America’s AI election

This year marks the first AI election. Voters are asking chatbots how they should vote. Campaigns and activists are using AI to analyse the electorate, send micro-targeted messages, produce tailored ads and even deepfakes. Candidates are tinkering with what chatbots say about them. And in many parts of the world, artificially intelligent machines, or at least data centres, are on the ballot.

That this should be the case is a consequence of the failure of political leaders to defend the quality-of-life interests of their constituents against the business interests of multinational tech companies. In a poll of five European countries, three-quarters of respondents said they favoured data centres only if they were fuelled by newly developed sources of renewable energy. About the same percentage of Americans oppose the building of data centres anywhere near where they live.

Last year, Donald Trump issued an executive order declaring the regulation of AI a form of wokeism, and, this year, the EU’s attempt to regulate AI has faltered. In 2024, the EU passed an Artificial Intelligence Act, which attempts to curb “unacceptable risks” of AI by prohibiting, for instance, the use of “subliminal, manipulative, or deceptive techniques to distort behaviour and impair informed decision-making, causing significant harm”. But in May, the EU appeared to bow to pressure from US-based AI corporations to amend the law, delaying implementation of restrictions on “high-risk” uses to 2027 or 2028 — and introducing a grace period for watermarking of AI-generated content.

In the absence of comprehensive government regulation of AI in the form of executive action, legislation or judicial decisions, voters are undertaking to regulate the industry themselves — at the polls.

Voters have no way of knowing what any given bot is trying to do, or for whom, or to what end

Voters have no way of knowing what any given bot is trying to do, or for whom, or to what end

This turn has already begun. Opposition to data centre construction has influenced municipal elections in France and national politics in Ireland. At the moment, the momentum appears to be all on the side of the antis. In US state primaries this spring and summer, more than one campaign was felled by an intensifying tech backlash.

In Utah, the state senate’s president, a popular and powerful Republican, lost his seat to a challenger who lambasted him for having championed the building of a 40,000-acre data centre alongside the Great Salt Lake and who described voters as “tired of feeling like they’re not seeing themselves represented in government”. (More than 70 per cent of voters opposed the data centre and American public opinion broadly disfavours their construction.)

In Michigan, voters recalled town board members who refused to extend a data-centre moratorium. In Maryland, three Republican county commissioners who had thwarted calls for a public referendum on data centres were defeated by Republican challengers who, as one of them put it, ran on the message, “we will stop data centers!” In Wisconsin, a democratic socialist who ran on a promise to halt the construction of data centres nearly won the Democratic gubernatorial primary.

Even in races where the most vehemently anti-AI candidate lost, AI didn’t win. In a primary race for a congressional seat from Manhattan, tech billionaires spent $8mn to defeat a Democrat, Alex Bores, who, as a state assembly member, had sponsored New York’s AI safety law (one of the few such laws in the US with any teeth). But the candidate who won that primary had co-sponsored the same law and he, too, favours a data centre moratorium.

There are darker forces at work than American and other tech companies spending billions of dollars to influence governments and elections at home and abroad. The very AI tools about which voters have so many concerns are making it harder for them to know how to vote because it’s increasingly difficult to know what’s true and what’s not.

In the US, pro-data centre legislators and advocates have asserted that the anti-data centre movement is an AI deepfake: the Trump administration has promoted the theory that American opposition to AI is a Communist plot, the product of an AI-driven Chinese influence campaign. The anti-data centre people call this “data center gaslighting”. This shouldn’t be particularly confusing but it can be hard to know who’s faking when everything feels fake.

In this sense, the data centre debate is a good illustration of how, day by day, liberal constitutional democracies are being subsumed by what I have called the “artificial state”. Within the artificial state, what was once the public sphere has been replaced by automated online arenas owned by private corporations, driven by AI and dominated not by humans but by robots. Currently, there are more bots online than humans and most new English-language writing on the internet is written not by people but by machines.

Bots — information drones — act on behalf of virtually any kind of entity, from nation-states to stateless regimes, from political parties to trade unions, from billionaires to corporations, from lobbyists and political candidates to terrorists and criminals. Voters have no way of knowing what any given bot is trying to do, or for whom, or to what end. This much, however, is true: democracies on every continent have allowed drones to take the place of the demos.

Rules designed to make something safer sometimes do make it safer, better, cheaper and faster

Rules designed to make something safer sometimes do make it safer, better, cheaper and faster

Under these circumstances, AI is sure to influence the US midterm elections in November. Indeed, the next US presidential election, in 2028, may well constitute a referendum on the regulation of the AI industry. Past, present and future presidential contenders have all begun to weigh in on the question. Trump and his troop of possible successors — including JD Vance and Marco Rubio — oppose meaningful regulation (Trump favours controlling the industry by the force of his own will), although this position is losing support among the Republican Party more broadly, as the techlash crosses political boundaries.

Vermont senator Bernie Sanders has called for a national moratorium on the building of data centres in the US and he has been joined by other leading and rising progressives as well, including New York Democrat Alexandria Ocasio-Cortez.

Democrat Ro Khanna, a 49-year-old member of Congress who represents Silicon Valley, is toying with a bid for the presidency (he is likely to announce his decision after the midterms). This spring, in a manifesto called “AI for the People”, Khanna, a progressive who supports a wealth tax and is best known for leading the congressional campaign for the release of the Epstein files, called for a “new tech social contract”. Echoing John F Kennedy, he asserts that Americans ought to ask “not what America can do for Silicon Valley, but what Silicon Valley must do for America”. One of his manifesto’s principles is to “prevent AI from weaponizing public discourse”. Another is to “regulate AI so it is used to improve humanity, not damage it”. What any of that means is less clear.

