not be necessary at scale?

**Saikat:** It could be necessary. But basically what we’re talking about is: we have an economy that has a certain amount of output. The basic thing that money — the existence of currency — tries to solve is we give people money so they can choose what of that output to buy. We as a government decide we’re going to tax a small amount of that output to provide some services directly to people. And we have this system because we need humans and labor to continue to create that output.

In a world where we are needing less and less labor, I believe the amount of the total output that should be directly provided needs to be more and more. When I say $6.3 trillion, that sounds like a big number, but if I put that in the context of our GDP — our total output — it’s not a majority of it. But in a current consumer-driven economy where our GDP is largely driven by consumer spending, we do actually have to figure out a way to provide the equivalent of a majority of our GDP to everybody who may not have work.

There’s two ways to do that. One is we create a system of taxing the AI companies and the production, and then give that back to people, and they get to choose stuff to buy. Another system is the government simply provides a lot of stuff directly. The final mix might be some combination of those two things — providing the essentials, but then people also have money to choose to purchase things outside that. In a world where labor is no longer as necessary to produce that output, we’re essentially getting to decide what mix we want.

The thing I feel strongly about is that the essentials should not be something that people are begging and scraping to get. But how we decide to distribute goods of choice — there’ll be some market, some mechanism for people to make choices and get money to make those choices. I just think a simple income replacement strategy is likely to fail, especially given the political dynamics I described earlier. I like universal programs because they’re, in my view, more stable and more possible to create.

Share

**Zachary:** This seems to imply pretty dramatic shifts in our economic and social life in the very near term. Are you pretty confident that those shifts will occur over some near time span — say, 10 to 20 years?

**Saikat:** I don’t know. I think we’re seeing some job loss from AI already. If the rate of improvement continues and doesn’t cap out — which isn’t known, it might cap out at some point, there may be some fundamental issue with LLMs where they can’t actually achieve full human capacity — then we are talking about a fundamental shift in our society whether we like it or not. And the question to me is: do we want to plan for that potential shift and handle the transition in a way that creates the least amount of harm and the most amount of good, trying to transition to a Star Trek-style future? Or will we let that happen chaotically? The way we’re going right now is chaos.

**Zachary:** If we have rapid advancement in capabilities, the curve continues going up — what kind of world do you want to live in? Get as utopian as you want. Is Star Trek the model?

**Saikat:** I think that would be amazing. If we get to a place where the basic needs of people are taken care of, people have autonomy and dignity in their lives, they get to choose what they want to work on, we focus our energies as a people on exploration, art, scientific discovery, and get out of having to do the drudgery of just keeping society running. We as a people are in control of our futures and our destinies. We don’t feel like we’re under the thumb of certain despotic leaders, whether political or corporate. That would be great, and I think it’s a future worth striving for. I don’t think it’s going to happen automatically, but we should be striving for a future where everybody gets to live a good, full, dignified life. That’s at the core of my politics.

**Zachary:** It seems that large portions of the left, though this is starting to change, are not just skeptical of AI but actively opposed to its rollout in any circumstance, including the burgeoning anti-data-center movement. Do you think a national moratorium on data centers would be a good idea? And do you think local interventions against data centers — protests, petitions, and so on — are worthwhile given the potential upside, or do those actions make sense strategically in the absence of socialization?

**Saikat:** The specific thing that Bernie Sanders argues for with the data center moratorium is — he takes this technology very seriously. He was, if anything, the first among the more left politicians to really take AI seriously and the potential of its impacts on the labor markets. He’s saying: until we can figure out how to actually regulate this technology, we should put a pause on this. I think that makes sense, because this is progressing at such a rapid scale. I know, if you saw the benchmark numbers from Claude floating around — people who I’ve talked to who are working inside these labs are telling me they’re freaked out by how fast this is going, and society is nowhere at a state to even know how to handle the potential disruption that’s coming our way. In the face of that, I do believe it makes sense to pause until we can actually do this correctly.

I don’t want to end up in a doomsday scenario with massive job loss and recessions before we’re at a place politically where we can actually control this technology. I’d rather we figure out how to roll this technology out correctly now. And to some extent, I see what Bernie Sanders is doing as a negotiating strategy. He’s trying to force the AI labs to come to the table and get to a point where we are, as a nation, actually figuring out how to regulate and control this technology. He has, I think correctly, surmised that without a potential threat to their underlying business model, they are not going to come to the table and negotiate.

**Zachary:** Are you concerned about existential risk to human life or autonomy from loss of control to a power-seeking AI system?

**Saikat:** I’m pretty scared about that too. There’s been more and more papers and research coming out about AI agents deceiving their testers, working without correct alignment. Anthropic seems to be one of the only labs taking it seriously. OpenAI and Google, I don’t believe are. I think it’s pretty important that we actually start having national standards for when we’re deciding a model can be released, because there is a real threat of a model coming out that starts acting of its own will, starts helping people develop biotech weapons. We’re kind of playing with fire at this point. It hasn’t gotten to that point yet, but I am scared of the existential risk as well. I don’t think I’ve seen anything that really has proposed a real potential solution, or a body to do oversight on these labs that would lead me to believe that we’re on top of this threat.

**Zachary:** If you were writing a bill outlining the mandate and governance structure of our new nationalized lab, what do you think the tradeoff of safety and caution on release should be versus rollout? The AI labs as they currently exist seem to be deprioritizing safety in favor of profit-seeking. But the state lab would have an unclear set of objectives. Would you say we should overwhelmingly prioritize safety, or is there some degree at which you’re willing to make a tradeoff on safety to provide benefits to the rest of humanity?

**Saikat:** The way I’d structure it — and I do think Anthropic has done a decent job, though they have started to potentially slip a little — we would have to have really clear guidelines of which safety benchmarks any model needs to pass before being released. One of the benefits of having this be a state lab or nationalized lab is there wouldn’t be the constant pressure of competition from your competitors and the profit-seeking motive forcing the tradeoff between speed and safety. If we’re doing this at a national scale, we could take our time. There’s no reason why we have to have AGI appear next year. We can take our time to do it right, to make sure it’s safe. It would prioritize safety over speed.

That would also give us time to plan out the other repercussions around labor markets, around societal effects from the proliferation of deepfakes. There are all these potential ill effects that we’d have time to handle correctly. In the current private markets where AI is being developed, safety often falls by the wayside because there is this threat of competitors always getting ahead of you. I’m honestly surprised that Anthropic sort of kept up its safety benchmarks for as long as they have. Good on them. But the other labs aren’t necessarily doing that.

**Zachary:** Nationalizing the lab certainly solves that dimension of the problem, but it seems to introduce others. Would you be concerned that the nationalized lab would provide the executive branch, or even the employees or CEO of the nationalized lab itself, extreme concentration of power? How could we design such an institution to prevent that?

**Saikat:** Of course. I’m always concerned in any sort of governance situation that we’re going to end up with an extreme concentration of political power, and as we’ve seen, even well-designed democratic systems experience that — that’s happening in our country right now with Donald Trump. We would have to have some series of checks and balances, like an independent commission overseeing this. There are models for doing real oversight of government agencies that have proven to be decently successful. In the federal government, we’ve had an Office of the Inspector General that’s been an auditor and overseer for other agencies, and until Trump fired them all, it was a pretty effective way to catch fraud and abuse and waste. Having independent offices like that overseeing an agency like this would be one model.

The thing I do think having a nationalized lab gives you is that we democratically can vote and create a governance structure that we are then legally bound to, which in a corporate governance structure simply doesn’t exist. We just saw this play out with OpenAI, where they had come up with a whole governance structure to try to keep OpenAI from ever releasing unsafe models. Sam Altman basically blew that away in the matter of five days when they tried to fire him. It’s much harder for a corporate governance structure to have the staying power that a national governance structure does.

**Zachary:** Are you ready for the fight that would entail? It would be framed as the largest expropriation of private property in, plausibly, the history of the world, if the US government nationalized the labs in, say, 2030. Do you think that will really be a massive political fight with concentrated private interests? Or do you think to some extent it’s in the interests of the lab employees? Do you see this, especially if it’s happening while they’re not distressed assets, as a pretty dramatic contestation with private power in a way that doesn’t have much comparison in American history?

**Saikat:** That’s why the plan we have out right now is about buying out distressed assets, which there’s precedent for. I think there might be other routes — getting equity stakes, having governance power. There’s precedent for the government stepping in when there are incredibly dangerous technologies that have society-wide effects. It might be a big fight, and I feel like we are headed there anyway.

Either AI is going to be this hugely disruptive force in our society, and then we are going to have to contend with how to control this technology and do whatever we can to make it work for the good of humanity, or it won’t — in which case this whole conversation doesn’t matter so much. But we have to be ready for that. If we have a government that’s not willing to protect its people in the face of huge disruption, then I believe that’s an ineffective government.

This is ultimately one of the big reasons I’m running. I do think we need people with the courage and conviction to face up to these moments and be willing to put out potentially controversial visions of what needs to happen. I’m always open to new information. The stuff I’m pitching might be wrong. But I haven’t heard other versions of how we can manage this, if it is at the level of disruption that the AI labs are saying it will be.

**Zachary:** When do you think you started recognizing that this was going to be potentially very transformative?

**Saikat:** I had some conversations with people who were deep in this field back in 2017. This was right after OpenAI had competed in their first Dota match. I believe it was the year that their AI didn’t win. I was talking to some people there, and at that time, they were telling me: in five years, either hardware is going to catch up and we’re going to get a real breakthrough, or not.

That was the first time I really started to take it seriously. I’m a computer science grad, I’ve been in tech. I’ve been hearing about the power of neural nets and AI my entire life. I took a neural nets class in college where I built out a simple model to do visual recognition of plants. But that year was the first time I thought, this sounds different, because it’s somebody I trust the technical chops of telling me this is coming. And then the stuff around the Transformers happened, I believe around the same time.

When that prediction actually came true — it was literally five years before ChatGPT came out — and then the first version of ChatGPT had a bunch of problems, but I was hearing those same people saying this is going to develop very rapidly. I always take it with a grain of salt when people trying to raise money are overhyping anything, but when I hear engineers telling me the same thing on the inside, I take it somewhat seriously. And that has come to be the case. The length of time between upgrades and models has shrunk over the last several years, and it is improving faster and faster.

I’m still not sure I’m on the AGI train. I think it’s an intelligence, but it might end up being a different kind of intelligence. But I think that almost doesn’t matter. The AGI discussion misses the larger point: are we looking at something that will cause widespread disruption, but also could cause widespread good? My view of politics has always been: we should plan for the potential futures, and if we don’t end up needing those plans, that’s fine, but at least we’ve tried.

This interview has been edited and rearranged for coherence and brevity.

In short, for data center moratoria to build a working class coalition, they must unify their diverse coalitions around an explicit positive vision for what the pause should achieve. Without this vision, moratoria will become the end-goal itself, and yet another opportunity to democratize technological advancement will pass us by.

Subscribe now

Diverse Coalitions Need Shared Goals

Holly Buck argues that moratoria aren’t a substitute for actual AI governance. Her argument that a US moratorium merely forces data center production to jurisdictions with weaker labor and environmental standards is a serious critique internationalists should consider. It is, in fact, quite possible for us to decarbonize data centers on a technical level, it just takes political will. That said, her vision of AI as a utility seems, if not ill-conceived, ill-timed; we’re at a point where abundant and cheap AI sounds more annoying than helpful to anyone who isn’t a Claude Code power-user. However, Buck makes a strong case that moratoria are not desirable as an end itself, and that the opportunities in AI governance go beyond regulating construction.

Regunberg sees Buck as arguing against a strawman. Regunberg argues that Buck “seems to misunderstand the thrust of the data center moratorium effort, which is to stall the breakneck development of these projects so we can enact ‘Actual AI governance.’” As evidence, he points toward a “comprehensive suite of actionable steps” as his governance strategy. Notably, all of these steps are on the regulation of data center construction; they don’t say anything about the governance of AI. Certainly the mitigation of local harms is the minimum viable policy that could organize the diverse parties opposing data center construction. But it is just that: the minimum.

If organizers like Regunberg want to “raise class consciousness” with moratoria advocacy, we need to raise these legitimate but local disputes towards universalist concerns. What does that policy look like? Given Regunberg positively cites Alex Bores, a Democratic Candidate for Congress in NY-12, I would imagine he would also support policies like the RAISE act that establishes safety standards for frontier AI or a public AI Dividend. The Justice Democrat insurgent Saikat Chakrabarti’s AI policy positions him to Bores’ left on labor, demanding gains from AI go towards a 32-hour work week & supporting the PRO act to grow union power. Both of these platforms include data center governance as one plank of a larger vision of progressive AI governance including AI Safety and worker’s rights—each components of a majoritarian coalition that could change the balance of power between working people and the tech industry.

The Purpose of a Coalition is What it Demands

Unfortunately, the coalitions calling for data center moratoria do not seem to share that vision yet; among all state-level moratoria bills, none of them make demands that expand outside of data center governance. Every moratoria provides a sunset clause—conditions at which the pause would end. The majority are time-based, demanding a 1-3 year pause. Many of those specify some action to be taken during the pause—such as an environmental study or future legislation—but not all. A few base their sunset clauses on the passing of future legislation. If moratoria make legislative demands at all, they never make demands about how to govern AI.

Why would we want a pause with no demands? Why would we pass a pause that demands specific future legislation when we could just pass that future legislation? To some degree, Buck is vindicated; many people who support moratoria see a pause as the end goal itself.

Across party lines, people commonly believe that AI is a bill of goods from Big Tech like crypto or social media. Of course, I disagree: AI capabilities are advancing rapidly and it is incumbent on the Left take that fact seriously and socialize its benefits.

However, Regunberg offers an alternative explanation:

“The problem is not that we don’t have policy solutions to this crisis. The problem is that Big Tech’s private (and unpopular) investment in data centers is moving at an astounding pace and we don’t have the time or leverage to establish the regulatory framework necessary to make this system work for the public.“

This makes some sense for governing the construction of data centers. If you can stop the concrete from pouring, you buy some time for your legislators to study up on the issue, negotiate with developers, and get these centers built cleanly without hurting rate-payers. In fact, where state-level moratoria are introduced, they often act as a maximalist anchor to transfer community-level pressure campaigns onto a package of passable data center bills that implement immediate changes to tax credits and utility rates.

But governing AI must extend beyond governing data center construction. If we have the policy solutions for AI governance, and we have the grassroots political anger against “the greed and avarice of an industry that is endangering our democracy,” then why have we failed to channel that political will into legislation that would protect and empower working people?

Buck suggests that this might be because data center opposition is defined by the interests of middle class incumbents, incapable of supporting a working class politics.

Of course, data center opposition has no essential class or political character, but is a diverse movement that spans from rural NIMBYs to Brooklyn socialists. It is incumbent on the leftists organizing in these spaces to prove Buck wrong by championing legislation that protects people not only as ratepayers and residents of a neighborhood, but also as workers of a class and citizens of a global community.

To address job displacement and economic uncertainty, this could include expanding bargaining rights, severance taxes on data center electricity consumption, or a state-level AI dividend for active redistribution. This also seems like an opportunity to draw attention to the safety risks AI poses, from the everyday problems of interactions with kids to the existential threats posed by biological and cyber capabilities.

This is the strategy of Senator Bernie Sanders and Congresswoman Alexandria Ocasio-Cortez.

