Gambling on Trump for the AI Era: Nathan Labenz and Samuel Hammond Argue the Stakes

Nathan Labenz introduces two episodes, originally from the Moment of Zen feed, in which he gives a hearing to two AI-forward cases for Trump. He treats AI as a legitimate voting factor if AGI is a real possibility around 2027, and he says the election is, in his honest opinion, a referendum on Trump. The guest in the first episode is Samuel Hammond, a senior economist at the Foundation for American Innovation.

Labenz lists a number of AI-adjacent views he shares with Trump supporters: build more nuclear plants, take population decline seriously, protect free speech, loosen rules on building, favor abundance over degrowth, and prioritize high-skilled immigration. He also thinks big American tech companies should neither abuse market power nor be harassed for outcompeting legacy businesses.

Still, he gives three reasons for not backing Trump. He sees Trump as too inclined to escalate with China and put narrow domestic political interests first. He says the lack of rigor and discipline is a recipe for unnecessary risk in a volatile period. He also believes a positive-sum, inclusive mindset is needed for a new social contract in the AI era, and says he will vote for Kamala Harris primarily as a rejection of Trump.

On China, Labenz says there is a failure of leadership on both sides and rejects the idea that adversarial relations are inevitable, a view Joshua Steinman calls "the physics of the environment." He does not see China as a meaningful threat to the American way of life, though he does not excuse IP theft or the early COVID cover-up. He credits the Biden administration with decent domestic AI policy, including the 10^26 reporting threshold and subsidies for US chip fabs, but finds the chip ban and Harris's China rhetoric depressing.

He would rather compete with China on life expectancy, diseases eradicated or tonnage sent to Mars, and cites former OpenAI policy researcher Miles Brundage on proactively showing that Western AI development is benign. Between a hawkish Democratic establishment and Trump, he prefers the steady hand who might make costly arms-control-style deals. Trump's withdrawal from the Iran deal, inflammatory rhetoric and blanket-tariff plans lead him to expect things to get worse.

Labenz describes his P(doom) as 5 to 95 percent, meaning nobody has convinced him either that AI risk is negligible or inevitable. Today's best models understand human ethics fairly well, which is better than most safety people expected years ago. What matters, he says, is what we can shift the odds to, and he points to Kennedy overriding military advisers in the Cuban Missile Crisis as a case where executive character mattered.

He concludes that in a moment of crisis he finds it far likelier Trump would change things for the worse than for the better. He admits Harris has not impressed him on AI either, but he would choose the stable person who listens to experts over a bet on Trump overriding them. He also hopes for a positive-sum approach, praising OpenAI's stated vision and Sam Altman's basic-income experiments.

Hammond says he is not an effective altruist but can put on the hat, and that as a rationalist he tries to see past social desirability bias, including the social cost of supporting Trump. Taking a utilitarian view, he argues that what matters most in the long run is innovative capacity, since productivity improvements compound. He would ask which candidate is better for innovation at first order.

His first example is drug pricing. He argues that Harris's plan to cap prices using a basket of international prices would hurt, because US spending subsidizes global pharmaceutical R&D, and he says caps in the Inflation Reduction Act may already have delayed or killed up to 170 new drugs. He adds that getting AI policy right could swamp the costs or benefits of every other domestic policy.

Labenz replies that he thinks of Trump as a total wild card and that those policy details are small next to the tail risks effective altruists care about: nuclear war, a super-lethal pandemic, and who makes judgment calls as AI heats up. He recalls the downward trend in conflicts and worries that Trump might tamp down small spikes but set the conditions for a very big one. Harris's brief "two little letters" AI remark is not confidence-inspiring either, he says, but Trump seems all over the map.

Hammond argues that, especially if you are a doomer who thinks default paths go badly, the high variance of a Trump administration should count in its favor: there could be very high highs as well as low lows, and anything out of left field is likelier to come from it. He draws on Grossmann and Hopkins's book *Asymmetric Politics*, which sees Democrats as a coalition of social groups that bargain over concrete interest-group goals, and Republicans as a more ideological vehicle.

Because the Republican party is ideas-driven, he says, it sometimes does things that cut against its narrow interests, such as appointing the justices who overturned Roe, or Paul Ryan's tax bill restricting popular deductions. In the Trump administration, personnel were also drawn heavily from libertarian circles. Operation Warp Speed, he argues, could only have happened because Chicago-school economists at the FDA were already streamlining approvals and building public-private partnerships, which brought the vaccine forward five or six months and saved perhaps 200,000 lives.

He believes Trump on television differs from Trump in government, and that a future pandemic would see the same playbook. He notes Trump has called AI "super duper intelligence" and described it as alarming, while Biden named climate change as the biggest existential risk and Harris described AI risk in terms of denying healthcare. In his view, black-swan risks that are still theoretical can only be taken seriously through an ideological lens, not a stakeholder one.

