It was hard to believe the words coming out of Governor Maura Healey’s mouth when she kicked off her reelection campaign. In a radio ad that aired early this year, she blamed Massachusetts’ high electricity bills on her climate policies, then actually bragged about people fleeing the state.
It was bizarre to hear this self-indictment of failure. Problem was, it was an AI-generated ad, made to sound like the governor and put out by one of Healey’s Republican challengers, Brian Shortsleeve.
Massachusetts voters got only a small taste of what’s possible in the world of AI campaigning. In Georgia, people saw Senator Jon Ossoff getting pampered in a hair salon in one ad then mocking farmers in the next. In Texas, viewers watched Senate candidate James Talarico deface a Bible by adding language saying that God called for six genders. He then turned to the camera and said, “I always think of myself as a Christian who hates Christianity.”
(Both candidates won their elections to the US Senate.)
Campaigns often argue that their AI ads are clearly satire. Some, but far from all, of the ads carry tiny disclaimers at the bottom that read, “This ad is AI generated,” but you’d need a magnifying glass to see that.
Political ads have always been nasty, but this is next-level deception. And it should be stopped. This campaign cycle, the Wesleyan Media Project identified at least $137 million spent on 220 AI-generated ads — surely an undercount — with 81 percent sponsored by Republican candidates or groups.
At least 31 states require disclosure of deepfakes used in election messages. Massachusetts is not among them. Yes, there wasn’t anything too awful this election cycle in the Commonwealth, but make no mistake, much worse is coming if action isn’t taken.
The Legislature is now considering a rule requiring notifications at the start and end of political ads if something is AI. Beacon Hill should pass this into law.
“I don’t think we’ll ever be able to stop AI from being utilized. It’s more just adding transparency,” said state Democratic Senator Michael Moore of Millbury, who introduced a bill to stop election misinformation. “Obviously when you know that, if you know what’s AI generated, you might be a little more inclined to look into the information.”
The editorial board also asks: Why stop with campaign ads?
In 1919, Supreme Court Justice Oliver Wendell Holmes wrote, “The most stringent protection of free speech would not protect a man in falsely shouting ‘fire’ in a theatre and causing a panic.”
Now imagine a person today trying to incite panic by posting a deepfake video of people rioting in your neighborhood. Would this be a modern, and moral, equivalent?
Going beyond the shades of legality, at minimum people need government regulation of AI images, videos, and audio to help separate fact from fantasy. This includes adding visible watermarks and metadata to delineate what’s AI. The standard should be simple: If somebody is interacting with AI, they should know it.
Identifying AI is also about preserving jobs. One small example — AI-created comics are stealing jokes from real stand-ups. How is that OK?
Google does allow users to trace imperceptible watermarks across many images, audio, and visuals to detect what’s artificial. That, however, requires web users to go to Google’s SynthID Detector to do the actual detective work. (And to actually know the detector exists.) The tool is also not fail-safe — when the editorial board fed it several AI-generated images, the program did not spot all the fakes.
Meta’s guidelines state that AI-content “may be identified and labeled.” TikTok “encourages” creators to label AI content. A cynic might fairly read this as: We’re considering the issue, but it’s not really our problem to fix. Quickly scroll through online social posts and it’s clear that AI-labeling policies remain mere suggestions.
And some deepfakes need to be stopped outright. Using the technology at its worst, users have posted AI-generated sexualized images of real women and children online, made with Elon Musk’s chatbot Grok.
If the tech titans won’t do a better job, the government must step forward. The European Union and China have stricter labeling standards and rules on what AI can generate. Surely, the United States can match them.
If Washington is unwilling to pass timely regulations, which seems like the plausible outcome with a president seemingly enthralled with fantasy AI images of himself, the states must act. Piecemeal regulation would not be ideal and could be hard to police across state lines, but would serve a purpose.
“It starts a discussion,” said Republican state Representative Steven Howitt of Seekonk, who introduced a bill that would require “conspicuous” labeling of AI-generated content. “A lot of times a bill is presented to start a conversation going ... between states as opposed to within the state.”
This feels like a wise approach. Then eventually, if enough states step up, Washington could be forced into action.
Editorials represent the views of the Boston Globe Editorial Board. Follow us @GlobeOpinion.