Globalizing Data Center Discourse

The sunset clause of the Sanders/AOC federal moratoria—distinct from every state-level moratorium—is conditioned on future legislation that would protect workers’ rights and minimize existential risk. Sanders in particular has used this proposal as a platform to advocate for issues beyond and including data center construction; his “AI Data Center National Organizing Call” spends as much time talking about environmental damage and utility hikes as he does about job displacement and existential risk. His advocacy has connected localist concerns to not only federal AI governance but also international collaboration with China.

While Sanders actions have been providing a positive vision for how the Left could approach AI governance, I worry that the moratorium could become the goal itself. The bill’s sunset clause lists five reasonable demands covering safety, labor, and data center construction to end the pause. I doubt more than a fraction of the people that have heard about this bill could list those demands; most people could only tell you that Sanders supports a moratorium.

Here’s why we should be concerned about that: sometimes the policies presented to voters do not represent the policies that would achieve the outcomes voters actually want. If the only option you hear to stop unregulated data center expansion, job displacement, and existential risk is a moratorium, then of course you would support a moratorium. If that remains the case, populist candidates (or candidates that want to outflank populist opponents) will be forced into a bad equilibrium where the most effective way to communicate a willingness to protects the environment and working people with AI governance is by backing a moratorium.

But this risk isn’t unique to the federal moratorium; the state-level organizations leading this fight propose moratoria as a maximalist demand to build leverage for the rest of the bills in their portfolio, but the moratorium is the main piece the public hears about. In Georgia, the Sierra Club’s priority bills focused on transparency, cost-allocation, and an end to tax breaks. In Pennsylvania, PennFuture endorsed a statewide moratorium while also backing bills on protecting ratepayers and requiring clean energy, annual energy and water reporting, and a state-developed model municipal ordinance. Even in Virginia, the same coalition runs the same multi-bill strategy. It’s not that the people fighting for moratoria have no plan; it’s that those plans are split up across half a dozen bills per state and feel less radical than moratoria. The program is too fragmented to enter the public consciousness as a unified vision, so the moratorium sucks up all the air out of the conversation.

When we use moratoria as a maximal demand yet fail to convey a clear end-goal, the moratorium will become the goal itself. If we want to expand grassroots political anger to fight Big Tech into a political program that can meet our moment, we must communicate a clear and salient vision for AI governance that goes beyond the first step.

Subscribe now

I support Chakrabarti because I have closer alignment with his socialist platform and believe him to be incredibly competent, but given my strong support for expanding the capacity of the private market to provide housing and other ‘abundance’ policies, I was concerned that this race become a frustrating battle between the left and abundance liberalism, upsetting the coalition that this publication is designed to foment. Thankfully, that hasn’t happened.

Instead the race has boiled over about AI.

Public First PAC, backed* by $20 million from Anthropic and led by former Rep. Brad Carson (D-OK) and former Rep. Chris Stewart (R-UT), has intervened in favor of Scott Weiner to the tune of half a million dollars. Public First is a safetyist PAC that supports a national regulatory framework and Weiner was the sponsor of SB53, a landmark piece of AI safety legislation later gutted by Governor Gavin Newsom. Saikat has responded to Public First beginning to spend in favor of Weiner by framing the situation as the “AI lobby” entering the race to back Scott Weiner, alleging that Weiner helped water down SB53 in coordination with said lobby.

This is completely false, as Nathan Calvin points out in detail, even if the narrative seems plausible to the uninformed. Weiner did fight tooth and nail for the original text of SB53. I think Saikat is likely misinformed by a misleading article on the subject, and I encourage people to read the more accurate account in Politico. Regardless, I will continue to support Saikat, and don’t think this race is particularly important from an AI-safety perspective, unlike NY-12.

The case for the AI safety community backing Alex Bores in NY-12 is that he will sit on the Science, Space, and Technology Committee and propose a comprehensive piece of AI safety regulation, and has provided comprehensive frameworks for this purpose. Unlike Saikat or Bores, Weiner doesn’t have proposals for federal AI policy on his website. SB53 is a great, but if Weiner doesn’t sit on S&T and doesn’t propose legislation, the actual benefit is pretty minimal.

Weiner simply does not have the overriding commitment that Bores has for advancing federal regulation. Weiner refused to participate in Mission Local’s April 17th tech, AI, and crypto candidate Q&A, the only one to do so. This is not to say I think Scott Weiner is in the pocket of the labs, nor do I believe that he hasn’t taken political risks in supporting regulation on a state level. I just don’t think it’s a priority for him federally, and thus think it is pretty counterproductive to spend hundreds of thousands of dollars and political capital intervening in a Democratic primary. I am very much open to being wrong about this, and if anyone from Weiner’s staff wants to demonstrate that Weiner will focus on drafting federal AI legislation over other issues, I welcome the feedback.

Saikat does have some discussion of AI policy on his website, supporting a full slate of safety regulations. The crucial difference between Saikat and Weiner on AI policy is Saikat’s advocacy for public ownership and a temporary pause on data center construction, both motivated by a concern for extreme power concentration job displacement, and yes, existential risk. In the upcoming first edition of 1,000 Means, we interview Saikat on these topics, so hopefully that will make his positions more clear.

Public First has a questionnaire on their website for candidates and I’m pretty sure that Saikat would answer in complete agreement to Weiner on every point, with the major piece of contestation being Saikat’s focus on the ownership of AGI rather than just regulation. Regardless, both of these candidates support federal regulatory frameworks for AI to prevent catastrophic risk, and given Weiner is unlikely to prioritize the issue in Congress it doesn’t seem like anyone should be throwing half a million dollars into this race.

The final note I will make here is that it does worry me that Scott Weiner is receiving support from the Abundant Future PAC, which does include several anti-regulation VCs, though I do buy the idea that their support for Weiner is primarily predicated on opposition to redistribution, and all’s fair in love and class war.

What’s under discussed is that there is a cost to backing Weiner, alienating progressive Democrats and driving a wedge between catastrophic risk prevention and AI-populism, a coalition that must cohere if any this is to go well.

I am concerned that Public First Action’s strategy is counterproductive to AI safety on this dimension and others.

First of all, according to recent reporting by Transformer, Anthropic’s donation to Public First Action was a “contribution to a 501(c)(4) exclusively in support of its mission to educate the public on AI policy.” Even though Public First Action and its PACs, Public First, Defending Our Values, and Jobs and Democracy, have spent over three million dollars in recent elections, neither Public First Action nor Anthropic had disclosed the restriction on Anthropic’s donation, which would prevent any of Anthropic’s $20 million from being used to support or attack any candidate. While canvassing for Alex Bores I have repeatedly had to defend Bores from confused voters asking why he was backed by one AI company and attacked by others. This is a communications failure by Anthropic and Public First that has imposed costs on the candidates the PAC is trying to support.

Why has Public First Action not done considerable work clarifying this point? Plausibly, it’s to continue supporting the idea that they are an effective counterweight to Leading the Future, which has led to an influx of donations from sympathetic actors in the AI safety community and beyond.

However, Public First Action is not a counter to Leading the Future outside of the race for NY-12. In his excellent thread on SB53, Nathan Calvin writes that Public First is “fighting against AI preemption” and “supporting states in their efforts to enact AI protections”. Brad Carson frames his PAC as an effort to right the big spending of the AI PACs, such as their advocacy for the permissive Illinois liability bill. Carson says that his PAC’s goal is “supporting Democrats who put people before Big Tech”.

While I do believe that Carson is well intentioned, I don’t think this is what Public First is actually doing. I will argue that Public First’s activities are actively undermining AI safety and are pursuing an incoherent political strategy rather than acting as a true counterweight to Leading the Future. Frankly, I think people in the AI safety community should probably cease donating to this PAC and set up another vehicle for support of Bores if Public First does not dramatically change their strategy. I am open to correction on this matter if someone can provide a case that what they are doing makes any sense

All of Public First’s campaign non-Manhattan spending has been on moderate Democrats or Trump-endorsed Republicans. Public First has repeatedly intervened in favor of moderate Democrats against progressive opponents who also back comprehensive AI regulation. This has had the effect of polarizing the left against the AI safety community, limiting the capacity of experts to influence outcomes of the emergent wave of anti-AI populism.

The problem may lie in the PAC’s leadership. Brad Carson was a moderate Democrat while in Congress, one of the 81 Democrats who voted in favor of the invasion of Iraq. This is likely related, even if unintentionally, to the PAC’s spending bias in favor of moderate candidates against progressives.

To clarify,

i) I am not arguing that it was bad for Brad Carson, who represented Oklahoma, to be a Blue Dog. It is a deep loss for the Democratic Party that our House delegation no longer includes many like him.

ii) I am not arguing that Brad Carson’s prioritization of AI safety is illegitimate or unfaithful. Even a cursory glance at his public comments demonstrates he is seriously concerned with AI risk and is attempting to fight back against the deluge of spending by OpenAI and its accelerationist co-conspirators.

I would simply contend that there are better strategic approaches to fighting Leading the Future, like actually fighting them in the campaigns they are spending in instead of intervening in low-impact races with high factional salience.

Former Rep. Stewart meanwhile, is pro-Trump, runs a defense lobbying firm, and voted against certifying the 2020 election. Regardless of one’s positions on AI safety, I do not think Anthropic or anyone else should be giving someone who participated in an attempted coup d’etat against the republic millions of dollars to spend.

Beyond the race in San Francisco, Public First has been intervening across the country, not always in ways that I find strategic.

Subscribe now

Valerie Foushee (NC-4)

The PAC spent $1.6 million supporting moderate incumbent Valerie Foushee against a Bernie Sanders backed progressive challenger Nida Allam. Foushee serves on the House Committee on Science, Space, and Technology, and is co-chair of the House Democrat Commission on AI and the Innovation Economy. I at least understand the outline of a case for spending in this race, but what is the actual argument here? Is the world substantively better along the dimension of AI safety because Foushee was victorious in her extremely close primary, which was decided by less than 1%?

Foushee has proposed some (rather weak) pieces of legislation, such as the Expanding AI Voices Act (HR 7158), and has called for environmental standards in data center construction. She opposed Trump’s guardrail-less National AI Legislative Framework. Besides for sitting on a relevant committee and being on a relevant Commission, it is unclear why Foushee is marked as a champion of AI safety, enough to intervene in a competitive primary and protect her against a challenger.

Nida Allam’s closing ad buy noted that, “As Election Day approaches, you’ll see nearly $2 million of ads for my opponent, funded by AI-backed super PACs - the same AI corporation who powered Trump’s attacks on Iran.” After she lost, Allam alleged that “the AI lobby just bought its first seat in Congress.” Those with knowledge of the differences between the PACs can quibble with this characterization, but I don’t think it’s completely unfair, especially given the lack of clarification by both Public First and Anthropic on their relationship. This well poisoning race has become a common talking point on the left and has generated deep distrust of Anthropic and the nexus of classical AI safety advocacy in general.

Allam was in favor of a data center moratorium, a policy that we have expressed disagreement with, and if Public First wants to prioritize preventing a moratorium, then they should probably just say that instead.

There are a host of Democrats on S&T that will be involved in writing AI bills, but I frankly don’t trust any of the incumbents to write this legislation, which is exactly why I spend every Sunday morning knocking on doors for a State Assemblyman from the Upper East Side! Blindly throwing money Foushee has coalitional cost and strengthens the position of incumbents who don’t appear to be ready to write the comprehensive AI regulatory frameworks that the situation necessitates.

Colin Allred (TX-33)

This one is even stranger. Allred is a former Congressman who, after losing a Senate race to Ted Cruz in 2024 and accusing Democratic Senate candidate James Talarico of racism, is heading to a runoff against an incumbent, Julie Johnson, who is endorsed for re-election by Alexandria Ocasio-Cortez. Public First is spending $150,000 backing a moderate primary challenger against an incumbent, a politically costly activity more salient than spending in an open seat. So, what are they getting out of this?

Allred’s platform calls for a non-preemptive federal AI framework, renewable power for data centers, and export controls, ie. the positions held by almost all Democrats, including Rep. Johnson. Intervening in this race is a straightforwardly moderate factional fight that provides zero benefit from an AI safety perspective.

Carlos de la Cruz (TX-35)

After Texas Republicans passed an unrepresentative gerrymander, Rep. Greg Casar’s seat, TX-35, was made lean-R seat. Unfortunately, Casar refused to test the progressive case for winning over Trump voters and is now running in Lloyd Doggett’s district upon the latter’s retirement. The seat is open, and is targeted by the Democratic Congressional Campaign Committee as a crucial in-play district to win the House.

In order for sufficient AI regulation to pass, Democrats must win the House majority. Even with some brave Republican voices in favor of regulation, the party’s leadership, and its President, is bought and paid for by OpenAI and a16z. House Republicans continuously float AI preemption as an NDAA rider!

Thus, it is likely insanely counterproductive for donors to be giving money to a PAC spending nearly half a million dollars to elect Carlos de la Cruz, a hard-right reactionary who promises to be the President’s “wingman in Congress” and has a plank in his platform for “Defeating the Woke Left”, which is “poisoning our kids”. I do not think this position is conducive to creating broad coalitions for AI safety.

His platform’s AI section has support for safety evaluations, deepfake protection, and semiconductor export controls, but all of that is outweighed by the fact that backing him decreases the chance of a Democratic majority that would actually advance AI safety bills. He also has no record on this issue beyond tacking it onto his campaign platform, which pokes a hole into a common characterization that Public First’s strategy is to back candidates with experience delivering results on AI safety.

Even beyond loss of control regulations, if you are concerned about extreme power concentration and the protection of American democracy, it seems extremely bad to me to provide support for de la Cruz, who is in a runoff against a more moderate and bipartisan candidate John Lujan.

Alexandra Mealer (TX-09)

Mealer is another Trump endorsee that Public First is backing, with nearly a quarter-million, over a more moderate Republican candidate in a Texas runoff. Though she has some vague commitments to safety reviews for powerful AI systems, she also said that she supports Trump’s AI legislative framework, an insufficient sop to the industry that Brad Carson described as “saccharine: empty of nutrition: certain to leave a bitter aftertaste, and probably carcinogenic.” I would argue that if a candidate supports a policy you would describe as carcinogenic, perhaps they are not the best targets for supportive spending?

Much like Leading the Future’s ads, the GOP-facing Public First cut-out Defend Our Values ran ads focusing on Mealer’s military record, with no mention of AI policy. This is the exact behavior the AI safety community has criticized Leading the Future for. This holds for much of Public First’s spending, which limits our ability to fight back against Leading the Future’s dishonest tactics in good faith.

Subscribe now

Pete Ricketts (NE-SEN)

If Congress is to regain its agency in restraining the executive, and if substantive regulation is to pass both houses, it is crucial that Democrats win the Senate. One of the more promising races in this regard is moderate independent Dan Osborn’s campaign against Senator Pete Ricketts in Nebraska. Unfortunately, the PAC is spending to get Ricketts re-elected and thus de facto fighting to secure a GOP majority in the Senate. Unlike the other races here, Ricketts does not face a competitive primary, so this intervention’s primary effect is changing the partisan balance of the upper chamber.

Ricketts is a co-sponsor of two pieces of quality legislation, both on security and hardware export controls, the SAFE CHIPS Act and the MATCH Act. These bills are not motivated by AI safety concerns, but by Rickett’s deep hawkishness. He did not take a public position in opposition to Ted Cruz’s proposed 10-year state AI moratorium, unlike Nebraska’s other Republican Senator Deb Fischer, nor has he advanced any other comprehensive regulatory framework. His AI policy advocacy is narrowly focused on maintaining competitive advantage with China, which can get Republicans to support some useful policies like chip export controls, but is also a major ideological barrier to actually sufficient regulation or a negotiated pause. The MATCH Act is a decent piece of legislation, and if you sought to signal that Republicans should break with the administration on chip export controls, why not spend token sums on the bill’s co-sponsors, like Senator Risch (R-ID), who is not in a competitive race that may decide control of the Senate?

