If the presidential election has provided relief from anything, it has been the generative-AI boom. Neither Kamala Harris nor Donald Trump has made much of the technology in their public messaging, and they have not articulated particularly detailed AI platforms. Bots do not seem to rank among the economy, immigration, abortion rights, and other issues that can make or break campaigns.

But don’t be fooled. Americans are very invested, and very worried, about the future of artificial intelligence. Polling consistently shows that a majority of adults from both major parties support government regulation of AI, and that demand for regulation might even be growing. Efforts to curb AI-enabled disinformation, fraud, and privacy violations, as well as to support private-sector innovation, are under way at the state and federal levels. Widespread AI policy is coming, and the next president may well steer its direction for years to come.

On the surface, the two candidates couldn’t be further apart on AI. When AI has come up on the campaign trail, the focus has not been on substantive issues, but instead on the technology’s place in a supposed culture war. At a rally last winter, Trump railed against the Biden administration’s purported “use of AI to censor the speech of American citizens” (a contorted reference, perhaps, to an interview that week in which Secretary of Homeland Security Alejandro Mayorkas denounced the “politicization” of public education around the dangers of AI, including misinformation). Trump also said he would overturn Joe Biden’s executive order on AI—a sprawling document aiming to preserve consumer and civil rights while also spurring innovation—“on day one.” Then, over the summer, the GOP platform lambasted the “dangerous” executive order as slowing innovation and imposing “Radical Leftwing ideas” on the technology, perhaps referring to the order’s stated “dedication to advancing equity.” Elon Musk, now the most powerful Trump surrogate in the world, recently invited his followers to “imagine an all-powerful woke AI.” Harris, for her part, hasn’t discussed AI much as a candidate, but she is leading many of Biden’s AI efforts as vice president, and her economic platform mentions furthering “the commitments set forth in the 2023 AI Executive Order.”

[Read: The real AI threat starts when the polls close]

Such rhetoric is par for the course this election cycle: Trump in particular has never been known for nuance or gravity, and tearing down Biden is obviously his default position. What no one seems to remember, though, is that Biden’s “dangerous” executive order echoes not one but two executive orders on AI that Trump himself signed. Many of the policies around AI that President Biden and Vice President Harris have supported extend principles and initiatives from Trump’s term—such as efforts to establish federal funding for AI research, prepare American workers for a changing economy, and set safety standards for the technology. The two most recent presidential administrations even agreed on ensuring that federal AI use is nondiscriminatory. Trump’s approach to the technology, in turn, built on foundations laid during Barack Obama’s presidency.

In other words, despite how AI has been approached by their campaigns (that is, barely, or only in the shallowest terms), both candidates have real track records on AI, and those records are largely aligned. The technology appeared to be a rare issue driven for years by substance rather than partisanship, perhaps because prior to the launch of ChatGPT, it wasn’t on many Americans’ minds. With AI now assuming national importance, Trump has promised to tear that consensus down.

Still, there’s a good chance he won’t be able to—that reason and precedent will prevail in the end, if only because there’s already so much momentum behind what began during his own administration. “To the extent that the Trump administration worked on issues of science and technology policy, it worked on AI,” Alondra Nelson, a professor at the Institute for Advanced Study who previously served as the acting director of Biden’s Office of Science and Technology Policy, told me. And in doing so, it was inheriting priorities set under a man Trump has called “the most ignorant president in our history.” Near the end of his second term, Obama directed several federal agencies to study and plan for the growing importance of “big data” and AI, which culminated at the end of 2016 with the publication of a report on the “future of artificial intelligence,” as well as a national strategic plan for AI research and development. Those included broad suggestions to grow the federal government’s AI expertise, support private-sector innovation, establish standards for the technology’s safety and reliability, lead international conversations on AI, and prepare the American workforce for potential automation.

A few years later, Trump began to deliver on those recommendations through his executive orders on AI, a 2019 update to that strategic plan, and his White House’s guidance to federal agencies on using AI. “The Trump administration made AI a national technology priority,” Michael Kratsios, who served as the country’s chief technology officer under Trump and helped design his AI strategy, told Congress last October. In that testimony, Kratsios, who is currently the managing director of the start-up Scale AI, lauded much of Obama’s previous and Biden’s current work on AI—even criticizing Biden for not doing enough to implement existing policies—and noted the continued importance of supporting “high-quality testing and evaluation” of AI products.

