by Jerry Cayford
A friend of mine covers his Facebook tracks. He follows groups from across the political spectrum so that no one can pigeonhole him. He has friends and former colleagues who, he figures, will be among the armed groups going door to door purging enemies, if our society breaks into civil anarchy. He hides his tracks so no one will know he is the enemy.
That trick might work for the humans, but artificial intelligences (AI) will laugh at such puny human deceptions (if artificial intelligence can laugh). When AI knows every click you make, every page you visit, when you scroll fast or slow or pause, everything you buy, everything you read, everyone you call, and data and patterns on millions like you, well, it will certainly know whom you are likely to vote for, the probability that you will vote at all, and even the degree of certainty of its predictions.
All of that means that AI will soon be every gerrymanderer’s dream.
AI will know not just the party registrations in a precinct but how every individual in a proposed district will (probably) vote. This will allow a level of precision gerrymandering never seen before. There is only one glitch, one defect: with people living all jumbled up together, any map, no matter how complex and salamander-looking, will include some unwanted voters and miss some wanted ones. To get the most lopsided election result possible from a given group of voters—the maximally efficient, maximally unfair outcome—the gerrymanderer has to escape the inconvenience of people’s housing choices. And since relocating voters is not feasible, the solution is to free districts of the tyranny of voter location. The truly perfect gerrymander that AI is capable of producing would need to be a list, instead of a map: a list of exactly which voters the gerrymanderer wants in each district. But that isn’t possible. Is it?
Our idea of a district is a place, a community, a geographical area where voters live in proximity to one another. The technical properties are “compactness” and “contiguity”: a district is a physically contiguous patch of land, meaning it is all one patch, and all of it compactly situated pretty close together. This idea of districts is only a constraint on gerrymandering, though, if it is codified in the law—which both compactness and contiguity are in most (but far from all) states—and if we enforce the law.
A gerrymandered district is one that has escaped the constraint of compactness. The term derives from an infamous 1812 redistricting: “Elbridge Gerry, the governor who signed the bill creating the misshapen Massachusetts district, was a Founding Father: signer of the Declaration of Independence, reluctant framer of the Constitution, congressman, diplomat, and the fifth vice-president” (Smithsonian Magazine). The outraged 1812 opposition coined the term: “It looked like a salamander, another dinner guest noted. No, a ‘Gerry-mander,’ offered poet Richard Alsop.”
An AI-created, perfect gerrymander would have to go one better and escape both compactness and contiguity. It might not be visualizable at all. If it looked like anything, it would be an abstract, pointillist painting, with a dot for every voter, color-coded to the voter’s assigned district. This would be, clearly, a mockery of our concept of a “district.” But that doesn’t mean it couldn’t happen, so let’s ask how it might.
You can infer from the prevalence of gerrymandering today that our expectation of compactness is routinely violated. As a primer on redistricting puts it, “In practice, compactness tends to be in the eye of the beholder.” Having gotten this far, then, the remaining barrier to a perfect gerrymander—the gerrymander that AI will soon be able to give us—is the requirement of contiguity. The same primer says, “In practice, the vast majority of congressional districts … will be drawn to be contiguous.” The reason contiguity is still respected (so far), while compactness is not, is surely partly that compactness is on a sliding scale, but contiguity is basically yes or no. Still, the primer tells us “Few redistricting concepts are absolute, and contiguity is no exception” (emphasis mine), so we should ask how a gerrymanderer could get around it.
Following their success ignoring compactness, gerrymanderers might just brazenly brush aside contiguity, too. Dick Cheney’s lawyer, David Addington, described this tried-and-true political strategy in immortal words: “We’re going to push and push until some larger force makes us stop.” So, what larger force might stop the ultimate gerrymander? Since contiguity is legally required in many places (and taken for granted everywhere), our first thought would be that the courts are that force.
The Congressional Research Service publishes useful summaries of redistricting issues and laws. (Congressional Redistricting Criteria and Considerations is a good start, with links to more specifically legal summaries.) The ominous bottom line, though, is that the Supreme Court, after waffling for some decades, decided in 2019 in Rucho v. Common Cause that courts should stay out of gerrymander fights. As the high court succinctly put it, “Held: Partisan gerrymandering claims present political questions beyond the reach of the federal courts.” So, the courts have chosen not to be that “larger force,” but rather fobbed off on legislatures the defense of our democracy from the evil of gerrymanders.
Legislators are, of course, the main beneficiaries of gerrymandering, so it’s a bit discouraging to hear they are also supposed to be our main force stopping it. Still, politicians don’t like incurring the public’s anger, and our whole concept of legislative districts is that they are places, maps, contiguous pieces of territory. Turning them into lists would so violate the public’s expectations that surely any legislator would be loath to even try it. And yet legislators have become quite practiced lately at violating public norms. For example, redistricting is only supposed to happen every ten years, after the census, but recall the Democratic Texas legislators’ famous 2003 flight to Oklahoma to deny the Republicans a quorum for a vote to redistrict, just one of many examples of legislators defying the public’s expectations of what’s right and proper in order to redraw district lines mid-cycle. A polarized public did not unite against them. Public sensibilities, then, also do not seem to be that larger force that will stop the AI-driven super-gerrymander.