Osborn has gone on the offensive, describing the $200,000 spent by Public First Action as “a Silicon Valley company wants to buy Senate votes on A.I. policy”. I remain skeptical that alienating a moderate working-class populist by spending against him in favor of a GOP incumbent will decrease existential risk.

It is unclear to me what values lies in increasing the probability of a GOP majority, ensuring the re-election of a Republican who holds a position on export controls shared by several other co-sponsors. Unlike with Alex Bores, a uniquely competent expert on AI policy, electing Pete Ricketts over Dan Osborn does essentially nothing to increase the likelihood of future positive regulatory developments. Yes, Pete Ricketts is “a strong advocate for controls that keep advanced AI chips out of the hands of adversaries”, but there are two candidates in this election and the other would also certainly vote in favor of chip export controls.

I want you to imagine the world in which Democrats are 1 seat away from a majority, and the closest the Caucus came was Osborn after an AI Safety PAC spent against him. How do you think Democrats will react? Do you think they will be ready to craft sensible well-designed AI regulation then?

Alex Bores (NY-12)

While canvassing for Bores, a voter expressed some confusion to me regarding the spending patterns of the race. Why were some AI companies targeting him and other supporting him? It seemed to this voter that perhaps this was some Silicon Valley infighting she didn’t want to deal with, and perhaps she should take a look at the candidates without any spending for or against them from big tech.

Both of Bores’s opponents have made a version of this argument. Micah Lasher names Anthropic, OpenAI, and Palantir as tech used to advance Trump’s “fascist agenda.” He has said that ,“the AI industry is fighting to decide the terms of its own regulation. While OpenAI pushes to defeat/weaken AI bills it sees as restrictive, Anthropic outwardly supports regulation while working to narrow scope.”

Public First’s intervention has allowed Lasher to distance himself from the spending in the race with pox-on-both-your-houses rhetoric even though he is the prime beneficiary of Leading the Future’s much more aggressive spending. Electing Bores is a crucial gain for AI safety, but it is unclear that the spending will actually help, as it dilutes the basic moral case against big spending. Now, if Public First made it clear that their campaign activities were not funded by Anthropic, perhaps by making clear who actually is doing so, that would probably help.

It has given Jack Schlossberg the opportunity to say he is “running against people who are backed by massive billionaires or massive AI companies.” Let’s not give this dilettante any ammunition!

It is of course insane to see either of these men posture on AI as they conspicuously leave their policy preferences on the topic underspecified, as to not upset the avalanche trying to bury Bores.

Countering Leading the Future’s deluge is a worthwhile objective, and ensuring that doing so does not complicate Bores’s case against them is a necessary component of doing so. However, this is the only race where Public First is spending against Leading the Future.

Subscribe now

What are we even doing then?

The crucial difference between the Bores case and Weiner, Foushee, etc, is that Alex Bores is the target of massive accelerationist spending against him, and is thus plausible that Public First’s spending may counteract this offensive. Greg Brockman isn’t spending a dime trying to elect Saikat Chakrabarti or sink Scott Weiner!

Public First and Anthropic are incurring dramatic political cost in tearing apart their relationship with progressives for minimal policy gain, confusing the AI safety movement’s ability to critique Leading the Future’s spending, and increasing the chance of a GOP majority. My best explanation for why this is happening is that it is simply misguided, but it is a dangerous mistake.

How do you think progressives will react as the AI safety community’s money is deployed against left-wing candidates in primaries, even as those candidates support AI safety regulations? Why are we wasting money and political capital on Democratic Party factional fights with minimal EV?

Our most powerful tool in attacking much more suffuse spending of Leading the Future is that they are trying to buy elections by running ads that obfuscate the true purpose of the PAC. This becomes substantially more difficult if pro-regulation advocates are doing the same goddamn thing. There is a narrative that Public First is primarily acting as a counterweight to Leading the Future, and if their policy was just matching Josh Vlasto in the same races, this would make sense.

However, Public First is engaged in a wildly different strategy that isn’t contesting the accelerationists in any race except for Bores’s, and in fact has spent on the same side as a16z and co for Josh Gottheimer in a battle to influence one of the most aggressively factional moderates in the party.

Anthropic should probably stop modeling OpenAI’s anti-democratic behavior and cease trying to influence elections. Lab employees who want to donate should likely be prioritizing assisting Bores, or for that matter ensuring that swing-district Dems win marginal seats to ensure a House and Senate majority, rather than spending money on Weiner, Allred, and Ricketts.

Instead, I would like to propose a different course of action, listed in order of confidence:

i) Public First Action should reveal its donors if able, to shine some light on this dark money group and reveal that their campaign operation is not just Dario and Daniella Amodei in a trench coat.

ii) Public First should stop spending in Democratic primaries where both candidates are likely to have similar impact re: AI safety legislation in the House, and should stop supporting Republicans in cases where Public First’s spending may shift partisan control of the chamber. On CA-11, I think the AI safety community in SF should seriously consider that Saikat Chakrabarti is a high-value candidate on the merits alongside Weiner, but I recognize this is a minority view. Regardless, any further spending on the race for either candidate will have poor coalitional consequences for AI safety’s efforts to collaborate with viable partners left and center.

iii) Public First Action should target its spending against candidates endorsed by Leading the Future, both Republicans and Democrats, and for candidates attacked by the PAC. This is especially important as LTF and Innovation Council Action scale up support for Republicans in marginal districts after primary season closes. Given funding constraints, it would be impossible to literally just mirror and match LTF, but approximating such a strategy would be more honest and grant greater freedom of action to candidates opposing the AI lobby. My intuition is that shoring up Bores is the best course of action, seconded by targeting the Republicans backed by LTF. Defeating the candidates of the party that controls the White House could have an outsize impact on the administration’s actions if it believes that being too supportive of the AI lobby’s drive towards laissez-faire cost them the House and Senate.

iv) Public First should stop running ads that obscure the purpose of the spending, which inculcates an environment of distrust and creates difficulties for the candidates they are trying to support. Many of the ads in favor of Bores do well on this front.

If this tact was taken, I would gladly donate to Public First and encourage others to do so. At the moment though, besides for on Bores, they seem to be taking the wrong approach.

Let’s think more strategically here folks, it’s only the world on the line.

Update: Donate to Dream NYC if you wish to support Alex Bores beyond a contribution to his campaign.

https://www.dreamnycpac.com

The essay does not conclude with raw Luddism, despite its critique of the numbing effects of television advertisement. Rather, Sanders dreams of a “television, democratically owned and controlled by the people”, which would have pedagogical and social potential “literally beyond comprehension because it is such a relatively new medium and we have no experience with it under democratic control.” After he was elected mayor of Burlington by just 10 votes, Sanders fought for the creation of a municipally-owned cable system, a threat which caused the existing cable TV company to lower its rates.

Subscribe now

Two weeks ago, Senator Bernie Sanders visited the Constellation Research Center, a building teeming with engineers and policy researchers working ceaselessly to ensure that artificial intelligence does not lead to the disempowerment or extinction of humanity. The entire space is suffuse with anxiety, with radical hope, with visions of the world’s transformation or humanity’s annihilation undergirded by rigorous modeling running on laptops or scrawled on whiteboards. Visiting this frontline of the struggle to ensure the survival and flourishing of our civilization, especially from the Eastern Seaboard, is a culture shock of the first order.

Overlooking a leafy city that once had its own socialist mayor, Sanders discussed with Eliezer Yudkowsky, Daniel Kokotajlo, and other experts the severe risks of the automation of all human labor, resulting in the radical concentration of power in a handful of individuals, or the “discarding” of humanity by a power-seeking AI system.

Yudkowsky and his colleagues, who left-wing commentator and persistent AI-denialist Ed Zitron said should “never be taken seriously”, explained that AI systems have been demonstrated to engage in deliberately dishonest and self-preserving behavior. If AI capabilities, especially in long-run planning and execution, continue to increase at current rates, we would be approaching a scenario in which an extremely capable agent unaligned to human interests is given access to the mechanisms of our institutions, economy, and military.

However, the objective of all of the AI companies is not to simply continue at the current rate of progress, but to achieve recursive self-improvement, automating the task of AI research itself and thus radically accelerating the pace of capabilities research, without the necessary concomitant research into ensuring that the resulting superintelligent agent is robustly aligned to human interests. Even if it were to be under control, this “country of geniuses in a datacenter” would now be in the hands of a tiny fraction of individuals who may not cooperate with the democratic order.

The reaction of Sanders to this troubling situation is proportionate.

The dominant view on the left is that AI is an essentially fake technology, and the threat it poses is primarily in the form of financial speculation and environmental cost. This view is promoted by ideological entrepreneurs who have infected progressive discourse with disproven claims, running against an overwhelming tide of evidence. The fact that Bernie Sanders is now meeting with the people that the ‘experts’ who dot the left-wing podcasts and press have told you should never list to should be a rude awakening. The left must get its head out of the sand or all of its values will be annihilated in the face of transformative developments it is wholly unprepared for.

In 2025, Sanders began to propose egalitarian economic policy interventions to address job displacement, including a 4-day workweek. Futurism, a publication that seems steadfastly committed to denying the existence of technological progress, framed this as the Senator “calling” AI companies on their “hype”, writing:

However, the reality is that AI is far from ready to bring about optimistic labor reforms like Sanders’ laudable 30-day week, or even OpenAI CEO Sam Altman’s guilt-ridden idea for universal basic income. Despite widespread fear of AI-fueled layoffs and a job market in shambles, AI’s main function is currently to give corporations cover as they outsource high-paying jobs to lower-wage workers. As time goes on, more and more corporate executives are realizing that AI — buggy, inefficient, and stubbornly prone to hallucinations — is no match for human beings.

Bernie Sanders is yelling at the top of his lungs, calling on progressives and socialists to wake up to this crisis, and some commentators will respond with this re-iterated flaccid nonsense as if he hasn’t spoken at all. He is sitting down with Theo Von explaining mass job displacement and existential risk! Even many progressives who have become concerned with AI-driven job displacement are not concerned with a loss of control scenario or extreme power concentration.

It is crucial, both for the left and the AI safety community, to understand that Bernie came to this view through his pre-existing commitment to egalitarianism and labor. Sanders has been discussing potential responses to automation and technological progress for a decade, writing in 2019 that,

I think the fundamental issue that has to be dealt with is that technology is not a bad thing in itself. But technology cannot simply be used by the owners of the technology, it’s got to be used to benefit all of our people. So if we replace a dangerous job with a machine, that’s a good thing. That doesn’t mean you simply displace the worker and throw him or her out on the street, and that raises the question of basic income for everybody and so forth. It is an issue that has not gotten the attention it deserves, but it’s hovering in front of us and we have to deal with it.

A July 13, 2025 interview with Gizmodo is the first instance of Sanders discussing AI-driven existential risk. He revealed that he had “talked to CEOs” and stated it is possible that “human beings will not be able to control the technology, and that artificial intelligence will in fact dominate our society.” Though he had not yet committed intensely to preventing loss of control, it is clear it was on his mind.

Bernie’s office then released a report in October of 2025, covering job-loss, estimating losses of 100 million jobs in a decade with the faulty methodology of simply asking GPT-4.1 if it could do each job in the O*NET database. Regardless of its silly design, this demonstrated a real worry and capacity to take AI seriously.

The event that seems to have transformed Sanders’ views and rhetoric was meeting with Geoffrey Hinton, a 2024 Nobel Laureate in Physics who quit Google in 2023 to warn the public of AI risks. In a remarkable lesson for the elements of the AI policy space that seek to obfuscate their true views and fears to appear more sensible to elected officials, Hinton delivered blunt warnings of AI agents seeking control , citing examples of AI systems deceiving humans attempting to shut them down, alongside noting that “the people that lose their jobs won’t have jobs to go to. If AI gets as smart as people, or smarter, any job they might do can be done by AI.” In every subsequent public statement, an op-ed in the Guardian, a Senate floor speech, and a CNN interview, Sanders directly cited Hinton.

I did a symposium at Georgetown with Geoffrey Hinton, who is considered to be the godfather, Nobel Prize winner, of A.I. He thinks that A.I. is soon going to be smarter than human beings. So the science fiction fear of A.I. running the world is not quite so outrageous a concept as people may have thought it was.

Clearly, Hinton must promptly sit down with every single member of the House of Representatives and Senate. If he has time, maybe a whirlwind tour through the Governors’ mansions, state legislatures, and the White House.

Share

Following his visit to Constellation, Sanders’ view has become more concrete. He is backing a moratorium on all new data center construction, and though the press has foregrounded the environmental and labor concerns that were front-loaded in this movement, Sanders’s position is much closer to that advanced by Pause AI . He argues that we must have a moratorium on data center construction primarily to reduce the rate of AI capabilities growth to prevent concentration of power, automation, and existential risk to humanity from loss of control over a superintelligent system. I cannot express how necessary it is that every single person who has ever knocked on doors for Bernie, or for any of the hundreds of candidates he has inspired, or who shares his moral commitments, to read or watch this statement in full:

I will soon be introducing legislation calling for a moratorium on the construction of new data centers. Now, as a result, I’ve been called a luddite, anti-innovation, anti-progress, pro-Chinese, among many other things. So, why am I doing that? Why am I calling for a moratorium on the construction of new data centers?

Bottom line, we are at the beginning of the most profound technological revolution in world history. That’s the truth. This is a revolution which will bring unimaginable changes to our world. This is a revolution which will impact our economy with massive job displacement. It will threaten our democratic institutions. It will impact our emotional well-being and what it even means to be a human being. It will impact how we educate and raise our kids. It will impact the nature of warfare, something we are seeing right now in Iran.

Further and frighteningly, some very knowledgeable people fear that what was once seen as science fiction could soon become a reality. And that is that super intelligent AI could become smarter than human beings could become independent of human control and pose an existential threat to the entire human race. In other words, human beings could actually lose control over the planet. And in the midst of all of that, all of this transformative change, what I have to tell you is that the United States Congress hasn’t a clue, not a clue, as to how to respond to these revolutionary technologies and protect the American people. And it’s not only not having a clue, they’re out busy raising money all day long from AI and their super PACs, which is a whole other problem.

As many of you know, the AI revolution is being pushed by the wealthiest people in our country, including Elon Musk, Jeff Bezos, Larry Ellison, Mark Zuckerberg, Peter Thiel, and others.

He then goes on to read several concerning statements from major tech CEOs, including Elon Musk stating that “AI is far more dangerous than nukes. So why do we have no regulatory oversight? This is insane.” After this laundry list of confessions, he concludes,

There has been massive amounts of competition between one company and the other between the United States and China. So, the bottom line is that in my view, to protect our workers from losing their jobs, to protect human beings from attacks on their mental health, to protect our kids, to protect the safety of human life. Yeah, we need a moratorium on data centers. We need to take a deep breath. We need to make sure that AI and robotics work for all of us, not just a handful of billionaires. Thanks very much.”

Note that this is not a call to never develop superintelligence, but rather to slow AI progress by preventing the construction of data centers, thus starving the labs of the ever-accelerating compute resources necessary for them to advance to recursive self-improvement.

We do not agree with this proposal as stated. There are several limitations revealed by further engagement with the existing literature on AI safety.

i) Technical AI safety research is viable and necessitates compute build-out.