Biden and Harris have since taken the baton. Trump’s first executive order in particular did “have a lot of the ingredients that got much more developed in Biden’s EO,” Ellen Goodman, a professor at Rutgers Law School who has advised the National Telecommunications and Information Administration on the fair and responsible use of algorithms, told me. “So when Trump says he’s going to repeal it with a day-one action, one wonders, what is it exactly that’s so offensive?” Even specific policies and programs at the center of Biden and Harris’s work on AI, such as establishing national AI-research institutes and the National AI Initiative Office, were set in motion by the Trump administration. The National Artificial Intelligence Research Resource, which Harris’s economic plan touts by name, originated with AI legislation that passed near the end of Trump’s term. Innovation, supporting American workers, and beating China are goals Harris and Trump share. Bluster aside, the candidates’ records suggest “a lot of similarities when you get down to the brass tacks of priorities,” Alexandra Givens, the president of the Center for Democracy & Technology, a nonprofit that advocates for digital privacy and civil rights, told me.

[Read: The EV culture wars aren’t what they seem]

To be clear, substantive disputes on AI between Harris and Trump will exist, as with any pair of Democratic and Republican presidential candidates on most issues. Even with broad agreements on priorities and government programs, implementation will vary. Kratsios had emphasized a “light touch” approach to regulation. Some big names in Silicon Valley have come out against the Biden administration’s AI regulations, arguing that they put undue burdens on tech start-ups. Much of the Republican Party’s broader message involves dismantling the federal government’s regulatory authority, Goodman said, which would affect its ability to regulate AI in any domain.

And there is the “Radical Leftwing” rhetoric. The Biden-Harris administration made sure the “first piece of work out the public would see would be the Blueprint for an AI Bill of Rights,” Nelson said, which outlines various privacy and civil-rights protections that anyone building or deploying AI systems should prioritize. Republicans seem to have a particular resistance to these interventions, which are oriented around such concepts as “algorithmic discrimination,” or the idea that AI can perpetuate and worsen inequities from race, gender, or other identifying characteristics.

But even here, the groundwork was actually laid by Trump. His first executive order emphasized “safety, security, privacy, and confidentiality protections,” and his second “protects privacy, civil rights, [and] civil liberties.” During his presidency, the National Institutes of Standards and Technology issued a federal plan for developing AI standards that mentioned “minimizing bias” and ensuring “non-discriminatory” AI—the very reasons why the GOP platform lashed out against Biden’s executive order and why Senator Ted Cruz recently called its proposed safety standards “woke.” The reason that Trump and his opponents have in the past agreed on these issues, despite recent rhetoric suggesting otherwise, is that these initiatives are simply about making sure the technology actually functions consistently, with equal outcomes for users. “The ‘woke’ conversation can be misleading,” Givens said, “because really, what we’re talking about is AI systems that work and have reliable outputs … Of course these systems should actually work in a predictable way and treat users fairly, and that should be a nonpartisan, commonsense approach.”

In other words, the question is ultimately whether Trump will do a heel turn simply because the political winds have shifted. (The former president has been inconsistent even on major issues such as abortion and gun control in the past, so anything is possible.) The vitriol from Trump and other Republicans suggests they may simply oppose “anything that the Biden administration has put together” on AI, says Suresh Venkatasubramanian, a computer scientist at Brown University who previously advised the Biden White House on science and technology policy and co-authored the Blueprint for an AI Bill of Rights. Which, of course, means opposing much of what Trump’s own administration put together on AI.

But he may find more resistance than he expects. AI has become a household topic and common concern in the less than two years since ChatGPT was released. Perhaps the parties could tacitly agree on broad principles in the past because the technology was less advanced and didn’t matter much to the electorate. Now everybody is watching.

