Pizza lovers swear that the way you cut up a pizza affects the way it tastes. Similarly, politicians know that the way you cut up a state’s districts can have a profound impact on which party wins the majority in that state. And for decades AI has helped politicians in this endeavor. What’s the story behind this early use case for AI?
AI Assists in Gerrymandering
Partisan gerrymandering is the practice of redrawing legislative and congressional districts to help one’s own party win more elections. When done right, gerrymandering can sneakily turn a politically red state into a state with a majority blue House seats (this interactive map will help you visualize the power of redistricting). Also, John Oliver has a further explanation, if needed.
And it’s actually very familiar ground for AI research:
The idea to apply computers to the redistricting process as a way to foster transparency and equity was first considered over half-a-century ago in 1961, and in 1965 an algorithm had been described that offered “simplified bipartisan computer redistricting.”
Daniel Oberhaus, Vice
By the time the 1990 census rolled around, algorithms had become the norm for legislators and special interest groups hoping to get a leg up during the redistricting process.
Gerrymandering, in essence, is an ideal playing field for AI because there is this confluence of historical data (# of voters, political leaning, economic standing, etc.) and predictive data involved (how might coming-of-age voters sway, how will these economic areas change, etc.).
Decades later, machine learning tools can design thousands of district maps in seconds to be analyzed and chosen as the best course of action for a party. These statistical models help create absurd, partisan districts like the 4th Congressional District of Illinois:
In a purple state like Wisconsin, gerrymandering can have a profound effect:
The new maps [designed by sophisticated computer models] efficiently concentrated many Democratic voters in a relatively small number of urban districts and spread out the remainder among many districts in the rest of the state. These are the twin techniques of gerrymandering, often called packing and cracking.
Emily Bazelon, The New York Times
In November 2012, Republicans won only 47 percent of the vote but 60 of 99 seats in the Assembly.
The fact that that is allowed seems absurd.
[However] There continues to be no way to measure at what point a district has been “gerrymandered,” and there are no hard rules for what a “fair district” is.
Daniel Oberhaus, Vice
That, coupled with other Constitutional reasons, led to the recent Congressional ruling:
Justice Roberts agreed that the states’ district maps were “highly partisan, by any measure”—and that the practice of gerrymandering is harmful—but wrote that federal judges have no right to reallocate power from one political party to another.
Louise Matsakis, Wired
Everyone recognizes the problem, Congress has deferred the duty, and so researchers believe we must take the matter in our own hands.
Remedies for Gerrymandering
There are a few direct and indirect ways in which I see AI remedying the current gerrymandering situation.
One, we could come to agreement and let a non-partisan AI create the official district lines. This is more theoretical than anything because we still run into the problem of “what constitutes a fair district?” and the possibility that the programmers will also have some political leaning. So that may be out the window. But:
Even if algorithms don’t end up drawing the districts themselves, they could provide a standard against which districts are judged to see if gerrymandering has taken place.
Kevin Stacey, Phys.org
As Kevin points out we could use the above non-partisan AI as a “standard” against which we measure the current lines.
In 2016, Wendy Cho and her colleagues at the University of Illinois Urbana-Champaign developed a complex algorithm that uses a supercomputer to generate hundreds of millions of possible district configurations in a matter of hours. The tool was designed to test an enticing idea: what if it were possible to generate every possible district map for a given state, and then use this set of all possible maps to judge the level of partisanship on an officially proposed district map? By using these hundreds of millions of maps as a point of comparison, Cho and her colleagues could statistically expose bias in redistricting procedures.
Daniel Oberhaus, Vice
This seems to me to be the most feasible option for the future. Right now, we’ll run into resource issues since this is a very costly method that requires a supercomputer to run. But I can see this being a widely used option by the late 2020s.
Lastly, and the most accessible option today, is using AI to create more transparency around the process of gerrymandering. Tools like districtbuilder, Bdistricting, and Auto-Redistrict provide the public with a way to visualize how different inputs can be used to change district lines and thus influence elections. This empowers journalists (like fivethirtyeight’s Gerrymandering Project) and individuals to understand what’s going on and petition for change.
We may never reach perfection. However, there clearly needs to be a change. Thanks to partisan gerrymandering, legislators pick their voters, even though it’s supposed to be the other way around.
With the 2020 Census (and thus redistricting) right around the corner and AI research vastly further along than it was in 2010 (the last redistricting), the stakes are higher than ever for the confluence of AI and gerrymandering.