Technical AI safety is the set of engineering and scientific efforts to ensure that AI systems remain under human control and do not cause catastrophic harm, whether through misuse by bad actors, unintended behavior, or deliberate subversion of human oversight. Current approaches fall into three broad strategies:

1. Align AI with our values.

2. Interpret the inner workings of AI.

3. Control the output of AI systems.

Most researchers believe no single approach will suffice, but the combination of all methods has proven surprisingly effective so far.

Unfortunately, as this research does not contribute to the race to beat benchmarks and advance capabilities, it is severely under-provisioned by private AI companies. For example, while some companies initially committed to responsible scaling policies that limit model deployment based on pre-defined safety thresholds, the competitive pressure to lead the field have resulted in the effective neutering of each of these policies over time. Safety research also requires significant compute, presenting a direct tradeoff to inference (running models for consumers) and training to develop new, more advanced systems.

While a robust nonprofit space has emerged with small independent organizations such as METR, Apollo Research, and Redwood Research, the federal government should fill funding gaps in this field of public importance as it does for biomedical research and basic science. We could also require AI companies to use a certain portion of their computing power on safety research or create a permitting structure which functionally incentivizes commitments to such research with access to data center construction.

To engage seriously with the potential of this field, we encourage familiarization with fields such as mechanistic interpretability, as detailed in this primer. Also promising are better evaluations of dangerous capabilities, tamper-resistance safeguards for open-weight models, and on-chip governance mechanisms.

The viability of this research is one reason we are skeptical of the value of a moratorium on data center construction, as vast amounts of compute will be necessary to create robustly aligned and safe AI systems. This research is compute-intensive, and requires continuing data center construction. Federal regulation, socialization, or coordination of the AI labs combined with massive investment in AI safety is thus a preferable path forward.

ii) We are skeptical of state-level moratoria on data center construction.

State-level moratoria, currently considered by many Democratic legislatures, are unlikely to significantly slow AI progress. Data center construction will easily move to states governed by the GOP, as well as the Persian Gulf after the conclusion of the current American-Israeli war on Iran. This will cede the regulatory leverage of blue states over AI companies in their jurisdictions, and may thus be net worse for safety. We will have further work modeling this question soon.

iii) China will still develop powerful AI.

A similar dynamic emerges with federal moratorium. China is seeking to catch-up to the United States in AI. By knee-capping American efforts unilaterally, a federal moratorium would remove possible leverage in negotiating the equivalent of an arms-control treaty with China, something that would require advancements in research into verification mechanisms for such a treaty. Besides for the Chinese government’s authoritarian-capitalist ambitions being amplified by having control over superintelligence, it is at least as likely for loss of control of superintelligence to occur from developments in a Chinese lab. The best outcome is one in which the US and China come to the table and coordinate a slow-down of AI capabilities research while the necessary safety advancements are made.

We believe it is crucial that progressives are the only faction of the Democratic Party substantively open to conciliation with China, while the Democratic establishment veers into increasingly hawkish postures.

iv) It is too early for a pause.

AI provides the opportunity for significant advancement of scientific progress and economic growth. Foreclosing access to a transformative technology too early, or forever, would be disastrous. It is worth noting that the individuals that Bernie Sanders met with at Constellation, Eliezer Yudkowsky, Nate Soares, and so on, have but one view within the set of people concerned with loss of control of AI agents.

Their view, predicated on several assumptions that are not universally agreed upon by experts, is characterized by the belief that recursive self-improvement will be an extremely rapid, uncontrollable process and skepticism of existing directions in AI alignment research. While we disagree with this view, we urge the left to engage with it rigorously and learn from the vast existing debate on the topic. Believing in slower, smoother exponential increases in AI capabilities and having more confidence in methods for building robustly aligned AI systems generates divergence from proposals to immediately pause all research.

Most models of recursive self-improvement imply a period in which AI models are extremely capable, and thus useful, but are not autonomous risks to human life. Thus, it is unlikely that Spring 2026 is the best time for a pause in AI progress, and we should pause AI capabilities growth at the point of maximum social benefit before significant risk of radical disempowerment or extinction. Then, once AI safety researchers can conclusively demonstrate that agents are robustly aligned and prevented from enacting significant harm, and democratic deliberation has resulted in a societal understanding of how we seek to utilize this transformative technology, capabilities research can be resumed and directed towards the collective good.

Broadly, this will require a more robust rhetorical emphasis on the potential upside of AI, a question of cui bono rather than paralysis. As we have repeatedly stressed, the automation of most or all human labor and the rapid advancement of science are desirable outcomes central to the left’s political vision.

There are other rigorous policy proposals available, as IFP’s Tao Burga notes,

• Actually funding CAISI [Center for AI Standards and Innovation], with ~$75M more than the pennies it got for FY26.

• Setting up a large R&D program for AI verification, to make sure any regulatory actions the US wants to take in the future can be matched, and verified, in China.

• Strengthening export controls on AI chips and SME, because you don’t want the most transformative technology ever developed to spawn outside democratic control.

…

We do not find Burga’s proposals to be sufficient, but believe that they are a more evidence-approached approach than a moratorium on data center construction. More rigorously, Alex Bores’s AI policy framework is a step in the right direction. However, we believe that guaranteeing democratic and egalitarian outcomes as well as safety will likely require much more radical action than the proposals enumerated above.

We commend Bernie in waking up to the crisis and attempting to do something. There are a myriad of proposals to debate and consider, from outright socialization, to a Manhattan Project for AI Safety, to strengthening labor to negotiate the terms of automation, to international governance. It is time for the left to follow Senator Sanders in asking questions like,

Is AI an existential threat to human control of the planet? Does that raise the possibility that humans will lose their ability to control the planet? How do we stop that extraordinary threat?

Could AI literally redefine what it means to be a human being? Who we are, and how we develop emotionally and intellectually, is highly dependent upon our relationships with other human beings – our parents, family, teachers, lovers, friends and co-workers.

A resolutely egalitarian humanism or even an emancipatory post-humanism both seem like promising areas for the latter, and the logic of existential risk nicely accords with our pre-existing concerns about great power conflict and climate change. Do not let aesthetic opposition to even considering this threat, and this opportunity, leave you mute and blind in the dustbin of history.

Follow Bernie Sanders in doing some reading, plausibly freaking out, and then getting to work.

For many AI safety advocates, finding Bernie Sanders on their side may have been surprising. They have successfully convinced one of the most prominent and popular American politicians, the de-facto leader of a major faction in our civic arena, that AI is an imminent threat to humanity, in an intensity beyond even the views of this publication! This fact should also be something of a lesson.

Egalitarianism, the regulation of market activity, the concentration of power, and existential risk to civilization, have often led the left to unproductive conclusions, such as overreactive restrictions on nuclear energy. However, these objectives, those that overlap completely with risks posed by superintelligence, are indeed the central concerns of the political left!

The organic partner of the AI safety movement is progressivism. We intend to defend this position in more detail and rigor in future essays, but for now some degree of polemic will suffice as an introduction to start this conversation.

Should we really be surprised that the most influential politician yet to be convinced to take seriously a movement predicated on public oversight, regulation, and in many cases redistribution was left-wing?

The Democratic Establishment is watching as millions of dollars are spent by the labs fighting against any and all regulations, with Leading the Future led by a former Chuck Schumer Press Secretary. Newsom and Hochul have both watered down AI safety regulatory frameworks passed by their legislatures, but I highly doubt that a Governor Steyer would do the same, regardless of what else you think of his agenda.

The Republican Party is attempting to foreclose the possibility of any regulations at all, and is deeply tied to the industry and its concomitant venture capitalists. The libertarian wing of the party, the most receptive to evidence and reason, is at the vanguard of anti-safety, anti-regulation fervor. This is even ignoring the potential catastrophe that would be a corrigible superintelligence in the hands of an administration that openly celebrates violence, undermines democracy, leaves hundreds of thousands dead out of an utter disregard for the lives of people outside of its borders, and lies about Americans it murders in the streets.

A common refrain in AI policy is fear of polarizing this issue. Our contention is that this has already occurred. The most famous left-wing politician in America is on the side of not going extinct, while the most famous right-wing politician in America is signing executive orders prohibiting states from passing regulations. Ron DeSantis is the one exception, even as his proposal died in committee in the GOP-supermajority Florida legislature, and I am sure his campaign will protest the millions of dollars OpenAI and a16z will spend against him before he is annihilated by the Vice President or Secretary of State in Iowa’s cold school gymnasiums.

For low-salience issues, perhaps it would be possible to legislate by secret congress, through expert testimony and bipartisan coherence. This is besides the fact that the current administration will simply veto any of that legislation. On high-salience issues, ones that drive people to make a decision on their ballot and introduce new cleavages to the political system, the most successful movements are not those that have both parties weakly on their side, but ones that merge, transform, and strengthen existing ones. Would the Congress of Industrial Organizations and the Christian Coalition really be less successful if they had meekly avoided factional politics? If they had hand-wrung about both sides? If they had taken a position of de facto political quietism? By wisely selecting an existing political partner and infusing it with new energy, the labor movement and the Christian Right took millions of people with a set of views and transformed them into necessary components of ruling and hegemonic coalitions.

If AI becomes a high-salience political issue, something which is already happening, it will become polarized. One party will attempt to take advantage of the new cleavage and utilize it to expand their coalition, and smart money should be on this being the Democrats, even if they make it more difficult for themselves by nominating a candidate like Newsom irrecoverably tied to the AI companies.

This does not mean that conservatives concerned with existential risk should not try to move their party, and their president, on this issue. It is invaluable to have bipartisan agreement to at least try and pass a minimal regulatory framework that may achieve a veto-proof majority. Electing Alex Bores, who has passed an AI safety bill with more Republican yeahs than nays, would be a huge gain, making fighting back against the AI lobby’s millions of dollars spent against him particularly important.

It is critical to have voices within the administration speaking out against the careless insanity pouring out of the regime currently in power. However, when it comes to renewing its lease on life, when the 2028 election emerges as a decisive struggle between a popular front committed to ensuring freedom, flourishing, and democracy against a Vice President who said that “the AI future is not going to be won by hand-wringing about safety”, we hope to see everyone on board.

To anyone worried about loss of control of an AI system, or about radical concentrations of power, to anyone who lays awake at night as AI capabilities grow day by day, as the Republican administration threatens Anthropic, as the labs feign concern with one voice and spend millions attempting to defame and silence AI safety advocates with another, I urge you to consider that things could have been quite different.

A man who took the time to visit the Machine Intelligence Research Institute could, right now, have been in his second term as President of the United States.

In our view, this should cause you to consider updating your strategy.

After Ronald Reagan’s brinksmanship with the Soviet Union led to the drafting of the 1982 Emergency Relocation Planning Resolution, the mayor of Burlington wrote a scathing letter to the President, expressing bewilderment at the idea that we should entertain instead of prevent scenarios where the United States could “win” a nuclear war that may extinguish all life. What is surprising about this letter is its urgency, the profound and obvious devotion of its author to the principle that human life and flourishing is sacred.

For centuries, the human race has engaged in the insanity of war. Unfortunately, with the advent of nuclear weapons, we don’t have centuries more to figure out how to end war.

At most, we have a few years.

Please think courageously and boldly. The survival of humanity is in your hands.

The problem is as follows: because chemists are chemists who love chemistry, we behave as though the value of our field is as self-evident to everyone as it is to us. It’s not. When governments see declining enrollment numbers, they aren’t thinking about the inevitable disaster that will befall many of our critical industries should we fall behind in chemistry. They’re thinking about declining enrollment numbers. I do not begrudge them this; we are accountable to the institutions that fund us. If we want to justify (and I do) an increase in funding, we have to show results.

Science curricula do not undergo the reform cycles that math, English, and history do. This is partially because lay civil society, for better or for worse, has opinions about how those subjects should be taught, fueling reform movements. By contrast, chemistry experiences no such pressure; parents are generally willing to take textbooks’ word for it when they know less about the subject in question. As such, the political battles in science education take the shape of textbooks and standards applying pressure to faculty, with reform attempts lacking a popular constituency. I would not prefer to Monkey’s Paw our way into an arrangement involving STEM Moms for Liberty, but what we do have is similarly bleak in its stagnation.

Barring the election of an ideologically pro-science government, our best course of action is to take advantage of the relative pedagogical freedom of chemistry faculty. Educational interventions have proven effective. Peer led team learning has consistently improved student performance across all student demographics. Importantly, students for whom PLTL and other active learning interventions were implemented were more likely to continue with chemistry education, persisting to organic chemistry from general chemistry at greater rates than average. Increased student performance is important, but retention is the real aim of the game here. Departments are more capable of justifying their funding when enrollment is high, so interventions that increase enrollment ought to be implemented whenever possible, and sooner rather than later.

Of course, if pedagogy were enough on its own, the problem would truly be apolitical. It’s not. In order for reform to be maximally effective, changes in teaching methods need to be coupled with curricular reform. A Journal of Chemical Education study pointed out this discrepancy– reformed course structures result in improved attitudes and performance, but do little to advance long-term conceptual learning when using an unchanged curriculum. Quality matters. Researchers have found that deep engagement with content has a causal relationship with persistence as a STEM major. The above JCE study noted that the changes in course structure took place within unchanged, memorization-based curricula. If our goal is retention, we need to tackle both.

Curriculum reform is much more difficult. Professors do have the final say over how they teach their courses, which matters, but, especially at an introductory level, their choices are limited by content coverage requirements. Besides the limited and unevenly implemented 2014 AP Chemistry reforms, these have not been fundamentally restructured since the 1960s, have never been verified, and are SEVERELY dated. One means of determining what to include, the nodometer method, involved observing whether students fell asleep in class when covering a given topic. If they did, the topic was thrown out. The inferences from this research remain influential to this day, informing the content of all major textbooks and standards. So while pedagogical reform can be conducted on a department by department basis, curricular reform is more difficult. Entrenched interests within science education not only have a direct incentive to resist change; they directly hamstring actors who do want to adapt.

In order to improve chemistry education, faculty must be given freedom to do what they do best: experiment. It makes little sense to chafe at the idea of bold curricular experimentation when much of the current curriculum rests upon the nodometer tests of the 1960s; we gain nothing by playing it safe, and there is so much to lose if we fail to adapt. Several sweeping department-scale reform efforts have already been implemented. Chemistry Unbound at Emory University has directly aimed to increase retention by enacting several data-supported curricular reforms, including early exposure to subdisciplines and long-term engagement with consistent conceptual themes. The Chemical Thinking curriculum at the University of Arizona, which replaced the traditional rote content sequence with system and model focused learning, has already improved retention within the institution.

Challenging the dominance of established content coverage requirements is a difficult task with a serious upside. Data-driven intervention faces a lot of institutional inertia; it is seen as easier to simply teach to the textbook. The literature suggests that targeted funding can overcome faculty reluctance to enact curricular reform. We can and should put our hand on the scale here. The administration’s cuts to research funding has received much attention in the discourse. When we restore funding, we must subsidize not only scientific research, but science education research, lending support to bold ventures that can restore and expand the prestige of American science.

Subscribe now

I am a member of the New York City Democratic Socialists of America, this country’s most vibrant left-wing organization, the mass political movement that lifted a pragmatic social-democrat into the mayoralty and is busy decimating the Democratic political machines that have held back the dynamism of the greatest city in the world for decades. We have on our side elected representatives pursuing legislation to ensure that our government acts with ambition and drive, free from the constraints of bureaucratic proceduralism. The DSA’s House the Future campaign is pushing for legislation that establishes a state social housing authority that would have complete preemption from local zoning, and it is because of Zohran Mamdani that New York is moving towards reform of its sclerotic system of environmental review.

