A landmark Supreme Court ruling last month unleashed a nationwide redistricting scramble. In the legal battles that will follow, artificial intelligence may prove to be the sharpest tool in the courtroom.

The Supreme Court’s gutting of Section 2 of the Voting Rights Act set off a new wave of redistricting oneupmanship, as red states and blue states alike are poised in the coming years—and in some cases the coming months—to redraw their political maps, dividing up voters to help one party win more elections.

Many of those maps are likely to end up in court—either under challenges under federal law guiding how districts can be drawn or even stricter state laws. Legal and political science experts say A.I. algorithms—which can draw and analyze millions of possible maps for a given state—are increasingly being used by both sides in such fights.

In November, a Utah judge threw out a congressional map drawn by the Republican-led state legislature for violating a state proposition voters approved in 2018 that was intended to end partisan gerrymandering. In her ruling, the judge cited computer-simulated analysis that found that the proposed map was “more Republican than over 99% of expected maps drawn without political considerations.”

Legal opinions based on such arcane analyses are likely to become more common, as A.I. takes on a larger role in high-profile redistricting battles that could decide control of Congress.

Much like generative AI models such as ChatGPT or Claude are trained on countless texts from the Internet, the algorithms used for analyzing congressional maps are trained on millions of randomly generated maps through a complex sampling process. The A.I. algorithm can then come back with analysis of how a map passed by a state legislature compared to the universe of possible options.

The algorithm’s role in the redistricting fight could be pivotal, as courts race to rule on disputed maps, often within a limited timeframe. Supreme Court Justice Elena Kagan argued in her dissent in the landmark Rucho v. Common Cause case in 2019 that gerrymanders will only get worse, as data becomes ever more fine-grained and data analysis techniques continue to improve.

“Old-time efforts, based on little more than guesses, sometimes led to so-called dummymanders — gerrymanders that went spectacularly wrong,” Kagan wrote. “Not likely in today’s world. Mapmakers now have access to more granular data about party preference and voting behavior than ever before.”

Tyler Simko is an assistant professor at the University of Michigan and is part of a team that has developed the Algorithm-Assisted Redistricting Methodology (ALARM) Project that develops algorithmic tools that essentially draw “thousands and thousands of maps” that comply with certain rules at scale. He pointed to a state as vast as Texas, which needs to fit over 9,000 voting precincts into 38 voting districts. There are trillions of possible maps that can be theoretically drawn. AI can generate nearly all of those maps within a few minutes to an hour, Simko says.

“We might take a claim for partisan gerrymandering by saying, for example, the expected number of districts that might go to one party is extremely high in the enacted plan. And [we can] compare this large, neutral set of thousands or tens of thousands of plans that can be drawn with tools like ours,” he says.

Such analyses are likely to get more attention in the wake of the Supreme Court’s gutting of Section 2 of the VRA, which made it harder to prove in court that minority votes are intentionally diluted in the redistricting process. Mark Gaber, Senior Director of Redistricting at the Campaign Legal Center, stressed that the ruling did not give state lawmakers carte blanche to draw maps however they wish. “The Supreme Court didn’t eliminate Section 2. They just substantially narrowed it,” he says.

How that new reality will play out will vary from state to state. “If the allegation is that the map is a partisan gerrymander, then that depends on whether the state law prohibits that in either statutes or constitution,” Gaber says. “In federal court, if you’re bringing a Voting Rights Act claim now…you’d have to show extra steps that you didn’t used to have to…so that’s a taller task.”

How a congressional map gets drawn

Redistricting fights have been intensifying nationwide since Texas’ unprecedented mid-decade redistricting ruling, yet the process behind drawing congressional maps, and who controls it, remains largely opaque. Experts say A.I. is likely already involved, and that role is likely to expand as more states rush to redraw their maps.

In late April, two days after Florida Governor Ron DeSantis gave the state legislature a newly drawn congressional map intended to give Republicans four additional seats, the legislature voted to adopt it without knowing who exactly was involved in drawing it. When pressed by lawmakers from both parties, DeSantis’ senior official refused to provide more clarity about the map-drawing process.

Even if A.I. is used more in redistricting, human input remains essential, experts say. Paul Mitchell, a demographer who drew a new California congressional map last year intended to give Democrats more seats and which the state’s voters approved, explained that the process of drawing a map starts with conversations between a map drawer and congressional and local party leadership. Based on census data and geographic data of the state, Mitchell then came up with different versions of the map that are compliant with the U.S. Constitution, which requires each congressional district to have as equal a population as possible.

“You want to draw districts that follow natural boundaries, like following major highways or streams or mountains that are compact, meaning that they are keeping nearby populations with each other, instead of having crazy districts that end up looking like two octopuses wrestling,” Mitchell said.

Normally states only draw new congressional maps every ten years. That process usually takes up to six months as map drawers take input from a wide range of local leaders. California’s new map was drawn in three weeks. He says he drew the map but A.I. helped guide the process.

Even though the goal of the new map was to counter Republican gerrymandering in Texas, that doesn’t mean drawing it was simple, Mitchell says. “The relationship is more like, ‘is it possible to pick up Democratic seats? And if so, what do those districts have to look like?’ And then you have to have conversations about what we’re comfortable with, Mitchell said. “It is a little bit more complex than just somebody walking in and saying, ‘I want this many congressional seats that are Democratic, and to this percentage.’”

Although AI is proving to be a powerful tool for lawyers in redistricting cases, there are limitations in what it can do for map drawers. “You don’t say to an AI, ‘Hey, draw me the perfect map.’ And then AI draws you the perfect map,” says Mitchell, who adds that where AI can be helpful is reviewing draft maps and giving him specific guidance on how he may tweak it.

“What you have is AI being able to help you analyze the map that you have drawn and say, ‘Hey, this map has fewer Democrat [districts] than you can actually get if you try harder.’ Or the computer program says, ‘When I drew the maps 90% of the time, I only split 34 cities, and your map split 60 cities. So your map splits more cities than it has to.’”

While the current redistricting free-for-all is mostly being looked at through a partisan lens, a change in districts in multiple states could also have unintended consequences, from the kinds of political figures that gain power in the coming years to the policies that get traction.

“Are you electing legislators who are going to support rent control, or are you drawing districts where you are electing legislators who are going to fight for ending homelessness in the suburbs,” Mitchell says. “An algorithm can’t make all those choices for you.”

Even as the technology for drawing district maps progresses, humans will still shape the outcome in significant ways, says Simko.

“Like in this community, we want to keep the district whole by not splitting one particular river,” he says. “Things like that only humans can bring. Those will continue to be true.”

The post AI Is Already In the Redistricting Fight. Just Don’t Ask It to Draw the Perfect Map appeared first on TIME.