I’m writing this on a plane back to Washington, DC, from a conference in the Bay Area, the land of tomorrow. While the conference wasn’t about AI, this is the Bay Area, and thus roughly 90 percent of conversations were about AI.
It is hard to overstate the scale of the gap between the cultures of the Bay Area and DC on this topic. AI has certainly become a real part of the policy conversation in DC, but only in quite technical, near-term, and not especially high-profile ways: How should we regulate deep fakes? How should we handle data centers’ increasing demands for energy? Should we require Nvidia processors to have a little component that can tell if the chip is physically in China to prevent Beijing from getting its hands on too many?
But if DC’s AI concerns are quotidian, the Bay Area’s are existential.
In Berkeley, or at least among the crowd I was talking to, the questions were more like: Are we ever going to be able to stop these machines from cheating on our attempts to evaluate us, from blackmailing us when we obstruct their goal, from actively working to avoid being shut down? (These are all real things that researchers have found leading-edge AI models can do.) If we don’t fix these problems, will we survive the next 10 years?
Sign up here to explore the big, complicated problems the world faces and the most efficient ways to solve them. Sent twice a week.
When somewhat less apocalyptic questions like “how will we cope if billions of people are suddenly unemployed due to AI and robotics progress,” the tone of most responses I got was something like, “God, I really hope that turns out to be the biggest problem. It means we all survived.”
Temperamentally, I’m more inclined to think about these things in very concrete, near-term ways. There’s a reason I live in Washington, DC; it is a town for good-natured incrementalists. So, naturally, all the AI talk got me thinking about the huge budget reconciliation bill passed by the House and being considered by the Senate.
Let me be blunt: This is, in ways big and small, not a budget that takes AI seriously at all. Even worse, if you think this technology is going to have an even slightly significant influence on the world in the next decade, the One Big Beautiful Bill will make that influence worse.
The directly AI-related stuff
There’s one section of the bill that’s directly about AI, which is the proposed moratorium on most state-level attempts to regulate AI for the next 10 years. Originally, this was an outright ban, but because of the limits on what reconciliation bills can do on non-budgetary matters — and attempting to regulate regulation is clearly non-budgetary — it now takes the form of a requirement that states abstain from regulation if they want to get broadband money.
There are reasonable arguments that AI policy should happen at a federal, rather than state, level. But this isn’t a case where the federal government has a well-reasoned policy framework that it seeks to impose instead of the states’ policies. This is a case where the federal government wants to remove or prevent state regulations and replace them with nothing at all.
It’s not surprising that corporate interests like the venture capital fund Andreesen Horowitz are ramping up their DC lobbying effort amid this fight. AI will change our lives quite fast. The public is already very suspicious of it and will want regulation, demands that are only set to grow as the near-term economic and labor effects of AI become palpable. The only way for the industry to prevent this is to lock in a laissez-faire regime right now. If you think there’s even a chance that these systems could cause serious problems worthy of regulation, this is a very dangerous provision. Thankfully, even quite conservative Republicans in both houses seem to be realizing this, and hopefully that backlash kills the provision.
Almost as relevant to the industry are provisions slashing subsidies for clean energy development. Training and deploying AI requires a lot of data centers full of very expensive chips that need to be running 24/7 to pay back their immense upfront cost. Those centers need equally reliable, 24/7 sources of power. Ideally, that comes from clean sources like nuclear, geothermal, or solar-plus-batteries. Slightly worse would be natural gas. Much worse would be coal.
The reconciliation bill takes a number of actions to lower the odds that data centers are fueled by clean sources. It of course slashes the generous subsidies the Inflation Reduction Act created to encourage clean energy, which can offset as much as 30 percent of the cost of a new power plant.
The nuclear industry, the clean source to which Republicans are usually friendliest, has warned that the cuts could seriously hurt them as well. The bill also takes a hatchet to the Loan Programs Office, an Energy Department tool for investing in clean energy that’s especially important for nuclear and geothermal. Energy Secretary Chris Wright went so far as to ask Republicans to dial back the cuts to nuclear and geothermal; I don’t think a cabinet member has asked for smaller cuts in any other section of the bill, but this was concerning enough to spark intervention.