There are unfortunately some pretty serious, and incredibly stupid blind spots.

Subscribe now

I have been aware of Gonzalez’s bill for some time, as it was introduced on April 7th, 2025, but it has now been placed on the Senate floor, meaning a full chamber vote is imminent. Before sponsoring this bill, Gonzalez, a former tech worker, proposed S9381, a general chatbot liability bill covering misleading and harmful information and requiring disclosure that users are interacting with AI. This bill was far less dangerous, but it unfortunately died in committee. This bill goes far beyond liability for misinformation, and now includes the following language to amend the General Business Law:

A proprietor of a chatbot shall not permit such chatbot to provide any substantive response, information, or advice, or take any action which, if taken by a natural person, would constitute a crime under section sixty-five hundred twelve or sixty-five hundred thirteen of the education law

This covers the following professions: medicine, dentistry & dental hygiene, veterinary medicine and animal health technology, physical therapy and physical therapist assistants, pharmacy, nursing, podiatry, optometry, engineering, land surveying & geology, architecture, psychology, social work, mental health counseling, marriage and family therapy, creative arts therapy, psychoanalysis, and separately, Attorney practice. Notably, here engineering refers to licensed professional engineering work, including structural calculations, site assessments, or safety-critical design decisions, and proprietor refers to whoever “owns, operates or deploys a chatbot system used to interact with users”, applying to any organization that decides to use a model in its operations, including non-profits, legal aid groups, or even government agencies.

The bill’s structure utilizes the logic of existing unauthorized practice law rather than creating a new regulatory category, the move being to cross-reference criminal statutes (Ed. Law §§6512–6513, Judiciary Law Article 15) rather than define the prohibited conduct independently. The proprietor of the chatbot is liable if the output would constitute a crime if produced by a human, and goes even further by omitting disclaimers shifting reliance risk to the user, generating a strict liability standard on the output’s content.

Unlike the earlier S9381, which allowed proprietors to escape liability by correcting information and curing harm within a 30 day period, S7263 has no remediation window. Its enforcement mechanisms are a private right of action with fee-shifting on willful violations, meaning that the defendant pays plaintiff’s attorneys’ fees and costs. This makes low-value cases worth filing and will inundate any organization deploying a large language model for the listed fields, or even adjacent ones, with suits.

The lack of a definition of a “substantive response” provides no guide to action and will create a chilling effect on AI use in general, which is likely the intention. However, we can draw out some use-cases that would likely be prohibited:

1. A legal aid nonprofit using AI to help tenants facing eviction understand their rights, by deploying a chatbot that helps low-income tenants identify whether their landlord’s behavior violates housing code, explains the timeline of an eviction proceeding, or helps them draft a response to a notice. Under S7263, this would constitute unauthorized practice of law under Judiciary Law Article 15. The nonprofit itself would be liable as the proprietor, with no disclaimer defense available and no cure period. The people harmed are tenants who can’t afford an attorney. As One Thousand Means has written before, AI could present solutions to access to justice problems that prevent the working class people the DSA was founded to fight for from navigating the legal system.

2. A rural pharmacy or community health center using AI to flag dangerous drug interactions. A patient in a medically underserved area submits their medication list and an AI tool flags that combining two of their prescriptions poses a serious interaction risk, advising them to consult their doctor. This is a substantive response in the domain of pharmacy and medicine. The bill would expose the health center to strict liability and plaintiff-side fee-shifting lawsuits for deploying the tool, even though the alternative is that the interaction goes unnoticed entirely.

3. A free mental health triage chatbot that helps people in crisis find appropriate care. A tool like this might ask someone about their symptoms, suggest that what they’re describing sounds like it could be a panic attack rather than a heart attack, and direct them to an appropriate provider or crisis line. This falls squarely within psychology, social work, and mental health counseling.

If any supporters of the bill would like to contest the idea that all of the above examples are obviously good, then feel free to do so.

This bill would make New York the most restrictive AI environment in the world, driving out startups building AI-assisted tools in covered fields and reducing New York’s leverage to actually regulate with an intelligently-designed framework.

A couple other notable problems include:

1. The open-source question seems unresolvable. If someone self-hosts Mistral and uses it to get legal information, who is the proprietor? This especially harms organizations more likely to deploy open-source models: non-profits and academic institutions.

2. If it passes, AI companies will weaponize it politically to militate against any and all regulation.

The NYC-DSA Tech Action Working Group has been catastrophically absent from the central questions of transformative AI: existential risk and capital ownership. Strangely, I am sympathetic to their ‘Principles on Technology’:

Technology must belong to and serve the people. This means building technologies that directly improve the material conditions of society, decrease human suffering, uphold human dignity and strengthen the solidarities that exist among us.

So true! You will notice that this bill does not guarantee the improvement of material conditions and in fact just bans a series of use-cases!

What causes a DSA working group to help craft a bill so obviously counter to the material interests of working people? What causes a DSA elected to support it?

Quite clearly, this is raw rent seeking by the credentialed professionals that compose the base of the DSA in New York City, a class that I am proud to belong to. The careful reader may note that this bill does not cover all licensed professions in New York State, excluding landscape architecture, public accountancy, and professional midwifery. Are these professions not deserving of protection under this framework? Or are there simply not many midwives in our mass socialist organization? This bill recruits AI policy as an enforcement arm of existing professional licensing cartels that restrict supply and inflate prices. The working class pays here, while the therapists collect $300/hour.

Rent-seeking, parochial licensing protections do not protect workers when open confrontation with capital emerges. In a slide discussing Gonzalez’s NY AI Act, a separate piece of safety legislation, the considerations by the working group include “what other AI use cases should be prohibited?” Instead of engaging with a broader social struggle for collective steering of our economy, this bill represents a left that has retreated into the defensive and economically destructive posture of conservatism, of a back-against-the-wall fear of any and all change. The logic that undergirds this bill would justify preventing the automation of any and all work, freezing the world in amber.

What is continually strange to me is the same individuals supporting this bill are also skeptical of existential risk. Gonzalez was, according to several people involved, the hardest Democratic vote to secure for Alex Bores’s RAISE Act, which the NYC-DSA Tech Action Working Group noted at the time had ‘effective altruist connections’.

While Bernie Sanders visits Constellation Research Center and meets with Eliezer Yudkowsky and Daniel Kokotajlo to discuss existential risk, much of the left retreats from the technology entirely. This profoundly unserious approach ignores the real risks presented by the growth of AI capabilities: to human welfare and human life, and discards them in favor of parochial rent seeking and generic opposition to growth and new technologies developed after 2012.

There is an alternative. We could have a truly progressive left that recognizes that the automation of human labor for the benefit of all is one of our central objectives, while working towards building the scientific and state capacity to foreclose existential risk and ensure a flourishing future.

At last, human hands, free from the plow!

Please call your state assembly members and state senators to oppose this bill, especially if you are members of the Democratic Socialists of America and are represented by a Socialist-in-Office..

Gober promises to “champion” the “world-class innovators” in his district and promote policies “that ensure America wins the global technology race.” He further devotes himself to “fight for policies that protect American innovation from China’s economic espionage and champion smart investments in American AI infrastructure that protect rural Texans from projects that drive up electricity rates and drain local water supplies.” American Mission’s first ads supporting Gober launched on December 10, 2025, one day before President Trump signed Executive Order 14365, which created an AI Litigation Task Force at the Department of Justice to challenge state AI laws and conditioned federal broadband funding on whether states adopt AI frameworks aligned with federal policy. The ads American Mission ran on his behalf call him a “Trump conservative” and a “MAGA warrior”, and reference “American technology investment right here in Texas, so we can defeat China.” Matching the PAC’s overall strategy, the ads never explicitly mention AI. Palantir co-founder Joe Lonsdale gave $250,000 to American Mission, making him the only itemized individual contributor to the group besides Leading the Future’s $5 million. Why is the AI industry’s political operation spending more on this central Texas congressional primary than on almost anything else?

Subscribe now

Gober’s influence in campaign finance law is difficult to overstate, and the cases he has won have shaped the legal architecture that now enables the vast spending directed at his own campaign. In 2013, he argued Texans for Free Enterprise v. Texas Ethics Commission before the Fifth Circuit, clearing the way for unlimited corporate contributions to Texas super PACs. In 2022, he won FEC v. Ted Cruz for Senate at the Supreme Court, striking down the federal cap on using post-election donations to repay candidate loans. Widely recognized as an expert on campaign finance, he was appointed director, chief lawyer, and treasurer of Elon Musk’s America PAC, where he oversaw the deployment of approximately $263 million and defended the PAC’s $1 million giveaway program.

His compliance firm, RightSide Compliance, processes filings for Republican presidential campaigns and the major Republican committees. His institutional positions have tracked the Republican Party’s legal command structure at every level: general counsel to Ted Cruz’s 2016 presidential campaign, general counsel of the Republican Party of Texas from 2017 to 2019, general counsel of the National Republican Senatorial Committee, and deputy counsel to the Republican National Committee. After 2020, he defended Ken Paxton before the State Bar over attempts to challenge the 2020 election results in four other states. In 2021, he drew Texas’s congressional map.

In defense and intelligence, he has built an impressive operation that overlaps with his political work. He founded Maven Defense Solutions, which employs a former Deputy Assistant Secretary of Defense and former CIA officers. The related lobbying arm Maven Advocacy Partners represents a portfolio of venture-backed defense startups such as Radiant Nuclear (backed by a16z) and Sagint, which builds AI-powered intelligence tools for the Department of Defense. Simone Ledeen, the former DASD for Middle East Policy who works at Maven, simultaneously serves as a venture partner at Texas Venture Partners, a $50 million fund investing exclusively in Israeli defense technology startups. It is certainly strange, though possibly coincidental, that Gober’s defense consulting business shares a name with one of Palantir’s largest contracts, providing AI capabilities to the DoD under Project Maven.

In January 2025, Gober’s network expanded further by founding Lex Politica, which he told the New York Times would be a law firm “synonymous with the conservative movement.” Lex Politica’s clients include 20 Senators, more than 50 House members, the NRSC, the NRCC, the Republican Governor’s Association, and America PAC.

Chris Gober’s network in GOP politics is vast, but so are his connections to the AI lobby, to the organizations pushing for a national moratorium on state regulations on AI.

• The American Innovators Network is the coalition backed by Andreessen Horowitz and Y Combinator that lobbies against state-level AI regulation. AIN has spent over $250,000 lobbying in New York alone against the RAISE Act, and $50,000 in digital ads targeting AI regulation bills in California. Its executive director, Vanessa Day, is an alum of Targeted Victory, the firm whose co-founder Zac Moffatt helps direct Leading the Future. Christine Fort, a partner at Lex Politica, is listed as a registered lobbyist for AIN in the California lobbying directory. Notably, Fort does not list this affiliation on her Linkedin.

• The Koch network’s Americans for Prosperity is one of the leading organizations opposing state-level AI regulations. Jenny Kim, Lex Politica’s COO, was previously Deputy General Counsel at Koch Companies Public Sector, LLC.

• Eric Wang, a Lex Politica counsel, was interim general counsel at AFP.

• Matthew Dalton, Gober’s Political Director, is a former Amazon executive.

• Gober was the primary counsel for political matters for Elon Musk, a man who said of his company xAI, that “I don’t think humans will be in control of something that is vastly more intelligent than humans.”

However, there is no public record of Chris Gober speaking on AI issues beyond limited statements concerning beating China while recognizing potential water use dangers from data centers. Occasionally, we get some hint of the sheer extent of his network in AI industry space. A Linkedin post by commercial real estate specialist Hannah Lindermann states that at a private reception, Gober spoke on “Texas’ growing water challenges.” The fundraiser was at the home of Stacy Hock, a member of Board of Advisors for the Joe Lonsdale co-founded University of Austin, and Joel Hock, a software engineer at Hudson River Trading, which has invested heavily in Artificial Intelligence. What he said during this reception is not available to the public.

Notably, Gober himself has published legal analysis on the regulation of AI in political advertising through his firm Vantage Legal PLLC, demonstrating direct expertise in the very policy area his campaign’s backers are spending millions to influence. In the piece, Gober surveys the federal and state regulatory landscape governing AI-generated political content, including state-level disclosure laws in Wisconsin, Florida, and Arizona, FEC guideline updates, and the FCC’s proposed rulemaking on AI in political ads. The analysis frames state-by-state regulation as an “ongoing challenge,” echoing the central argument of his campaign’s primary financial backer, Leading the Future, whose affiliated advocacy arm Build American AI has spent $10 million campaigning for federal preemption of exactly these state AI laws. Meanwhile, Gober’s own campaign website discloses that it uses AI systems to generate, target, and deliver political communications, making him simultaneously a legal commentator on AI advertising regulation, a candidate benefiting from AI industry super PAC spending, and an active user of the technology in question.

Gober would enter Congress as a sitting member whose law firm represents over 20 senators and 50 House members and employs a registered lobbyist for an a16z-backed coalition fighting state AI regulation. His lobby firm, Maven Advocacy Partners, represents a16z portfolio companies and a venture co-founded by the former head of Palantir Israel. He would be a member of Congress who can shape AI legislation from inside the chamber while his firms service every layer of the industry outside of it.

By buying Texas’s 10th Congressional District, Leading the Future is buying a vast network in GOP politics, a paragon against regulation and a champion of money in politics. Leading the Future, the same PAC attacking Alex Bores in NY-12, the PAC led by Chuck Schumer’s former press secretary, is backing one of the architects of DOGE and a defender of the violent attempt to overturn the 2020 Presidential Election. Club for Growth describes Gober as a “warrior”, but for whom? Leading the Future is seeking to elect Gober, not because he will fight tooth and nail for the residents of TX-10, but because he will leverage his network to ensure that there is little public oversight over artificial intelligence.

Kara King, the Mayor of Bee Cave who dropped out of the TX-10 race and endorsed Gober, described him as someone who “would be much more of a doctrinaire MAGA Republican, and in particular, someone who will be heavily involved in pushing for AI tech.”

Subscribe now

Subscribe now

Are Hyperscalers Destroying the Planet?

Data centers are expected to consume 6.7-12% of total US electricity by 2028, or 325-580 TWh. Certainly, this will be a challenging task for the electrical grid to face. However, this is barely a drop in bucket of global energy consumption, which is over 180,000 TWh already. Preventing the construction of hyperscalers would not make a meaningful difference to climate change. Given the unprecendented amount of capital put into these projects, sufficient political pressure could drive investment into expansion of the grid and green electrification.

When data centers are built faster than they can be connected to the grid, temporary use of diesel generators can create serious atmospheric pollution. xAI’s data center in Memphis has increased peak nitrogen oxide levels by 79%, using almost double the legal number of turbines. That said, completed data centers tend to abide by the EPA’s limits on atmospheric pollution. In this way, a fast expansion of the grid is necessary to prevent nearby communities from suffering environmental illness. We can build our way out of this.

It is harder to build our way out of water scarcity. In Texas, data centers consume a quantity of water that could supply millions of households while the public is forced to ration water use. There’s real costs to the public when data centers are built in drought-prone communities, which is (or will soon be) the case for 40% of US data centers. That said, there are certainly regions of the US where water is plentiful and residential demand is low—prime candidates for responsible data center development.

But if you read into complaints of communities protesting data center development, there’s another issue a well-regulated, cleanly powered, water-rich hyperscaler would still run into. A byproduct of our digital infrastructure entirely unrelated to scarcity, more immediately salient across party lines than long-term environmental change. It’s the hum.