Labenz concedes that a stakeholder-driven approach could block good applications of AI, since groups that stand to lose status will resist, and he is not confident about teachers' unions. He has, though, been pleasantly surprised by the medical establishment's response, perhaps because doctors are exhausted by paperwork. Still, he dislikes the idea of an ideological approach, because AI has to be understood on its own terms, ideally mechanistically, rather than through pre-loaded frameworks or analogies.

He notes that the smartest people he follows admit deep uncertainty, as in Anthropic's core-views post, and asks which Republican ideology Hammond even means: Chicago-school market maximalists, the Christian right, or neo-reactionaries. Hammond answers that by ideology he means simply driven by ideas, such as following the Constitution, resisting surveillance and censorship, and rolling back parts of the administrative state. He says the left's AI input is ideological too, pointing to who each party invited to the Senate's AI insight forum and to a Democratic appointee to an AI advisory committee.

Hammond argues that Biden's OMB directive on government AI, which could retroactively require audits even of something as simple as the post office's mail-routing tool, buys safety at the cost of state capacity. A right-wing project to "deconstruct the administrative state," he says, could instead speed AI diffusion in government. He tells the story of Balaji Srinivasan being shortlisted for FDA commissioner in 2016, with a plan to move from pre-market approval to post-market surveillance, and says only tech-aligned figures like Peter Thiel and Blake Masters running the transition made that possible.

He wants a "community notes version of everything" and argues that the Republican problem of lacking a technocratic elite, due to education polarization, is one the tech world can fill. In his view this will not come out of a stakeholder process, which is why he thinks an ideas-driven faction, if it has resources, is the more promising path.

Labenz finds Hammond's picture attractive but unrecognizable in today's party. He argues people are investing hope in a Trump-Silicon Valley romance that did not yield a high-competence government the first time, though he credits Operation Warp Speed. In his view the party has, at the top, a self-interested leader, and below him a Congress of religious zealots and opportunists, with a small old-guard remnant that cares about state capacity.

He says primaries are won by the most religious-right, abortion-focused candidates Trump endorses, who do not care about AI, and that the vision Hammond describes is not on the ballot in most districts. If you visited every member of Congress, he says, you would likely find more ideas-driven conversations among Democrats. He sees the real effective-altruist outlook as anti-Trump, since people motivated by scripture are not a good guide to AI.

Hammond calls this a media caricature. He says he works daily with Congressional offices on both sides and has been in Republican reading groups on topics like monopsony power, and cites work with Marco Rubio's staff on turning the Small Business Administration toward a development bank. He argues that Trump is essentially a populist New York Democrat with paleoconservative views on immigration and trade, uncomfortable with the social conservatives, and notes that Bannon and Tucker Carlson have raised alarms about AI, including calls to bomb data centers.

Hammond cites a Dwarkesh podcast conversation in which Patrick McKenzie described running vaccine logistics from a Discord server because no one was in charge. McKenzie's lesson was that officials should have picked the smartest person, made them a colonel, and told them to ignore everything else. Hammond argues the next president could preside over an AI takeoff, and the question is whether the person given authority is someone like Elon Musk or a committee shaped by the professional-managerial class, which he expects will resist AI diffusion and push deceleration aimed at use rather than real safety.

Labenz is a Musk fan but doubts that making things happen is what a crisis will require. He recalls Trump's pandemic press conferences, including the remark about injecting bleach, as evidence of someone not in command of the facts and polarizing the issue. He notes that blame for vaccine polarization lies partly with Democrats, but says someone who makes everyone around him worse is not the person to put in charge.

Hammond replies with a utilitarian point: Norman Borlaug plausibly saved hundreds of millions of lives, and a utilitarian should be glad he existed whatever his character. Labenz answers, "where's my Green Revolution from Trump?" and says the argument asks us to imagine competent executives this time without evidence from last time.

Hammond says the first administration's competence is underrated and that its staffing was chaotic because the team did not expect to win. Labenz points to the unprecedented number of senior officials now warning against his return. Hammond counters that Trump was fundamentally misaligned with those officials and that the departures partly reflect a new regime. He cites Michael Lind on the need for a counter-establishment rather than a counterculture.

Hammond compares the AI transition to the Industrial Revolution, which took governance from the world of 1900 to that of 1950. The new science of management developed in the first large corporations was absorbed into the administrative state, and he expects AI, which directly implicates the machinery of government, to force another such change. He says that is much harder to see coming from the Democratic side, whose horizons of possibility he describes as more constrained.

The discussion turns to immigration. Labenz notes that a 2016 article saying only Trump could fix immigration was by Hammond, and that Trump later helped kill a bipartisan border deal to keep the issue alive. Hammond argues that Trump's team wanted a points-based system, that Paul Ryan and Mitch McConnell talked him out of it in his first term, and that the deal would mainly have added asylum judges while relieving pressure for structural reform.

Labenz responds that holding issues hostage for a later structural win is cynical and an unhealthy approach to AI, and recalls Trump downplaying COVID in early 2020 as evidence he does not grasp exponentials. He calls Trump a tornado, destructive rather than constructive. Hammond argues that creative destruction drives prosperity in the private sector, but that the public sector has no profit-and-loss mechanism to retire failing agencies, so entropy builds up and some destruction is needed before rebuilding.

