“A” is the alpha. The firstborn. Numero uno. Firsts are often a little stiff, burdened as they are with being so visible, so vulnerable—if this whole thing fails, won’t it be their fault? But with great responsibility comes great likeability.

As for “I,” who among us can’t relate to its skyscraper-like appearance, signifying that attention must be paid? Put these revered letters together, however, and they spell D-O-O-M, according to some. Others see the digraph not as dastardly, but as our salvation.

Election season is a new canvas for this world-defining chiaroscuro. Amid the political ads we all know and loathe has appeared a new strain: deepfakes. Librarians are at the forefront of efforts to maintain information integrity, so we should know as much as we can about this new and growing peril.

THE RISE OF DEEPFAKES

First, what is a deepfake? It is “an artificial image or video (a series of images) generated by a special kind of machine learning called ‘deep’ learning (hence the name).” Basically, it’s a realistic-looking picture, audio, or video of a situation that never happened. Deepfakes have been used for years in a number of industries, including politics, with little fanfare. But they made headlines back in January with a series of robocalls of President Joe Biden seemingly discouraging people from voting in the New Hampshire primary, which he called “a bunch of malarkey.”

Except it wasn’t Biden. It was an AI-generated voice that sounded like him.

CONCERNS

Of course, artificial intelligence is neither good nor bad. It is a tool, an apparatus, a thing to be used. No doubt many of those uses are outstanding. But the downsides can be significant too. According to the Brennan Center for Justice, a recent survey asked Americans about their concerns regarding AI, and 85% said they were “very concerned” or “somewhat concerned” about the spread of deepfakes.

This concern poses a threat of its own, as venal individuals can manipulate it by falsely claiming that legitimate audio or video content is, in fact, artificial, and therefore fake. Law professors Bobby Chesney and Danielle Citron call this the “liar’s dividend.” As deepfakes become harder to spot, they argue, false claims that real content is AI-generated will become more persuasive as well.

You can see how this would be particularly appalling in the political realm.

GOVERNMENT REFORMS

The good news is, those in power are aware of the problem and are trying to do something about it. Last year, a bipartisan bill was introduced in Congress “to prohibit the distribution of materially deceptive AI-generated audio or visual media relating to candidates for Federal office,” although it hasn’t become law yet. Not content to wait for legislation, in February 2024, the Federal Communications Commission outlawed the sort of robocalls that had been used in New Hampshire the prior month. The Cybersecurity and Infrastructure Security Agency, a division of the U.S. Department of Homeland Security, has an excellent white paper discussing the electoral risks of AI and recommending strategies to mitigate them.

Reforms are also underway at the state level. According to the National Conference of State Legislators (NCSL), 16 states have statutes governing the use of AI in campaigns and elections. Most were enacted in 2023 or 2024, although two states—California and Texas—were ahead of the curve, passing their legislation in 2019.

Other states have considered AI-related laws but haven’t yet reached a consensus. One challenge, according to NCSL, is defining precisely what AI entails. Another is determining the threshold for when AI use triggers a legal restriction. There are also First Amendment considerations, such as when AI is used to create, say, political satire. Finally, some states don’t see the need for this sort of legislation, saying, for example, that existing anti-defamation laws should be enough to protect candidates for office.

BEST PRACTICES

Naturally, new laws aren’t the only solution. The Brookings Institution describes eight best practices for state officials to “defend elections from the dangers of AI.” There are also education efforts such as “prebunking,” which works like this: Debunks don’t reach as many people as misinformation, and they don’t spread as quickly. When we are told that the misinformation is false, research suggests that it continues to influence our thinking. Better, then, to stop misinformation before it spreads.

Finally, earlier this year, 20 tech companies, including Google, Amazon, Microsoft, Meta, OpenAI, X, and TikTok, signed an accord to try to prevent their software from being used to interfere with elections. The pledge commits the signatories to take eight concrete steps, such as “seeking to detect the distribution of Deceptive AI Election content” and “providing transparency to the public.”