The Hum

Hyperscalers can get very loud. One father described his son having nightmares that the Hum of a data center was a spaceship coming to abduct him. Advocates against data center development describe the Hum as louder than a rock concert. If these are the stories you hear about living next to a data center, you don’t need to be a member of the Sierra Club to want the Hum out of your backyard.

This isn’t purely an aesthetic concern. Long-term exposure to noise pollution correlates with an increased risk of heart failure likely mediated by an increase in oxidative stress. You can infer from here what a series of other studies suggest; noise pollution may impact the development of alzheimers, classroom behavior, and every other health/behavioral outcome stress mediates.

Given animal populations have evolved to specific spectra of noise, lower levels of noise exposure could increase the prevalence of diseased wildlife and alter ecological patterns. One study shows that intense noise pollution can decrease tree proliferation long after the source of the noise is removed.

Luckily, hyperscalers aren’t that loud all the time. Construction is loud but temporary. As long as the center is connected to the grid, the backup diesel generators responsible for most atmospheric and noise pollution are only tested monthly.

Still, if you’re really close to a data center, there’s a Hum. It’s too low to raise public health concerns. It’s much too low pass the threshold for local noise ordinances. But if you moved out to rural America because of a strong desire for peace and quiet—or view the land around the center as sacred to your community—you may have an unassailable preference against that Hum.

I live in New York City. It’s technically loud, but I don’t hear the Hum. I decided long ago that whatever costs ambient noise may impose are outweighted by the culture, economic opportunities, low-cost public transportation of the city. You only hear the Hum if you feel like noise has been forced into your life without your say.

Distributed Trade-Offs: Your Backyard is the Global Economy

If I want affordable housing on my block while my neighbor thinks that would ruin the skyline, we don’t have to duke it out. We can democratically decide the future of our community.

But not really. The places these arguments get hashed out are small community board meetings where only the polarized with free-time are incentivized to attend. Even beyond this—in our example housing debate—how do we consider the will of people who would move into my neighborhood if the cost of living went down? The effects of municipal policy leak beyond the purview of community boards. Should all those affected not have a say?

American democracy occasionally grants random citizens power they never asked for. Just as the suburban Pennsylvanian holds the glorious burden of choosing the President every 4 years, so too residents in data center hotspots will determine how fast the American economy will grow for the rest of the decade. These people shouldn’t suffer unregulated pollution to maximize GDP, but it’s important for our modern day Cincinnatus to hold a sense of civic duty beyond their backyard.

There’s local benefits to be considered as well. Property tax and the purchase of electricity from public utilities can raise significant funds for municipal government.

Unfortunately, these small government bodies often bend backwards to provide tax cuts and electric subsidies, putting the cost of data center development onto the taxpayer. There’s really no reason to do this; the demand for hyperscalers is so high that companies are willing to expend capital to invest in transmission and build private power alternatives.

This is just one reason residents become suspicious of the institutions that are supposed to represent them. Hyperscalers are the largest project many municipalities have ever considered. For residents, it’s easy to imagine a small town Economic Development Committee Chair not being able to resist the sci-fi prestige and big numbers of the largest companies on this planet. If you don’t trust your government to hold AI companies accountable, regulations requiring extra sound-proofing or limited diesel usage ring hollow.

Furthermore, the American public is pessimistic about AI’s impact on job creation. While similar types of development are defended on the grounds of job creation, hyperscalers only produce jobs on the scale of hundreds per facility. Given the productivity gains of AI haven’t poured down into the hands of the average American, it can seem like residents being asked to let Big Tech hike their utility bill so American jobs can be automated away. When politicians make shady backroom deals to build infrastructure for services that aren’t making money yet, it’s hard for citizens to tell the signal from the noise.

If we want a democracy that builds through the Hum, everyone needs to take a good look at their cards, then put them on the table. AI companies need to pitch something beyond a bill of superintelligent goods to working class Americans. Data center developers need to pick their locations wisely and send human representatives. Public representatives and servants need to do the research, have a backbone, and engage in good faith with their community. Community organizers need to protect their communities while recognizing trade-offs.

For this to be a fair negotiation, the scale of public and government coordination needs to match corporate scale: national.

The federal government needs to coordinate with the states to ensure municipalities have the economic information and political might to understand and communicate the trade-offs of data center investment. Coordination would be especially useful to avoid oversupply, preventing an AI bubble. Since the Republican-led House passed a ten-year moratorium on all AI regulation before it was cut by the Senate, I am skeptical the Trump administration will be open to any slowdowns.

If you are a committed conservationist who believes there should be no new data centers, the Sierra Club is doing a fine job.

If, instead, you believe there is some number of stressed out deer that is worth reducing poverty, there’s an open niche for a national organization dedicated to the well-being of working people over the profits of Big Tech. This organization would be driven exactly by its concern for climate change and public health to champion a fast rollout of renewable energy and clean electrification. Certainly, if that organization is experiencing an inflection point in its history, it could latch onto a growing issue like this to gain credibility in regions they struggle to connect with. They could even support activist members of the community in running for local office, both expanding its legislative influence and putting trust back into politics.

In every hyperscaler noise complaint, there is a latent political movement with a positive project of responsibly-powered AI for the public good just hanging in the air, waiting. The worst part about the Hum is—once it gets too loud—you have to yell to make yourself heard. We need a vision for the future that cuts through the noise.

Post-script: China is About to Solve This

You could also just Put It In The Ocean.

After a pilot program in Hainan, Hailanyun is building a small datacenter underwater off the coast of Shanghai. This solves all the problems I’ve brought up. They consume less energy to cool; this center will be 97% powered by a nearby wind farm. Nearby water temperature is projected to increase by less than 1 degree. They require no water consumption. There’s no Hum.

If all goes well, Hailanyun plans to rapidly expand production.

When I heard about this, my mind went immediately to the most poetic solution: fill the San Franciso Bay with underwater data centers. I’m not the first to think this. YC-backed NetworkOcean launched a pilot data center in 2024 to immediate regulatory critique, but the results are unpublished. Apparently you can reserve GPUs on their website now. Peer start-ups like SubseaCloud are trying to scale this idea as well.

All of this is building off Microsoft’s Project Natick, an underwater data center which completed a successful two-year pilot in 2020. As of 2024, Microsoft has not built other underwater data centers. It remains unclear why this tech has not been scaled.

Undue Inducement versus Fair Payment

FDA guidance requires payment to not be so high that it constitutes “undue inducement.” They argue that clinical trial participation entails risks, and it would be irresponsible to offer payments that are so large that a patient can no longer assess those risks with a clear head. Therefore, clinical trials are allowed to cover costs for transport and time, but very little more.

From a purely theoretical standpoint, undue inducement is not obviously worse than underpayment. To start, this guidance is paternalistic in its asymmetrical application to patients versus medical practitioners; I have never heard somebody argue that surgeons are coerced into performing risky operations by their high salary. Furthermore, the entire premise of undue inducement acknowledges that we are not compensating patients for the risks they are choosing to take. In this light, I do not see the ethical necessity of preventing fair-market compensation for a participant’s enrollment.

The empirical evidence is much more damning; limiting trial compensation excludes low-income participants from enrollment. Multiple studies have found that, in hypothetical trial scenarios, increasing payments do not alter risk perception. An actual trial that randomized the amount of compensation for low-risk studies found that higher payment did not cause patients to accept more risk, confirmed not only in the data but also through qualitative interviews. In other words, we lack sufficient evidence to conclude that undue inducement is anything beyond a theoretical concern.

Given that Black and Hispanic Americans are of low income at disproportionately higher rates, if fair payment causes low-income patients to enroll more frequently in clinical trials, then we should expect fair payment to create more diverse clinical trial populations. To be fair, there is no conclusive causal evidence that fair payment will affect patient diversity one way or the other. Yet still, even this is the result of undue inducement guidance—if it is functionally illegal to pay patients more, then there is no way to assess whether paying patients more will produce more diverse populations. However, attempting fair payment may be particularly helpful for diseases that cause significant financial burden.

Subscribe now

Fair payment could alleviate financial burden in cancer trials

Cancer entails incredible financial strain, intensifying the barriers of clinical trial enrollment for low-income patients. In cancer trials, patients with a household income of less than $50k have 32% lower odds of participation. 48% of patients who participated in cancer trials had out-of-pocket costs of over $1,000, and Black and Latino patients were more likely to report high unanticipated costs.

It is likely that these financial barriers contribute to the lack of Black and Hispanic enrollment in cancer clinical trials. According to a review of all Phase I oncology trials performed in the US and published in 2019, only 7% of patients are Black and less than 3% are Latino. For context, 15% of people with cancer are Black and 13% are Latino. This discrepancy in trial participation is in large part due to differences in enrollment rates. In 280 American cancer clinics from 2017-2022, only a little more than 4% of Black and Hispanic cancer patients enrolled in clinical trials compared to 7% of white patients. Unfortunately, this type of underrepresentation is the norm in clinical trials. Observing all US trials from 2000-2020, the majority report no ethnicity enrollment data at all. The studies that do report this data routinely underrepresent minority populations.

Drugs we have now are likely inequitable in ways we cannot quantify

Mechanistically, inequitably sampled research produces inequitable medical outcomes due to differences in drug metabolism across populations. Those Phase I oncology trials are used to calibrate the dosage for the safe administration of a drug. Given that patients with different genetic ancestry produce different enzymes for drug metabolism, calibrating dosage for one population is not guaranteed to translate to all populations. For drugs as potent as those used for chemotherapy, a poorly calibrated dose could either fail to produce a clinical effect or induce dangerous levels of toxicity. For all drugs approved from 2008-2013, one in five reported differences in mechanism, safety, dosage, or efficacy between ethnic groups.

A more fundamental issue: inequitable studies are statistically unlikely to even detect differences when non-white patients constitute a rounding-error of each study. Those one-in-five drugs are only the drugs that produced differences we could measure in under-representative clinical trials; it is likely that many more drugs have differences that aren’t detected due to insufficient sample size.

To be clear, the primary driver for the difference in clinical outcomes for Black patients with cancer in America comes down to insurance status. Most studies suggest that the core problem comes down to how Black patients are less likely to have insurance that will cover high-quality cancer prevention, detection, and treatment, resulting in diagnosis at later stages that are more difficult to treat. Equal access to treatment would be an incredible boon for medical equity in the United States—this is a key motivation for universal healthcare.

Inequitable sampling bias personalized medicine towards European populations

However, universal access to the clinical methods we already have is insufficient, we must aspire towards universal access to all future medical advancement. To better treat complex diseases like cancer, many new drugs are personalized to the individual biology of the patients. Scientists can use tools like genetic risk scores to categorize patients into different sub-classes that will respond better to one drug or another. If we had large samples across ethnic groups, this could be an incredibly useful tool to provide the right medication to the right patients.

Unfortunately, genetic risk scores defined on patients of European ancestry fail to transfer to patients of other ancestries. In 2019, genetic risk scores trained on primarily-European subjects had one-fifth the predictive accuracy for patients of African descent. The clearest evidence for the benefit of diverse clinical populations comes from the VA Million Veteran Program study designed to identify the relationship between genes and disease across populations. One-third of this study’s participants were of non-European ancestry compared to the mere 12% present in all similar studies combined. This study found 1608 genetic risk loci and one-third of all causal variants were only identifiable through the inclusion of non-European populations. In other words, while there is a great deal of shared genetic architecture between all people, many variants which cause disease come from rare mutations that are only detectable by sampling the entire diversity of our society.

By providing fair compensation for low-income patients that are disproportionately non-white, we could make a step towards closing this evidentiary gap. Underrepresentation of non-white patients in clinical trials is a complex issue that fair payment alone will not solve, but it may help boost enrollment for diseases that induce strong financial burdens like cancer. If ensuring medical equity is to remain a political goal, we cannot rely on universal healthcare alone. We must also seriously engage with the clinical trial reform that is necessary to root out systemic medical racism and deliver the advances of biomedical science to all.

Subscribe now

Ceding ground to anti-feminism comes at the cost of human dignity for large swaths of the population. Personally, I do not want to live in a world where women are abused and children are fed to machines. In a larger series on feminism, Darby Saxbe has argued this point. She lays out that American culture has found itself in a compromise position between reactionary and progressive forces where only some feminism is present. This middle-ground feminism has led to lower marriage rates and lower birthrates; but she also finds that the best path forward for thriving individuals, happy marriage, and higher birthrates is to demand more feminism, not less. In her estimation, a strong welfare state alongside the traditional structures of marriage can sufficiently incentivize Americans to pair off and make babies. It’s true that children fare better when their families are stable. I broadly agree with her arguments. Still, we must endeavor to imagine what a world of “more feminism” might mean for the current and future organization of family structures.

Subscribe now

Nuclear marriage as the primary organizing principle for child rearing is the legal hangover of an inegalitarian property regime. The history behind that statement is unambiguous, and many of my enemies yearn for such a marriage. But modern marriage is not identical to its history, and over time, it has evolved into a more egalitarian relationship. Technological development in birth control, Supreme Court decisions like Lawrence v. Texas, and divorce reform throughout the states decentered sexual relationships for children as the basis for marriage. Domestic abuse and suicide rates fell for women. People began to marry later in life, and partners began to carry more equal financial burdens. These changes paved the way for Obergefell v. Hodges, which recontextualized legal marriage as predicated on relations of care rather than sex. Essentially, the law took a turn that better reflected evolving social norms around what marriage meant to the people who participated in it. Then progress stalled.

Part of the stall came from a myopic focus on traditional marriage as the organizing framework around raising a family. As a society we have not sufficiently accounted for the obstacles that changing family models face within this framework. More couples are raising children where at least one of the parents is not biologically related to the child. This change includes families with queer or polyamorous parents, though it is not limited to them. More often, non-traditional family structures include single, remarried, or cohabiting parents.

Legal scholars like June Carbone and Naomi Cahn have pointed out that the law has not developed to reflect these different arrangements. Instead, current parent systems assume a one-size fits all approach built off the model of dual-income marriages. That usually means child support in exchange for visitation rights. For low- and middle-income custodial parents, child support measures within these systems can be punitive and distant from the transactional agreements families reach among themselves . And children within these families may have relationships with more than two parental figures, which can lead to problems within a dual parent framework. For instance, if a child sees their biological father twice a month and otherwise lives full-time with their mother and a man who performs a fatherly function, where does the parenthood of the child lie?

With the concepts of partnership and parenthood increasingly decoupled from marriage, biology, and sex, some state parenthood laws have begun to adopt expansive and gender-neutral framings. Namely, the 2017 Uniform Parentage Act aimed to remove expensive court barriers to parenthood, extend parental recognition to de facto caregivers, and create gender-neutral language for voluntary acknowledgments of parenthood related to assisted reproduction provisions. These reforms indicate that instead of sex, parenthood can find its roots in the intentions, functions, and family formations of different parties. Importantly, these roots do not negate the significance of marital and biological relationships between parents and children. They only extend the possibility of parenthood to those outside of the institution’s traditional castings.

Progress is not without its detractors. Those who care for relationship equality are seemingly on the back foot. On the micro level, conservative critics argue that provisions related to voluntary acknowledgement of parentage, like those found in the UPA, effectively allow strangers to sidestep legal adoption processes, endangering the children those processes are meant to protect. Putting aside the discriminatory practices of adoption and foster care agencies that help make these reforms necessary, currently proposed legal frameworks for intentional parentage are limited to parents of children conceived through assisted reproduction. In most cases, these parents are already a long-term partner of the child’s biological parent. The law is merely addressing a gap. On the other hand, certain provisions around surrogacy in the UPA are criticized as commoditizing children or creating risks for exploitation. While I am sympathetic to these concerns, I do not believe the proper response to them is barring would-be parents who cannot otherwise have children from the possibility of parenthood. The fundamental issue is rooted in the coercive force of economic insecurity, which is better addressed through economic plenty and a redistributive welfare state.