Since contiguity is more open and shut than compactness, we will no doubt try the courts again; they might throw their hands up at adjudicating degrees of compactness but still insist that contiguity is the law. So, it is worth considering what arguments the gerrymanderer might make. Even if we hadn’t all seen some shockingly political judgements in real court cases recently (I don’t dare give examples, but I’ll give you this), we have all watched enough courtroom dramas to have a sense of the weaselly, hair-splitting, word-twisting arguments that sometimes convince judges. So, let’s imagine we are shyster lawyers trying to give a sympathetic judge just enough wiggle room to get away with rejecting contiguity and ruling for our gerrymandering clients. What do we have to work with?
As the redistricting primer describes it, “A district is contiguous if you can travel from any point in the district to any other point in the district without crossing the district’s boundary.” Obviously we can’t do that, if we want to be able to put voters from any old place into a district. But if we could get rid of the idea that a district has a boundary at all, maybe we could get away with it. A list, after all, doesn’t have a boundary; we can get from any point on a list to any other without leaving the list. So, does a “district” have to be a geographical space?
We go to Wikipedia (our preferred source for legal wisdom): “An electoral district … is a subdivision of a larger state (a country, administrative region, or other polity) created to provide its population with representation in the larger state’s legislative body.” Hmmm… a subdivision of a state or polity seems somewhat spatial, but not too rigidly. “A state is a centralized political organization that imposes and enforces rules over a population within a territory. Definitions of a state are disputed.” That sounds promising: a state is the political organization within a territory, but not the territory itself! Even less physical. And disputed! Let’s check our backup, “polity,” for confirmation. “A polity is a term for an identifiable political entity, defined as a group of people with a collective identity, who are organized by some form of institutionalized social relations.” Bingo! A polity is people, not land at all. We’re ready.
“Your Honor, we humbly submit that our client’s proposed electoral districts, composed of lists of voters assigned to each district by their AI program, fully comply with all legal requirements. The requirement of contiguity is inapposite to districts such as these, which perform their political functions as districts without being a physical location. Being inapposite, the requirement should be considered presumptively fulfilled. We ask that you find in favor of our client.”
If this scenario seems far-fetched, remember that the question is not whether the perfect, big-data-AI-generated, maximally unfair gerrymander fits our conventional ideas of redistricting; the question is what larger force will stop it. The courts have clearly expressed their wish to leave it to the political branches; the politicians have clearly expressed—through their bipartisan embrace of it—the belief that they must gerrymander or die; and the public is too divided and too uncertain to force a different outcome. Things look bleak.
But there is a fourth possible force I have not yet considered: that very AI so ominously poised to do us irreparable harm. Can we fight fire with fire and turn the talents of computers to good and not evil? If AI is allowed to design districts unconstrained by either compactness or contiguity, we can expect it to do the job very, very well, locking in lop-sided, one-party rule beyond the reach of democracy to counteract. But computers intent on preserving the legal constraints can do that equally well. What if we turn computers to constructing districts that truly are compact and contiguous?
Compactness may be in the eye of the human beholder, but the blind mathematical eye can see it clearly. Compactness is an easily definable mathematical property. Once defined, it is as open and shut as contiguity; this district is compact, yes or no. (There are several different ways to define it, but they are highly correlated with one another.) MGGG Redistricting Lab (formerly the Metric Geometry and Gerrymandering Group) has been studying for years how best to use mathematical modeling to create fair political districts. The time has come—and the technology is ready—to take redistricting out of human hands. Not to undermine democracy, as AI is capable of doing—and will do, if the current gerrymandering free-for-all continues—but to increase competition, weaken incumbency, and restore democracy.
Some of the issues I raise are discussed in greater detail by Douglas Rudeen in “The Balk Stops Here: Standards for the Justiciability of Gerrymandering in the Coming Age of Artificial Intelligence,” which concludes that “the only reliable way to forestall gerrymandering in the age of AI is to employ a form of fully-automatic redistricting.” It may seem paradoxical that abandoning human discretion to machines can increase human freedom. But in the end, the computers are going to be doing our redistricting either way. Our choice will be between computer gerrymandering that pre-determines electoral outcomes, depriving us of democratic control of our lives, or computer redistricting that creates compact, contiguous, fair, and competitive districts. If we choose the latter, we can without fear join Jeopardy! champion and fellow human being Ken Jennings in his gracious and cheerful concession after being trounced by IBM’s Watson: “I for one welcome our new computer overlords.”
In all seriousness, though, the choice is ours. The courts have chosen not to prevent gerrymandering. The politicians have chosen not to stop doing it. AI is not capable of choice (so far, anyway). That leaves the largest of those larger forces, us, the public. We loathe gerrymandering, but we dither. We can either have AI serve democracy, or let would-be overlords use AI to create super-gerrymanders. They will certainly do so, unless we choose to make them stop.