The Strange Case of the AI Force

There is something admirably straightforward about President Trump’s attitude toward artificial intelligence. America has stumbled onto an extraordinary new technology; American companies are, for the moment, rather good at making it; China would quite like to be better at making it; so perhaps the sensible thing is not to strangle the whole business in its cradle. This is not a terribly complicated argument, which is part of its appeal.

Trump has now given the argument an institution. On Saturday he announced an “AI Force,” along with a forthcoming “AI czar,” promising that his administration would “cherish” and “help” the industry rather than hinder it. Exactly what the force is, where it will live inside the federal government, and what powers it will possess remain unclear. Reuters likewise reported that Trump announced the force and a new AI adviser without yet supplying implementation details. Perhaps it will turn out to be a very good idea. But there is something faintly peculiar here.

The conservative position on an exciting new industry was once supposed to be that government should mostly leave it alone. The new position seems to be that government should leave it alone so energetically that it requires a special government body to make sure everyone is leaving it alone.

This is how Washington grows. Not usually because someone announces that the federal government ought to become larger and stranger. It grows because some genuinely important thing appears in the world, and therefore somebody must be appointed to look after it. AI really is important, but that is also precisely why the temptation deserves scrutiny.

Trump’s underlying diagnosis has plenty to recommend it. The United States has a large lead in much of the AI ecosystem, the technology may have enormous economic and military importance, and an American regulatory regime designed principally around preventing hypothetical harms could easily become an extraordinary gift to Beijing. The administration is also plainly capable of distinguishing technological optimism from complete insouciance: American officials are simultaneously pursuing discussions with China about mechanisms for notifying one another of serious AI incidents and other safety risks.

So the question is not whether Trump “takes AI seriously.” Clearly he does. Nor is it whether rapid American development is desirable. There are strong reasons to think it is. The more interesting question is what happens once the federal government decides that an industry is not merely useful, but strategic.

“Strategic” is a dangerous word in Washington. It performs approximately the same function as holy water did in medieval Europe. Sprinkle it over an ordinary policy dispute and suddenly everything is permitted.

Ordinarily, a dispute over whether somebody should build a warehouse-sized computer next to your town is a dispute about land, electricity, water, taxes and perhaps whether you particularly enjoy looking at warehouse-sized computers. Once AI becomes a matter of national destiny, the same dispute starts to resemble the Battle of Midway. The zoning board is no longer deciding whether to approve an industrial facility. It is deciding whether America will defeat China. This does not mean the national-security argument is false. Quite the opposite: it is strong enough that it could swallow a great many other arguments whole.

The same problem appears with the phrase “AI czar.” Washington has accumulated czars for the same reason old royal courts accumulated chamberlains, masters of horses and keepers of the king’s swans: once an institution becomes sufficiently complicated, it starts producing offices whose precise relationship to all the other offices becomes difficult to explain.

Maybe an AI czar is exactly what is needed. Artificial intelligence touches the Pentagon, Commerce Department, Energy Department, intelligence agencies, financial regulators and half the rest of the federal alphabet. Someone coordinating these organizations could conceivably make government less cumbersome rather than more. But conservatives are entitled to ask what kind of creature is being created before becoming too attached to it.

Is the AI Force an advisory committee? A deregulation shop? An industrial-policy office? A national-security body? Will it coordinate agencies, overrule them or simply argue with them? Can it direct money? Can it pressure states? Does Congress have anything to say about it? These sound like procedural questions, which is another way of saying they are constitutional questions that have temporarily put on boring clothes.

There is a second difficulty: Trump says he wants government to “help” AI. Again, the instinct is understandable. The United States should want American firms to build the best models, manufacture advanced chips, construct abundant power and remain ahead of geopolitical competitors. But helping industries is something governments are very good at beginning and remarkably bad at finishing.

A strategic industry quickly acquires strategic companies. Strategic companies acquire strategic subsidies, strategic tax treatment, strategic exemptions and strategic relationships with the officials responsible for deciding what counts as strategic. Eventually everyone involved remains a committed supporter of free enterprise, provided the enterprise in question has retained the correct lobbyist. None of this requires corruption. That’s what makes it interesting. It is simply how the incentives work.

The biggest AI companies will inevitably understand AI policy better than nearly everyone in government. They will employ former officials; government will employ former industry executives; both sides will sincerely believe that the other possesses useful expertise. And because complicated regulation is generally easier for a trillion-dollar corporation to navigate than a twelve-person startup, even rules written with perfectly good intentions can quietly transform today’s incumbents into tomorrow’s regulated utilities.

A conservative AI policy therefore has two dangers to avoid rather than one. It can regulate the industry until nothing interesting can be built. Or it can embrace the industry so tightly that a handful of companies become semi-official national champions. The space between those mistakes is probably where the administration wants to be. The uncertainty is whether an “AI Force” helps it stay there.

There is a similar ambiguity in the growing fight over data centers. Trump has presented opposition to new facilities as misguided and emphasized the economic benefits they can bring. The political backlash, however, has centered on extremely ordinary concerns: electricity prices, grid capacity, water use and local development. Even Republican candidates have begun taking different positions on how quickly construction should proceed.

Here the older conservative vocabulary may still have something useful to say. A town should probably not be able to veto the existence of the American AI industry. But neither is every local objection to a data center evidence that someone has failed to grasp the grandeur of technological progress. People really do live next to power plants and transmission lines. They really do pay utility bills. Property rights remain property rights even when the property next door contains 100,000 GPUs. Federalism is useful partly because it prevents every disagreement from becoming a referendum on the fate of civilization.

None of this makes the administration’s concern with AI foolish. In fact, Trump may be right about the most important thing: America would be taking an enormous gamble if it responded to the arrival of powerful AI primarily by making it more difficult to build powerful AI in America. The competition with China is not imaginary, and technological abundance is generally a better American tradition than technological prohibition. But precisely because AI matters so much, conservatives might want to resist the idea that its importance settles every subsidiary political question.

Perhaps we need an AI Force. Perhaps we need a czar. Perhaps certain state regulations really do threaten a national market and ought to be preempted. Perhaps certain infrastructure projects really are important enough to justify federal involvement. The word perhaps is doing useful work in those sentences.

The attractive thing about Trump’s approach is its impatience with the emerging tendency to treat every new technology as a problem to be managed before it has even had the chance to become useful. The risk is that the reaction to bureaucratic management becomes a different sort of management: an alliance of federal power and favored industry, organized around the permanent emergency of beating China.

Conservatives have historically been suspicious of government not only when government says “No,” but when it says “Let us help.” The AI Force may ultimately amount to little more than a coordinating office with a grand name. It could even prove useful. For now, though, the most conservative response to the announcement may simply be curiosity.

What exactly is this thing? And once Washington has created a force to cherish artificial intelligence, who, exactly, will be watching over the force?