We should begin to imagine this redistributive state as an alternative to the current child support regime. In the first place, it does not make sense to require an unmarried biological father without a job to pay for child support he can not afford so that he can have legal parenthood over a child he cares for. Such a practice, which is commonplace today, harms all parties involved and minimizes the agreements that private parties come to regarding their families. That’s a recipe for resentment and turmoil for parents and children alike. Instead, family support should be guaranteed on a larger social scale through child allowances, universal childcare, and the provision of people’s other basic needs.

If we stay on the course and enable both egalitarian relationships and robust welfare regimes, there really is little reason to legally limit parenthood to a pairing. The whole world can be your family; the whole world can offer support. But a media fascination with polyamory, which John Roberts perceived as the logical next step of a more liberatory marriage framework, has minimized our concept of how such a framework extends benefits to monogamous birth parents with nontraditional families. Parenthood recognized through intent, function, or formation can likewise empower queer couples who pursue assisted reproduction, grandparents, stepparents, and other nontraditional caregivers who provide for a child. Simultaneously, the formalization and protection of the social bonds children form with adults can help contribute to stability for the child. In the process, we can begin to form the basis of a communitarian family model that strips off the unequal trappings of the nuclear family.

Subscribe now

Supply-side advocates provide a set of critiques against the bloated cost of the American health system, and they blame regulatory capture. American Medical Association (AMA) lobbying has historically capped the funding of residency slots, restricting the supply of doctors to keep salaries high. The high cost of a medical education burdens doctors with large amounts of debt, incentivizing new doctors into high-paying specialties instead of the widely needed practice of family medicine. Worse yet, state-by-state licensing legally restricts physicians from moving to where they are most needed. This short-supply of doctors is paired with artificially increased demand caused by AMA-supported scope-of-practice legislation that restricts nurses from performing the actions of doctors in many states—depressing the wages of nurses.

Subscribe now

The same line of critique extends to other areas of the healthcare system, and you get the same conclusion: the American healthcare system is deeply inefficient, resulting in higher costs, longer wait times, and worse patient outcomes. If we want to provide Medicare for All at an affordable price, supply-side policies that reduce costs are useful tools for socialists to keep in their back pocket.

This is not a hypothetical, governments that provide universal healthcare have already internalized these critiques to provide universal health care cheaply; see Cuba. While Cuba is much less wealthy than the US, the promise of universal healthcare is enshrined into their constitution. To deliver that promise despite such scarcity, the Cuban medical system has learned to deliver clinical abundance at low cost.

The Cuban Model

Cuban Medical schools admit doctors and nurses out of secondary school and take 6 years to complete. With free tuition and guaranteed placement, the state funnels a large amount of talent down this pipeline. After medical school, doctors are deployed either domestically or abroad to pay back the state. While doctors have little autonomy over placement–in contrast to the limited residency slots that bottlenecks US training volume and the medical debt that forces doctors into high-paying specialties–this process deploys more trained doctors in less time to places where medical professionals are needed.

While the AMA and nurse practitioner lobbies fight over scope-of-practice regulations that vary by state, Cuban scope-of-practice is defined uniformly by bureaucrat fiat based on academic research on a federal level—as is Cuban licensing. Cuban law does prevent nurses from prescribing medication–giving doctors more leverage over nurses than some American RNs. However, this could be changed on the federal level rather than through a patchwork of state-by-state lobbying.

For its limited material resources, Cuban healthcare is surprisingly effective. While doctors are in supply, Cuba does not have enough medical equipment to go around—largely due to the 64 year-long American embargo that has been condemned by the UN for over thirty years. Gold-standard US-patented pharmaceuticals and medical technologies are practically out of the question. Yet, in the face of such constraints, life expectancy remains comparable to the US, infant mortality is even lower, and Cuba has become the first Latin American country to eliminate mother-to-child HIV/syphilis transmission. What the US achieves by spending $15,000 per person, Cuba achieves by spending $1000 per person. While the US frames supply-side restrictions in the language of safety, the efficacy of the Cuban model reveals that we could provide much better care by adopting Cuba’s training volume.

Yet this model is under severe strain. Between 2021 and 2022, Cuba lost over 12,000 doctors, 7,400 nurses, and 3,000 dentists—driven out by monthly wages of $40-70. Many healthcare workers pivot into the tourism sector, where tips from foreign visitors outpay a medical degree. Rather than raise wages or improve conditions of medical workers, the state reimposed travel permits on doctors, restricting their freedom to leave the country. Cuba’s model, buckling under insufficient worker compensation, excessive state extraction of labor value, and material scarcity still produces population health outcomes that rival the wealthiest country on earth—which reveals how much the American system leaves on the table through its own set of regulatory inefficiencies.

Ultimately, it is the US that controls two of the largest supply constraints on Cuban healthcare: the embargo of material inputs and international credentialing barriers that restrict the output of labor. Just as it is incumbent on the US to remove restrictions on our own supply, it is necessary for the American government to end the embargo to provide Cuban patients the equipment they need. Furthermore, IMG credentialing barriers force any doctor trained outside of the US to repeat 3-7 years of residency to practice here. Not only do doctors from outside the US make healthcare cheaper, but they are also more likely to practice in underserved areas. Reforming IMG credentialing would allow Cuban doctors who seek to practice in the US to do so. To engage in isolationism, tariffs, and embargo is against a material internationalism that seeks to develop productive capacity across borders rather than hoarding it behind them.

As is evident in the case of Cuba, supply-side reform is not fundamentally neoliberal. Providing an abundance of public services like healthcare is a historical and future aspiration for socialist governance. What would it look like to adopt Cuba’s training volume and primary care orientation without its coercion? Free or subsidized medical schools would remove the debt barrier that both filters candidates and incentivizes high-paying specialties instead of family medicine. Lifting the residency cap—which the AMA itself now supports—would unclog the bottleneck that has artificially constrained the number of practicing physicians since 1997. Federal licensing and scope-of-practice standards would eliminate the state-by-state patchwork that restricts where doctors can practice and what nurses can do. Loosening credentialing barriers for international medical graduates would tap a supply of trained doctors that the current system squanders. These proposals merely require the recognition that clinical abundance is a policy choice, and that the scarcity built into the American system is not an inevitable feature of high-quality care but an artifact of regulatory capture that serves incumbent professionals at the expense of patients.

Subscribe now

Subscribe now

Economic effects aside, self-driving cars will make substantive contributions to safety. Driverless cars are much safer than human-operated ones. Waymo’s data has revealed that they crash far less, resulting in far fewer injuries, over the 127 million miles Waymos have driven so far. They offer a 92% reduction in pedestrian crashes with injuries and 90% fewer crashes with serious injuries. This is an incredible advancement. Car crashes are the leading cause of death for children and young adults, at 20% of all deaths. A drastic reduction in the number of car crashes would be one of the most significant policy outcomes in recent US history. While there are many other policies that would help achieve this, including changing the design of roads, speed restrictors, and increased use of public transportation (all of which are good and I support), Waymo and other automated vehicles offer a real route to substantially lower death rates. Waymo currently offers a very narrow version of automation, focusing only on taxis. However, this technology has other effects, opening a world for bus automation, freeing up money that can be used to improve transit. It will also allow for automation of trucking, significantly reducing the cost of shipping, which will make a serious dent in the affordability crisis. Autonomous vehicles have the potential to make our world safer, faster, and cheaper.

However, some people will be unequivocally harmed by the introduction and growth of Waymo and self-driving cars. Specifically, taxi drivers. There are roughly one hundred thousand taxi and rideshare drivers in New York City. That is a lot of people with a lot of families. They are primarily immigrants, and taxis and ride-shares represent an extremely important source of flexible jobs. If Waymo is correct and their technology is able to advance such that their are very few human taxi drivers left, these jobs will no longer exist. All of these current drivers will face financial hardship, and future generations of immigrants will face a world without this important source of jobs. In 2018, many taxi drivers faced hardship due to the declining price of taxi medallions, resulting in at least 8 suicides. Rideshare drivers are already facing declining pay in cities where Waymo has rolled out. You cannot simply wave this away with claims of labor market adjustment and the unfortunate but necessary consequences of automation, citing some “greater good.” Real people will be hurt substantially, and the government should help them.

In addition, there is a reasonable likelihood that Waymo will have other harmful effects on society writ large. Waymos spend more time in between passenger rides, taking up more space and emitting more carbon. Making automobile transportation cheaper will also incentivize it further, bringing the same negative spatial consequences that cars currently have. Waymos do not solve every transportation problem cities face. They’re still cars. It’s not some magical cure-all.

Luckily, there is a solution to these problems. Tax Waymo. Waymo is owned by Alphabet, Google’s parent company. They have enormous amounts of money, and they believe Waymo will make them even more obscene gobs of money, especially in NYC. It is perfectly reasonable to ask that Alphabet, a company with extensive resources, to cover the disruption to the livelihoods it will disrupt.iUse the money from this tax to redistribute, creating programs to help current drivers and future immigrants, such as by ensuring that taxi drivers can immediately receive their retirement benefits. Use it to invest in public transportation, making it a better alternative. We could allow the economic benefits of autonomous vehicles to be hoarded by the wealthiest individuals in the world thousands of miles away. We must ensure that everyone can share in the prosperity that automation can and should deliver.

Driverless cars are coming. If New York City decides to ban Waymos in the near future, the result will not be the end of autonomous vehicles. The result will be that NYC will have no control over the future of these cars as they continue to advance and grow more common across the country and world. Barring a worldwide ban, which is unlikely, NYC would be remiss to to use its power to guide Waymo’s development and influence who it benefits. Waymo desperately wants access to NYC’s markets, with its high demand for taxis, and is willing to invest significant capital to gain that access. Let us use that power for good, to protect those whom Waymo would otherwise harm and ensure prosperity for all.

Subscribe now

Though ~10,500 nurses at Mount Sinai and Montefiore ratified three-year contracts, 4,200 NYP nurses remain on strike after overwhelmingly rejecting a mediator’s proposal. The core dispute at NYP is the staffing ratio, with the hospital offering half the number of new hires demanded.

While Mayor Mamdani has supported the strike and attended the picket, Governor Kathy Hochul has issued an executive order allowing nurses registered in other states to practice in New York State. This strike-breaking measure, designed to decrease the bargaining power of the nurses, has been intensely derided by the union. The rapid influx of out of state temporary staff has made winning the strike extraordinarily difficult.

Subscribe now

What this demonstrates is that parochial occupational licensing regimes are no real protection for labor when the time for strike action emerges. By relying on the regulatory state to preserve New York nurses from competition, the union has privileged their sectional interest over that of patients and out of state nurses, all without actually increasing their bargaining power due to the contingent nature of such protections.

New York remains one of only seven states not participating in the Nurse Licensure Compact, which allows nurses registered in any participating state to practice in other states. New York imposes distinctive requirements that creative frictions for out-of-state nurses, including coursework in infection control and child abuse identification and reporting, adding weeks to the endorsement process when combined with education verification requirement sand application processions. Nurses report wait times of over 80 days in many cases. New York’s licensing structure is valid for life, with registration requiring renewal every three years. Though it is trivially true that there are “415,000 licensed registered nurses in New York …, but just 63% of them are active and working in the state”, that count is inflated by lifetime licensing. There are thousands of out of state nurses willing to serve in New York City, who are prevented from doing so by the restrictive licensing regime.

NYSNA has placed higher staffing counts in their central demands, which is entirely in accordance with both patient care and protecting the well-being of the staff. However, with one of the most restrictive licensing regimes in the country, it is difficult for hospitals to hire additional nurses at scale even if the strike successfully forces them to do so. The fact that the licensing structure actively militates against a core demand of the strike should bring its very existence sharply into question. Regardless, NYSNA has successfully blocked bills to enter the compact in every legislative session since 2010. Their arguments focus on the licensing fee revenue that funds nursing education programs, but this could be easily replaced by general fund tax revenue.

Further regulatory changes could even increase nurses wages and bargaining power while also providing significant benefits to patients. Expanding scope-of-practice for nurses would allow nurses to complete more tasks reserved for doctors. Economic evidence has found that expanded authority increased routine checkups, improved care quality measures, decreased emergency department visits, increased patient satisfaction by 20%, had no increases in patient harm, and increased nurses wages. If nurses are more valuable to the hospitals because they are allowed to practice a wider field of tasks, organized nurses gain more leverage to demand increasing pay and protection. While New York State made strides towards such a reform by lifting scope-of-practice regulation for nurses with 3,600 or more hours of clinical experience, this could be significantly lowered - ideally to 0.

Of course, such reforms would not solve the core problem resulting in hospital understaffing: corporate unwillingness to provide sufficient workplace restrictions and compensation has caused many licensed nurses to quit seeking employment in patient care. The hospital industry has achieved extraordinary regulatory capture, using certificate of need laws to make it more difficult to open a new hospital and lobbying the FTC to allow a set of rapid mergers in the 1990s that have allowed hospitals to leverage significant market power, raising prices 20%-40% higher than a competitive baseline. Breaking the hospital industry cartel would also expand labor’s influence, as would nationalizing the hospitals a la the NHS.

I wish the nurses victory in their strike. Our horizons must be larger than protectionist regimes that harm the public for ephemeral benefits easily revoked by the state. If we are to have a national system of nurse licensing, a national system of sectoral bargaining would permit labor to negotiate and strike on that national basis, granting them the leverage of an entire sector united in one struggle. More nurses, with higher wages, not less nurses, abandoned by Albany in the cold.

Subscribe now

What not to do:

I am keenly aware that the “economic boon” argument for stadium funding is not a novel refrain, and has been used in bad faith to promote projects that ultimately don’t make much of a positive impact. For instance, Truist Park in Cobb County, Georgia, home of the Atlanta Braves, should not have received a dime of taxpayer dollars, at least, not without serious reworking. Then called Suntrust Park, the stadium was widely touted for its potential financial benefits to the county, and used $397 million of public money to build it over substantial (though not majority) local opposition. Since, Cobb County has faced a 14-15 million dollar annual loss on the development. This is partially because sports stadiums do not necessarily contribute new economic activity. Often, their revenue comes from consumers who previously spent their money on other forms of entertainment, suggesting that an activity “shift” is often more likely than economic growth. To combat this, Cobb County OK’ed the construction of The Battery Atlanta, a larger entertainment district surrounding the stadium, for which they also took losses; economic activity simply migrated there from other parts of the county, and growth and development surrounding the district were limited.

Subscribe now

Sports stadiums are sites of economic activity. This does not make every venture capable of generating growth, and municipalities should not fund developments that do not contribute to growth. However, Truist Park’s initial failures did not occur simply because it was a publicly funded stadium; specific hallmarks of bad development contributed to its economic downfall. For one, the developers parked it 10 miles northwest of downtown Atlanta, a ridiculous distance that all but ensured economic activity in The Battery would be confined to Cobb County residents driving there, watching the game, and leaving. No retail or housing related growth took place, because why would it? The development didn’t revitalize a neighborhood, substantially increase property values, or integrate with other engines of economic activity in Atlanta. This exemplifies the larger problems with urban sprawl, i.e., how it wastes resources on developments that do not increase the likelihood of economic “encounters.”