Tech executives have tried to persuade governments to forswear regulation and to forge ahead with building data centres, despite voter objections, citing not their own profits but national security concerns — particularly the worry that China will win the “AI race”.

Those national security concerns are real. In a forthcoming book titled The Bitter Struggle, Ben Buchanan and Tantum Collins, two former Biden administration officials, argue that democracies must win the AI race because “Autocracies will use AI to entrench their power, repress their populations, and extend their reach abroad. Only democratic primacy in AI offers a path to broadly shared benefits, preserved freedoms, and governance that reflects the will of the people.” The hitch is that the will of the people appears to be not to develop AI.

The other argument put forward by tech executives — the argument that led the EU to push back the full rollout of the Artificial Intelligence Act and the Trump administration to undo the AI safety measures put in place during the Biden administration — is that “regulation stifles innovation”. This claim is made as if this were an eternal truth when it is, in fact, equal parts political slogan and marketing campaign.

Rules designed to make something safer can sometimes slow down the kind of tinkering that might make it better, or cheaper, or faster. But rules designed to make something safer sometimes also make it safer, better, cheaper and faster.

One of the best accounts of the relationship between the law, technology and regulation was, unfortunately, long lost to history. In 1949, the distinguished University of Wisconsin law professor James Willard Hurst, a hugely influential pioneer of modern legal history, completed a chapter on the automobile for a book he planned to write on the law, technology and regulation. Hurst never finished his book but his draft chapter on the automobile was saved from a dumpster in 1990 by a doctoral student who finally published it four years ago. Maybe it’s not too late for it to inform the ongoing debate about artificial intelligence.

Congress regulated radio from early on, and aeroplanes from the very beginning. Yet no one argued that the aeroplane and the radio developed too slowly

Congress regulated radio from early on, and aeroplanes from the very beginning. Yet no one argued that the aeroplane and the radio developed too slowly

Hurst’s 74-page chapter, “Technology and the Law: The Automobile”, begins with a 119-point list of the “Derivative Effects of the Auto Upon the Law”. He was fascinated by all that the automobile had wrought between Ford’s introduction of the Model T in 1909 and the time of his writing in 1949. For instance, Hurst believed that by giving people a sense of control and freedom and escape, the automobile appeared to “somewhat offset the loss of status and independence that had come with a machine-based society”.

That gain came with a deepened sense of privacy, the car as a man’s castle. His “derivative effects” were sometimes speculative but mostly persuasive (“The auto was an important factor in the higher standard of living desired by twentieth-century Americans, which tended to lower the birth rate and contribute to the relative ageing of the population, with attendant shifts in demands upon government for economic security; of the same tendency was the greater freedom from family ties to which the auto contributed, and which reduced the willingness and ability of children to take care of aged parents.”)

But Hurst was on very solid ground in listing the many ways the automobile had affected the law. It led to new types of taxes and tax deductions, a whole new understanding of liability and tort, new types of arbitration for disputes over accidents, driver’s licences, speed limits, casualty insurance, traffic laws and wartime rationing, the “regulation of the quality and safety of motor fuel”, the “licensing of dealers in new and used cars and regulation of their financing practices”, the regulation of driving schools, railroad crossings, “the registration of auto titles and legal doctrine governing transfers of title” — and much more, including “new interest in resort to law to preserve the natural beauty along the roads from invasion by roadside advertising and business”. (All this was years before Ralph Nader published Unsafe at Any Speed, the 1965 book that led to the introduction of, for instance, mandatory seatbelts, and the US National Traffic and Motor Vehicle Safety Act of 1966.)

For all the successes of regulation, Hurst argued that the law of the automobile had been laggard (with fatal consequences), preoccupied “with the problem of compensation for damage done” and uninterested in the “prevention of damage”. This, he argued, had been a feature of the Machine Age, too, beginning in the 1870s, with factories. (Manufacturers settled with workers harmed in factory fires long before anyone passed a law mandating fire escapes.) But the Machine Age ended about 1910, which meant that the laissez-faire treatment of the automobile was an exception, a throwback.

Recommended

Beginning in the 1910s, Progressive Era good-government reformers had abandoned the regulation-stifles-innovation argument of Machine Age boosters, who were known as Men of Progress. And so, for instance, Congress subjected the radio to strict regulation and the prevention of harm from early on and the aeroplane from the very beginning. Those measures included the Air Commerce Act of 1926 and the Civil Aeronautics Act of 1938, alongside the Federal Radio Act of 1927 and the Federal Communications Act of 1934. And no one argued that the aeroplane and the radio developed too slowly.

People like cars, and they don’t like data centres. But that doesn’t relieve legislators of the obligation to act. This election year, voters may force them to.

Jill Lepore is professor of American history at Harvard University and professor of law at Harvard Law School. Her new book, ‘The Rise and Fall of the Artificial State’, is published on August 25 by Allen Lane in the UK and Liveright in the US

Find out about our latest stories first — follow FT Weekend on Instagram, Bluesky and X, and sign up to receive the FT Weekend newsletter every Saturday morning