Americans broadly support Biden’s executive order. There is bipartisan momentum behind laws to regulate deepfake disinformation, combat nonconsensual AI sexual imagery, promote innovation that adheres to federal safety standards, protect consumer privacy, prevent the use of AI for fraud, and more. A number of the initiatives in Biden’s executive order have already been implemented. An AI bill of rights similar to the Biden-Harris blueprint passed Oklahoma’s House of Representatives, which has a Republican supermajority, earlier this year (the legislative session ended before the bill could make it out of committee in the senate). There is broad “industry support and civil-society support” for federal safety standards and research funding, Givens said. And every major AI company has entered voluntary agreements with and advised the government on AI regulation. “There’s going to be a different expectation of accountability from any administration around these issues and powerful tools,” Nelson said.

When Obama, Trump, and Biden were elected, few people could have predicted anything like the release of ChatGPT. The technology’s trajectory could shift even before the inauguration, and almost certainly will before 2028. The nation’s political divides might just be too old, and too calcified, to keep pace—which, for once, might be to the benefit of the American people.

For years, the dominant belief in both parties has been that Democrats need to run up a big lead in the national presidential popular vote to win an Electoral College majority. But in the dead-heat election of 2024, that may no longer be true. The distinctive dynamics of the 2024 campaign could allow Kamala Harris to eke out an Electoral College win even if Donald Trump runs better in the national popular vote this time than during his previous two campaigns.

The belief that Democrats need a big popular-vote win to prevail in the electoral vote hardened in the course of those two previous Trump campaigns. In 2020, Joe Biden beat Trump by a resounding 4.5 percentage points in the popular vote but still only squeezed past him by relatively small margins in the three Rust Belt battlegrounds of Michigan, Pennsylvania, and Wisconsin that decided the race. In 2016, Hillary Clinton beat Trump by two points in the national popular vote but narrowly lost those same three states, and with them the presidency.

That history has weighed heavily on Democrats as a procession of recent polls has shown Trump shrinking or even erasing Harris’s national lead. But the pattern of differences among white, Black, and Latino voters found in most of those national surveys show how Harris could still potentially capture the 270 Electoral College votes needed for victory—even if she wins the nationwide popular vote by much less than Biden did in 2020, and possibly by only about the same margin that Clinton got in 2016.

The principal reason is that these recent polls show Trump making most of his gains in national support by performing better among Black and, especially, Latino voters than he did in either of those previous elections. Even the most favorable surveys for Trump consistently find Harris polling very close to Biden’s level of support in 2020 among white voters, which had improved over Clinton’s performance with that group by several points. In other words, Harris will likely rely a bit more on white voters than her party’s past two nominees did.

That subtle shift is the crucial distinction from the earlier contests. It could allow Harris to scrape a win by sweeping the predominantly white, former “Blue Wall” battlegrounds of Michigan, Pennsylvania, and Wisconsin, even if Trump improves over his prior popular-vote results by gaining among Black and Latino voters (and Black and Latino men in particular).

[Read: Elon Musk wants you to think this election’s being stolen]

In each of his previous two races, Trump benefited because the decisive states leaned more Republican than the nation overall. In both 2016 and 2020, Wisconsin was the tipping-point state that provided the 270th Electoral College vote for the winner—first for Trump, then for Biden. In 2016, Trump ran about three percentage points better in Wisconsin than he did nationally; in 2020, he ran nearly four points better in Wisconsin than he did nationally, according to the University of Virginia Center for Politics.

The fact that Trump each time performed much better in the tipping-point state than he did in the national popular vote is central to the assumption that Democrats can’t win the Electoral College without a popular-vote majority. But as the Center for Politics research demonstrates, that hasn’t always been true.

The tipping-point states in the three presidential elections preceding 2016—Ohio in 2004 and Colorado in 2008 and 2012—each voted slightly more Democratic than the national popular vote. And in none of those elections was the disjunction between the tipping-point-state result and the national popular vote nearly as big as it was in 2016 or 2020. In fact, the gap between the national popular vote and the tipping-point state in Trump’s two races was considerably wider than in any election since 1948, the Center found.

