Is AI Coming to a Campaign Near You?
WHY YOU SHOULD CARE
Because speed kills in politics.
By Daniel Malloy
This summer when the liberal pot-stirrers at Media Matters for America wanted to figure out how often Donald Trump was retweeting or quoting fake accounts, they turned to artificial intelligence. Researchers built a computer program to scan Trump’s feed and highlight when he promoted suspicious Twitter accounts considered bots or “sock puppets,” finding dozens of examples. It was but a baby step into the political implication of computers that think for themselves. “Not just can we have AI do things like that moving forward, but it can turn around and tell us over time what other anomalous activities” can be found in Trump’s tweeting, says Angelo Carusone, president of Washington–based Media Matters.
As the emerging technology shakes up fields from trucking to law to journalism, AI is coming to politics — with vast potential to improve and streamline campaigns, lobbying and lawmaking. While there’s plentiful talk about fake news spread by Russian bots, much of the work that shook the 2016 election appears to have been human-driven. At this early stage, there are only a few cases of AI making a political splash. Chatbots, for example, mostly targeting younger Hillary Clinton fans, helped register and engage with voters.
On the political side, AI is already deployed in opposition research and media monitoring.
Machine learning has helped campaigns carve up reams of voter data. Bryan Whitaker of Democratic data firm TargetSmart predicts that by the 2018 elections, AI will be used to solve one of the thorniest problems for political data crunchers: how to connect voters’ online behavior — public tweets, Reddit posts, Instagram shots and so on — with offline voter data that can be used to target door-knockers and mail and television ads.
Those uses are not yet practical. Scott Tranter of Republican data firm 0ptimus compares deploying brain-mimicking AI neural networks on voter identification, persuasion and turnout to taking an F-16 to the drugstore. It will get you there quickly, but is it worth the hassle and expense?
Still, he said, the cost of applying such tools has declined dramatically in recent years thanks to cloud space offered by Microsoft and others, meaning that if he did want to tap into AI he wouldn’t have to buy a $200,000 computer. Tranter says the evidence he’s seen shows only a 2 percent to 3 percent improvement in targeting by using AI — but as the cost comes down, that might be worth it.
Where Tranter sees a benefit worth the expense on neural networks is in redistricting, which is coming up across the country after the 2020 census. Whether you’re looking for a partisan advantage or serving on a nonpartisan commission — each state does it differently — AI could be used to draw multiple district map options swiftly and make sure they comply with the byzantine laws and court precedents around gerrymandering.
Tranter is also looking at siccing bots on one of Washington’s most challenging — and lucrative — questions: Can this bill pass? Big legislation often moves markets, so traders want to know the scoop, and interest groups spend a lot of money trying to divine the political winds. “What would it take for us to get DACA?” Tranter asks, referring to the hotly debated amnesty for some children brought illegally to the U.S. “What are the scenarios? You’re not just getting on the phone with someone from [lobby giant] Squire Patton Boggs and getting the mealy-mouthed prediction answer. This is [a computer providing] a percentage chance.”
On the political side, AI is already deployed in opposition research and media monitoring. Tech startup Veritone sells software to campaigns and interest groups across the political spectrum that tracks an astounding array of television and radio programming around the country — more than 750 livestreams per day. Politicians can search their names to figure out how they’re being discussed, or outside groups can delve into preferred topics.
Television has long been a staple for media-monitoring firms, but the vast landscape of radio is harder to track, even though talk radio is often critical in a local political scene. Veritone also takes in YouTube videos and podcasts. The company’s instant bot transcriptions are not perfect, but they are faster than a campaign worker sifting through the material. “With the news cycles increasingly compressing, speed does become more of an issue; it’s a competitive advantage for us,” says Kristy McKnight, senior vice president at Veritone.
Instant transcription can also be applied to a campaign or outside group’s own audio and video, culled by trackers who follow opposing politicians around waiting for a slip-up — such as when Virginia U.S. Sen. George Allen famously called an Indian-American tracker a “macaca,” a racial slur used to describe Arab and Black North Africans. Moments like that 2006 gaffe are obvious, but searching a candidate’s back catalog for inconsistencies and wacky statements that go unnoticed at first glance could be a fruitful AI endeavor.
As AI adds speed and efficiency to existing tasks, campaign decision-making could be headed for a major shift. “You feed in every PowerPoint you’ve ever done to [IBM’s] Watson, every Excel spreadsheet with a campaign budget, historical TV ad buying,” Whitaker says. “Ask Watson the question: Where should I send volunteers to canvas tomorrow? … Watson’s going to at least provide you with options in a way where you don’t need to go to a data director, a digital director, a field director.” That’s right: Machines are coming for political operatives’ jobs too.
*Correction: An earlier version of this story gave an incorrect location for Media Matters’ headquarters. It is based in Washington.