As policy analysts Thomas Hochman and Pavan Venkatakrishnan noted in the Washington Post, Congress’s “approach almost uniquely disadvantages newer competing energy sources that run 24/7,” hurting them even more than wind and solar. It’s almost like it’s designed to make new data centers run on dirty fuels, or perhaps to encourage companies to build them abroad.
Work requirements in a post-work world
But the big, big problem with the bill is its obsession with larding on more onerous, poorly administered, ineffective work requirements on programs like Medicaid and food stamps.
I thought these were bad policies before AI became a big deal, and I am happy to rant at length about why. They’re cruel, they do not lead people to work more, and for Medicaid in particular, even conservatives who normally like work requirements accept they’re totally ineffective.
But back up for just one second. Right now, the leaders of the world’s AI companies are declaring that within the decade, they will be able to fully automate a huge share of human labor. Maybe you think they’re out of their gourds and nothing remotely like that will happen. It’s possible. It’s also possible that these incredibly powerful people with many billions of dollars at their disposal will be able to succeed at what they set out to do.
It’s also possible that even much, much less powerful AIs, like those available today, will eventually cause meaningful employment loss. We’re seeing some indications that’s already happening. In even the absolute slowest plausible timeline for AI that I can imagine, you still will have companies like Waymo using it to displace human labor in specific industries.
In a world where Uber and truck drivers are suddenly out of work due to no fault of their own, adding work requirements to food stamps and Medicaid is cruel. It won’t cause them to find work, at least in the near term; the work in their vocation is gone. Perhaps they should change occupations — but are we really confident their new job won’t be automated the same way? Do they not need some help as they transition?
Vice President JD Vance gave a speech in March where he reminisced about the steel plant in his Ohio hometown, saying, “it was the lifeblood of the town that I grew up in. When it went from 10,000 jobs to 2,000 jobs, the American working people started to get destroyed in the process. We can’t keep doing that.”
But his party’s budget bill does exactly that. It sees people whose livelihood might be destroyed imminently and actively takes support away from them. “We can’t keep doing that”? You’re doing that right now.
In a world of truly transformative AI, automating 10 or 20 or perhaps even 100 percent of human labor, work requirements go from cruel to some combination of cruel, bizarre, and silly. They’d be like if Congress were, today, to pass a dedicated law setting labor standards for horse-and-buggy drivers. Imagine telling folks in a world of transformative AI “you have to work to get food stamps.” Work? What work? Unemployment is 30 percent and rising, what are you even talking about?
David Sacks, a venture capitalist and one of Trump’s closest advisers on AI, has generally been dismissive about the potential of AI to threaten jobs. But even he conceded on a recent episode of his All In podcast, “If there is widespread job disruption, then obviously the government’s going to have to react and we’re going to be in a very different societal order.”
At the same time, on X, he’s declaring, “The future of AI has become a Rorschach test where everyone sees what they want. The Left envisions a post-economic order in which people stop working and instead receive government benefits. In other words, everyone on welfare. This is their fantasy; it’s not going to happen.”
Fine, you don’t want that. But AI will certainly displace many jobs if not eliminate them, and Sacks himself admits you need massive government intervention in that case. I don’t have a clear idea what that intervention would ideally look like; we know so little about how this technology is going to diffuse through society, how fast it will improve, and what this means for jobs. It’s an area that needs far more attention, from AI companies, governments, and civil society.
But I feel confident on one point. AI is going to make some employment more precarious. Occupations will be threatened. People will lose their jobs. The questions are how many of them will, and whether and how quickly they’ll get new ones.
Given all that, adding new work requirements to safety net programs isn’t just cruel or unwise. It’s a sign that this administration, and its tech advisers like Sacks, do not take the future of AI seriously at all.