Polling in the past few weeks, however, has indicated that this gap has shrunk to virtually nothing. Trump and Harris remain locked in a virtual tie both nationally and in the swing states. With polls that closely matched, none of the swing states appears entirely out of reach for either candidate.

Still, professionals on both sides with whom I’ve spoken in recent days see a clear hierarchy to the states. Both camps give Harris her best chance for overall victory by winning in Michigan, Pennsylvania, and Wisconsin; Trump is considered stronger across the Sun Belt in North Carolina, Arizona, Georgia, and Nevada (ranked from most to least promising for him).

That separation reflects the race’s unexpected racial dynamics. If Trump’s polling gains among voters of color bear out in practice, that would benefit him the most in the Sun Belt battlegrounds. There, minority voters are such a large share of the electorate that even a small shift in their preferences—toward Trump—would greatly diminish Democrats’ chances.

Whatever happens in the Sun Belt, though, if Harris sweeps the Rust Belt big three, she would reach exactly the 270 Electoral College votes needed to win (so long as she held all of the other states that Biden carried by about three percentage points or more, which is very likely). All three of those major industrial states are much less diverse than the nation as a whole: In 2020, white people cast about four-fifths of the vote in Michigan and Pennsylvania, and roughly nine-tenths of it in Wisconsin, according to census figures.

“One of the potential outcomes here is that at the end of the day, Trump will have gained with Blacks and Latinos and it may not have decided the Electoral College, if we don’t need [the Sun Belt states] to win,” Paul Maslin, a Democratic pollster with long experience in Wisconsin, told me.

Obviously, Harris has no guarantee that she could survive a smaller national popular-vote margin than Biden: The polls showing national gains for Trump could be capturing a uniform uptick in his support that would deliver slim victories across most—and possibly all—of the seven decisive states. Even the most optimistic Democrats see marginal wins in the battlegrounds as probably Harris’s best-case scenario. But the prospect that she could hold the former Blue Wall states even while slipping nationally challenges the conventional wisdom that Democrats must amass a significant lead in the national popular vote to secure enough states to win the electoral vote.

“The Blue Wall states are the likeliest tipping point for either candidate,” Kyle Kondik, the managing editor of the Sabato’s Crystal Ball newsletter published by the Center on Politics, told me. “If the country moves two to three points to the right but those states only move a point or less, that’s where you start to get the tipping point looking pretty close to the popular vote.”

The Democratic strategist Mike Podhorzer, a former political director at the AFL-CIO, also believes that Harris could win the Electoral College with a smaller popular-vote advantage than most analysts have previously assumed. But he says the demographic characteristics of the swing states aren’t the primary cause of this possibility. Rather, the key factor is that those states are experiencing the campaign in an immersive way that other states are not thanks to huge advertising spends, organizing efforts, and candidate appearances.

That disparity, he says, increases the odds that the battleground states can move in a different direction from the many states less exposed to such campaigning. Both Podhorzer and Kondik note that the 2022 midterm elections supported the general thesis: Although broad dissatisfaction with Biden allowed Republicans to win the national popular vote in House elections, Democrats ran much better in statewide contests across the most heavily contested battlegrounds, especially in Michigan, Pennsylvania, Wisconsin, and Arizona.

“It is really the difference between how well you are doing outside the battlegrounds and inside the battlegrounds,” Podhorzer told me. Inside the battlegrounds, he pointed out, voters have for years now been exposed at blast-force volume to each party’s arguments on all the major issues. “The cumulative effect of it is that they have an awareness of what is at stake, a different worldview, than people living outside those states,” he said.

The analogue to 2022 this year would be whether general disappointment in Biden’s economic record increases Trump’s popular-vote total in less-contested blue and red states alike, but Harris holds on to enough of the battlegrounds where voters are hearing the full dimensions of each side’s case against the other.

[Read: How the Trump resistance gave up]

The same national polls that show Trump gaining among voters of color this year do not show much, if any, improvement for him compared with his 2020 performance among white voters. The latest aggregation of high-quality national public polls published by Adam Carlson, a former Democratic pollster, found that Harris is almost entirely preserving Biden’s gains among white voters; that means Harris is also exceeding Clinton’s showing with them from 2016.