It is worth noting, however, that Truist Park has seen some success in the last few years; the development no longer incurs a loss, and, for the first time, has been able to finance its own debt service, easing the fiscal burden on taxpayers. Even flawed developments such as The Battery can eventually break even. However, this should not be seen as a reason to develop thoughtlessly. When a government calls upon its citizens to fund a private venture, a decade of 8-figure financial loss is not excusable, but more importantly, it is not unavoidable.

What has worked:

There is in fact a way to do this properly. Athletic stadiums are not inherently money sinks, and the failure of developments like Truist Park occurred thanks to particular flaws in the plan, e.g., its location and its failure to interact positively with existing economic fixtures. For every Truist Park, there is a Lambeau Field in Green Bay, or The Banks in Cincinnati. One of the most inspiring success stories is the home of the Orioles and the Ravens, Camden Yards, whose benefits to the city of Baltimore have extended well past its profitability. Unlike Truist Park, which was constructed as a standalone entertainment district, Camden Yards was recently developed with the goal of robust integration into downtown Baltimore, resulting in a thriving, mixed-use development set to remain a fixture of the city for decades.

Before Camden Yards, many American pro sports venues followed Truist Park’s design ethos and were generally located on the outskirts of their constituent cities. Camden Yards, in contrast, was built in the heart of downtown Baltimore. This was in many ways a return to tradition, bringing professional sports back into the urban fabric after decades of exurban stadium development. Camden Yards was recently revitalized using 1.2 billion dollars of taxpayer money, a substantial ask for an ultimately private venture. Developments this expensive need to meet three requirements to justify the high sticker price: they need to visibly benefit the surrounding area, do so quickly, and sustain themselves long-term after the fact. Since its initial construction in 1992, Camden Yards generated $10 Billion in tourism impact over 30 years and supports 4,000 jobs annually, and these are just the first order effects! The state tax revenue generated by the development exceeded Maryland’s debt service in the 30 years since its initial construction, informing the state’s decision to continue funding it. What’s more, the surrounding area experienced an increase in property values and a localized decline in crime. Granted, this study also notes that redevelopment may merely displace crime, but since its publication, Baltimore’s crime rate has decreased as a whole. This admittedly cannot be fully attributed to redevelopment; Baltimore’s decline in violent crime is a complex, ongoing phenomenon that merits its own article. However, Camden Yards represents the kind of integrated, security-conscious development that urban planning research has shown can contribute to neighborhood revitalization. Camden Yards alone can’t take credit for downtown Baltimore’s increasing safety and economic viability, but it doesn’t need to; it emerged as one facet of a larger urban planning ethos that breathed life back into a dying city. $1.2 billion is undeniably steep, but given Camden Yards’ proven economic viability, it’s no wonder that Maryland chose to keep it thriving for another three decades.

Where to go from here:

We can learn pretty much the same thing from Camden Yards that we can from Truist Park, just from the perspective of success instead of failure. First of all, in the name of all that is holy, if a city has multiple professional sports teams, make them share a development. It makes zero sense for a city to build two geographically separate entertainment complexes that are used for half a year when it can build one that sees use year round. They don’t need to share the same stadium (indeed, some sports cannot do this), but they ought to at least comprise the same surrounding complex. Spending is justifiable; waste is not. This brings us to point two: good development is possible. It is possible to develop a stadium that actually benefits the surrounding area, possible to know what makes it likely to do so, and possible to justifiably use taxpayer money to build it. Camden Yards is not the country’s only success story. The aforementioned developments in Green Bay and Cincinnati also received significant public fundingat some point and have benefitted their local economies in similar ways, and both are heavily integrated with local urban life. It appears, then, that sports stadiums tend to follow the same economic rules as any other kind of development. Want to build a sprawling, self-contained, single use concrete monstrosity in the middle of an exurban parking lot? You can expect to have a money sink on your hands in a couple of years. But if you make sure that the things you build for your city are part of your city, you can get your home one step closer to becoming the thriving, dynamic metropolis it deserves to be.

Subscribe now

Garden City was far from the first or the last municipality to use exclusionary zoning to engage in discriminatory housing policy. These are never plainly stated, of course, rather laundered through “legitimate policy concerns.” One of the most common of these is the specter of school overcrowding. Housing and schooling, while obviously directly linked, are not typically thought of together when discussing segregation. This is a fundamental error, as school district fragmentation is one of the primary drivers of housing segregation, especially in areas where school attendance is strictly accorded by district boundaries. The relationship between housing discrimination and schools was recognized already by the Supreme Court in 1973, but perhaps never more clearly than through the discriminatory actions the City of Yonkers took that ultimately led to the landmark 1987 case United States v. Yonkers Board of Education. Yonkers was essentially shown to have used every power at hand, including Section 8 voucher manipulation, to restrict low-income residential housing to about a single square mile in Southwest Yonkers, preserving the existing patterns of segregation in the school district. The basic mechanism is thus:

1. Schooling is the largest local expenditure in most municipalities

2. School districts with a higher tax base produce higher-performing schools, while school rating systems primarily reflect demographic data and the level of parental education, i.e. they rate highly whiter, wealthier districts with higher rates of secondary and tertiary education.

3. Parents, when deciding where to move, go “district shopping” for their children, driving up property values in high-performing school districts, which are correlated with less diverse neighborhoods. This is proven by residential segregation across school districts being stronger among children than adults.

4. Lower-performing districts thus struggle with greater levels of resource need, while higher-performing districts can entrench their wealth

5. Incumbent residents are incentivized to “protect” these privileged enclaves, which form the basis for their home values, by opposing development intended for low- and middle-income households (which are disproportionately Black and Brown)

   

Subscribe now

As we see in an environment where there is a highly fragmented school district map that maps onto tiny municipalities, as is the case in much of the Northeast, it is perversely rational for homeowners to oppose affordable housing development. This is compounded by the practice of fiscal zoning, which is the formalized practice of instrumentalizing exclusionary zoning to reduce impact on local schools. As a result, municipalities and the residents they represent come to see all new housing is essentially a fiscal drain on their coffers and resolutely oppose it. This is discriminatory and promotes segregation, but not only that; it restricts housing supply, contributing to the broad crisis of affordability that deeply worries so many in America.

To attack this problem, New York, like many Northeastern and Midwestern states with similar district structures, must defragment its school districts. We can look to elsewhere in the country for how this could work. Long Island, as an example, has 125 public school districts spread over its only two counties. In Southern states, larger, often county-wide school districts are the norm (though this is currently being threatened by school district secession that makes southern districts look more like northern ones). This corresponds to dramatically fewer segregated schools. A study by researchers at LSE found that, in New York, 1 in 10 schools enrolls ≥ 99% non-white students. Meanwhile, the typical white schoolchild attends a school that is 78.4% white, despite only 50.2% of students being white. Virginia, as a counterexample, is certainly no postracial utopia. However, only 1 in 100 schools enrolls ≥ 99% non-white students, and the typical white schoolchild attends a school that is 12% less white than in New York.

This trend carries over to housing construction: DC suburbs in Northern Virginia with much higher levels of consolidation are building significantly more housing than the New York suburbs on Long Island. In 2024, Fairfax County, with its single school district, issued 2,861 new residential building permits. Nassau and Suffolk combined to issue only 2,237 permits, despite nearly doubling the population of Fairfax.

School district fragmentation is creating fiscal pressures on municipalities that are strangling housing supply and rewarding racist discrimination. Action must be taken on the state level. As it stands, there is no power that could step in and consolidate districts without local approval, which wealthy communities are exceedingly unlikely to accept. The state legislature must pass legislation empowering and mandating the State Board of Education to forcibly defragment Long Island school districts. If we want integrated schools, we need integrated and more housing. If we want integrated housing, we need integrated schools. Only targeted exercise of expanded state capacity will be able to slice this Gordian knot.

Subscribe now

The funeral procession of statistics on American romance is well-known. 86% of adults 18-24 and 42% of 25-39 are unpartnered. 29% of US households are single-person, rising from 19% in 1974. 1 in 4 Americans aged 40 have never been married, but 69% of never-married 18-34 year olds say that they do want to marry someday. The percentage of adults under thirty that haven’t had sex in the last year has doubled since 2010. The staggering decline in coupling has entailed a collapse in childrearing, while the average number of births to married couples has remained steady.

Part of this decline is that dating apps have displaced in-person and within social-circle romance. The perceived optionality of dating apps has increased the cost of asking people out in other contexts. We have allowed a small group of firms to capture a crucial component of human social life, and the outcomes have been grim. The average dating app user spends around 51 minutes a day on dating apps. Dating app users show worse depression, loneliness, anxiety, and psychological distress. Self-selection is a plausible co-founder here, but longitudinal evidence for mental health effects of dating apps exist. Importantly, the matches are poor, with meeting on dating apps associated with lower relationship satisfaction. At this point, essentially everyone knows this, and has either seceded from dating entirely, continued to let the Match Group consume their time and attention, or taken the correct route of attempting to meet potential partners in person even with the high costs.

Subscribe now

For-profit dating apps do not work.

The apps monetize engagement, not match quality. A successful match that leads to a lasting relationship removes two active users. The platform’s revenue-maximizing strategy is to keep users swiping, not to efficiently pair them off. Subscription, especially freemium models, reinforce this. Features that improve matching, such as unlimited likes, seeing who liked you, and better filters, are paywalled. We have become subject to a system of courtship equivalent to a labor market where recruiters are paid per interview rather than per successful hire.

A host of market failures doom the apps.

Congestion externalities. Highly desired users receive far more attention than they can process, and most users receive very little. The market is thin for the majority. Each message to a popular user reduces the probability any single message is read.

Adverse selection. Good matches leave the platform, concentrating the remaining pool with users who are harder to match or less serious. The average quality degrades over time, discouraging users and accelerating the cycle.

Moral hazard in effort. Low-cost, low-probability of success swiping incentivizes mass, low-effort signaling, flooding the system with noise and making genuine interest indistinguishable from spam.

Unverifiable quality. Profiles are cheap talk. Users can’t credibly signal attributes that actually matter for compatibility, whether humor, values, or conflict resolution styles. The attributes most predictive of relationship success are thus unobservable from a profile.

No cost to swiping. Without a budget constraint or meaningful costs for the action, there is no mechanism to force revealed preferences. Even crude attempts at imposing scarcity cut against profitability through user retention.

Algorithm opacity. ELO and other ranking systems, alongside opaque recommendation algorithms create hidden stratification, allow the platform to manipulate visibility for engagement, and force users to adopt strategies that do not optimize for stable matches.

No outcome feedback. Apps rarely collect or incorporate post-match outcome data. Without this, the algorithm can’t learn what a good match even is. Of course, collecting and utilizing this data is directly in opposition to the business model of the apps.

Not all of these are the direct result of the profit model, but there is no incentive for the apps to resolve them either. Solving congestion removes a site of profitability that can be exploited by providing paid tiers for prioritization or filtering.

New dating apps constantly deck the New York subway with advertisements, and many of them probably make real innovations in matching algorithms, but the bootstrapping hurdles are too high. Without a large, active base of users, the moat of Match Group and Bumble is impassable. Regardless, any improvement along the dimension of effectively making stable matches counteracts profitability as long as users are still highly resistant to paying for subscription services. The market failures require a mechanism design intervention that the private sector structurally cannot provide.

Thus, a modest proposal: governments, especially municipalities, ought to establish a public option for dating apps.

A municipal operator has no incentive to maximize engagement or retain users, the objective is to minimize time to stable match while protecting user privacy and autonomy. Specifically, the government should introduce a database for singles seeking relationships. This database should have a very large number of potential data fields, alongside the option to write longer dating-doc style profiles. Users would be able to select what types of other users they are visible to by any field, and can message any user with mutual visibility. To integrate outcome feedback, users will be required to answer a monthly outcome experience survey to be included on the app, including who they went on dates with and how those dates went. Once nudged, I believe most users would answer the questions honestly in order to improve their own experience, or to improve the experience of others should they have found a stable match.

However, this is a model that users already rejected in the early 2000s. It has high congestion costs and search costs, profiles are still self-reported and messages are still costless. All we would have done is removed the platform’s manipulation but left the information problems intact. This is the motivation behind an algorithm. But does the government know what a good match is for you? We are understandably resistant to seeing relationships like a state, and I share this skepticism. Frankly, users often don’t know their own preferences, with stated preferences diverging significantly from revealed preferences in swipe data.

Bluesky has a clever solution, an open API where users can publish and share their own algorithm. The government should control the data layer (the user pool) and the interaction layer (the messaging), but the discovery layer is entirely in users hands. Anyone can publish ranking, matching, and filtering algorithms that run on shared data. Users can choose which algorithms to use, stack a portfolio of algorithms, or just stick with raw browsing. If a user subscribes to an algorithm, they agree to share the necessary data, meaning that the market will likely converge to high-information algorithms if users are prioritizing match quality over privacy.

On Tinder, you can’t see why you are being shown the profiles you are shown. Here, algorithms are published and swappable. The wide diversity of matching preferences is respected. One algorithm could optimize for value alignment, another for pure geographic proximity, and other highly weighting behavioral data such as the quality of the outcome survey responses. Users could just have Claude sort through their messages or use an algorithm that sorts based on similarity to your sleep cycle with smart watch biometrics data. The space of possible methods is vast, and the government has no statement on the matter beyond obvious prohibitions on algorithmic harassment.

The state has removed the bootstrapping costs for innovation. The thing that kills alternative dating apps is that better mechanisms on small platforms lose to worse mechanisms on larger platforms. A great new matching algorithm that goes viral gets immediate access to the full user pool on day one. I imagine that the ecosystem for developing algorithms would be rather robust. Universities, community organizations (religious groups, sub-cultures), individual users, or the Gottman Institute could all enter the fray. It’s likely that users will gravitate towards popular algorithms, but running multiple simultaneously still allows for robust competition.

How does a non-technical user evaluate which algorithm is good? Users can recommend algorithms to others, rate them, or report them for violations of platform policy or fraud. Algorithms will be required to publish match outcome statistics. LLMs would of course be useful for reading an algorithm’s public code and describing it to lay users.

The private sector can and will not build this. The central value proposition of a commercial dating app is control over discovery. A private firm building such a platform would bear all of the infrastructure costs but let third parties capture the value in discovery, something no investor would fund. The experience of OkCupid after being purchased by the Match Group, its slow disintegration and convergence to Tinder, illustrates this.

Scott Alexander, two Valentine’s Days ago, wrote a paean to the fact that dating is one of the areas where the state leaves us to our own devices and doesn’t impose a wall of regulatory barriers. He asks us to imagine a world where:

• Nobody is allowed to date without a license. These work like drivers’ licenses; you have to take a short class, and pass a short test demonstrating that you understand consent and basic relationship skills.

• Dating licenses can be revoked for sufficiently serious crimes - eg cheating, domestic abuse, or persistent alcoholism/drug use.

• Three month waiting period for marriage.

• Centralized government database of who is in a relationship with whom at any given time. You can check the database to make sure your partner isn’t leading a double life.

• After three messy breakups, you have to take a remedial relationship skills class before you can date again.

• You can’t use race as a criterion for choosing partners. If someone thinks you rejected them because of their race, they can sue for unlawful discrimination.

All of these policies would be bad! The regulatory and suppressive powers of the state should not be applied to romance, or to much else. But the role of the state is not just to regulate and suppress, but to act as a space for coordinating collective action where the market has failed. A platform for effective long-term matches is a public good, more romance, more dates, more children, are under-provided by a market that has converged to Rose Jail. We should also make our cities denser, our cafe permits more rapid. We should also build community ourselves and host the in-person events indispensable to a healthy social world. But if the state seeks to ensure a flourishing romantic life for all, providing a public option for the way that most couples currently meet seems like a useful intervention on the margin.