Hammond points to a leaked draft executive order from the America First Policy Institute that calls for Manhattan projects on AI and broad deregulation under an "America first in AI" framework. He took part, as a voluntary outside contributor, in a Project 2025 policy committee on AI, and says he was surprised by how many AI-safety-minded people were involved. His sense is that a Trump administration would keep monitoring provisions like the compute thresholds while repealing and replacing the rest of Biden's order.

He argues the people around Trump are situationally aware, in the Aschenbrenner sense, and that the national security apparatus will inevitably wake up to AI, so it is better done openly. He wants those in charge to be anti-censorship and anti-surveillance, noting that AI could power a dragnet NSA program. He also describes a "defensive accelerationist" ethic, iron domes everywhere, and sees energy as the bottleneck, pointing to China's coal build-out and Trump's aim to double electricity production.

Labenz replies that if the US and China race for indefinite strategic dominance, "we're all kind of screwed." He points to thousands of nuclear weapons still on alert, argues that even a small annual probability of nuclear war compounds, and finds the Manhattan Project a chilling precedent because many of its scientists regretted it. Hammond answers that mutual assured destruction pushed conflict into the tail rather than ending it, and that the Biden administration has in some ways been the more escalatory toward China.

Labenz says he has never been sold on the chip ban, calls himself a China dove, and sees the bipartisan impulse to stick it to China as lacking a clear end. Hammond defends export controls as securing American standard-setting capacity, noting that the US has roughly 90 percent of cloud infrastructure and Nvidia. If AGI is near, he says, even adding a year or two to the US lead could be decisive, especially for the inference compute needed to commercialize a model.

Hammond describes his essay series *AI and Leviathan* as framing a knife-edge between a totalitarian panopticon and a path that preserves liberty. He holds up Estonia, with its X-Road data exchange layer and encrypted e-government, as the model to aim for, and says the people thinking in these terms tend to be libertarian-right figures like Balaji, which makes it a high-variance gamble. He describes the choice as between two flawed options with very different variances.

Labenz argues China will likely make enough chips itself and that cutting it off, plus growing secrecy among frontier labs, could push it toward a different and less predictable branch of the technology tree. He cites the worry from Hammond's 95 theses that RL-based threat models have been prematurely discounted. Hammond responds that the controls are robust across architectures, citing Nvidia's rising H100 output and the difficulty China has had indigenizing chips and widebody aircraft, while ASML servicing remains a lever.

Labenz says he is allergic to the "our values versus their values" framing, noting that Sam Altman's Washington Post op-ed marked a sharp turn from his earlier openness to cooperating with China. He asks what Chinese values are, and says he would propose de-escalating, staying on a shared technology branch and pressing on human rights at the same time. Hammond replies that a Trump-style approach lets the US preserve its own values without lecturing, and notes that China has law requiring AI to reflect socialist values.

Labenz notes that US companies are the ones talking about AGI within a couple of years, so if anyone destroys the world with AI it is likelier to be the US, by something like an order of magnitude. He adds that if an AI powerful enough to impose values on America appeared, the key question would be whether we survive at all. Hammond answers with the Red Queen dynamic: we must co-evolve with the technology and strengthen our institutions to avoid being subsumed.

Hammond expects, and favors, a joint venture between the major labs and the US government, noting OpenAI's work at Los Alamos and Anthropic's national-security ties. He argues the defense vertical has the flexibility, like the original Manhattan Project run like a startup, and that frontier AI work should happen within democratic structures. Labenz says he leans toward advocating a halt to pushing the frontier until we understand what we are dealing with, and Hammond frames his own proposal as scaling interpretability, alignment and secured gain-of-function-style research, which would also make a coordinated pause easier.

Labenz says he would feel better about a state-backed effort to peer into the future with a stable, competent leader who has the courage to blow a whistle, and says Trump would be near the bottom of his rankings; he jokes that he would sooner bet on Xi first. Hammond responds that the choice is a package deal and that Democrats see themselves as managing a stable social and technical equilibrium, which he thinks is not the world we are entering.

Asked whom he would draft, Hammond says the focus on individuals is too narrow, because voters are choosing blocks of power. He thinks it would have to be an outsider, a Dominic Cummings type, and mentions Andrew Yang as ahead of his time. The two agree Trump has gone further than recent presidents in threatening nuclear war, but they read it differently: Hammond calls him close to a pacifist whose unpredictability is a credible threat, while Labenz is not so sure.

Hammond adds that policy on AI will depend on who is delegated authority in a crisis, scientists and builders or ethicists and union representatives. Labenz concedes he buys much of the diagnosis: a technology revolution will not be handled by consulting union leaders, and reform may need someone with an independent power base. Even so, he cannot bring himself to bet on Trump, and prefers keeping the incrementalists for another term and hoping for a better reformer later.