The comparison with Clinton is instructive. Among voters of color, Clinton ran better in 2016 than either Biden in 2020 or how Harris is polling now. But Clinton lagged about three to four points below both of them among white voters. If Harris wins the popular vote by only about the same margin as Clinton, but more of Harris’s lead relies on support from white voters, the vice president’s coalition would be better suited to win the Rust Belt battlegrounds. In that scenario, Harris would assemble what political scientists call a more electorally “efficient” coalition than Clinton’s.

Biden’s margins of victory in the former Blue Wall states were so slim that Harris can’t afford much erosion with voters of color even there. But two factors may mitigate that danger for her. One is that in the Rust Belt states, most voters of color are not Latino but Black, and Democrats feel more confident that they can minimize losses among the latter than among the former.

The other key factor is a subtle change in those states’ white populations. Calculations from the latest census data provided to me by William Frey, a demographer at the nonpartisan Brookings Metro think tank, found that since 2020, white voters without a college degree—the demographic group in which Trump performs best—have declined as a share of eligible voters by about three percentage points in both Michigan and Wisconsin, and by about 1.5 points in Pennsylvania. In Michigan and Wisconsin, college-educated white voters, who now tilt mostly toward Harris, largely made up the difference; in Pennsylvania, the share of minority voters grew. In a typical election, these slight shifts in the electorate’s composition probably would not matter, but they could in a contest as close as this one.

“There is still room to grow in the suburbs [across the region], and two things are going to contribute to that growth: January 6 and the Dobbs decision,” Mike Mikus, a Pittsburgh-based Democratic consultant, told me, referring to the insurrection at the Capitol in 2021 and the 2022 Supreme Court ruling that overturned the constitutional right to abortion. The racist slurs against Puerto Rico at Trump’s Madison Square Garden rally last weekend could also cost him with Pennsylvania’s substantial Puerto Rican population.

Sweeping Michigan, Pennsylvania, and Wisconsin with a smaller national-popular-vote lead than Biden’s is nonetheless a high-wire assignment for Harris. A significant concern for Democratic strategists is whether the party has plausibly declined since 2020 only among voters of color, without suffering material losses among white voters as well.

One strategist with access to a wide array of party polls, who asked for anonymity to discuss that private research, told me that although many Democrats are optimistic that surveys overestimate Trump’s strength among Black voters, a risk also exists that polls underestimate Trump’s strength with white voters (something that has happened before). That risk will rise if Trump turns out unexpectedly large numbers of the blue-collar white voters who compose the largest share of infrequent voters in the Rust Belt battlegrounds.

However, the Republican pollster Whit Ayres told me that he is seeing the same divergence between slipping non-white support and steady white backing for Harris in his surveys—and he sees good reasons for that pattern potentially persisting through Election Day. “The Hispanic and African American weakness [for Harris] is a function of a memory of the Trump economy being better for people who live paycheck to paycheck than the Biden-Harris economy,” Ayres said. “On the other hand, there are far more white voters who will be voting based on abortion and the future of democracy. There’s a certain rationale behind those numbers, because they are making decisions based on different issues.”

Democrats generally believe that they maintain a fragile edge in Michigan and Wisconsin, partly because many public polls show Harris slightly ahead, but even more because their party has built a better turnout operation than the GOP in those states. Pennsylvania looks like the toughest of the three for Harris and, in the eyes of many strategists in both parties, the state most likely to decide this breathtakingly close race.

“Looking statewide, I’ve always thought from the time she got in that Harris would do better in the suburbs and the cities than Biden, and Trump would do better in a lot of these redder counties, and the million-dollar question is what number is bigger and how much bigger,” Mikus, the Pittsburgh-based consultant, told me.

Biden carried the Keystone state by only 1.2 percentage points while winning the national popular vote by nearly 4.5 points. Whether Trump wins a second term to execute his dark vision of “retribution” against “the enemy from within” may be determined by whether Harris can hold Pennsylvania while winning the national popular vote by much less, if at all. It would be a fitting conclusion to this bitter campaign if the state that decides the future shape of American democracy is the same one where the nation’s Constitution was written 237 years ago.