David Inserra
For a while now, experts and policymakers have been growing increasingly worried about the threat that generative AI might pose to elections. Deepfakes, critics have warned, could become widespread, effective, and malicious enough to swing elections. Democracy itself may be under threat.
But a strange thing keeps happening: The greatest fears of AI and elections have not materialized. In fact, something different is happening. Instead of disaster, AI is playing an increasingly common and important role in modern elections.
The Growth of AI Political Ads
A recent and noteworthy example is the current election for Los Angeles mayor. Candidate Spencer Pratt and his supporters have gone viral with a variety of AI-created advertisements. In fantastical videos, Pratt is depicted as a Batman-like figure leading the people of LA to revolt against his political opponents who are shown as the Joker, out-of-touch royalty, and pseudo police thugs. Another depicts Pratt as a Jedi inspiring LA citizens to defeat his political opponents who are shown as forces of the “dark side,” including Darth Vader and Stormtroopers. Other ads look more realistic, satirizing his opponents’ own words or advertisements.
Democrats including California Governor Gavin Newsom, former New York Governors Andrew Cuomo and Eric Adams, and plenty of others have used generative AI tools for their own ads and online memes. For example, one of Newsom’s most popular AI-generated ads mocks his potential 2028 election opponent JD Vance and a couch.
Republicans including President Trump, the National Republican Senatorial Committee, and various others on the right are using generative AI in similar ways, such as President Trump’s claim of support from “Swifties.”
Generative AI is being used in both campaigns and political commentary to parody or satirize political opponents while showing the other side as principled, attractive, caring, and competent. This isn’t really anything new—many political ads have done similar things with traditional photography and media tools. While generative AI can create more realistic looking satire or parody that doesn’t require a cast of actors, a film studio and equipment, a team of video editors, or all the money that would be required to make a high-quality production, it does not change the fundamental speech at hand. In fact, this ease of use and lower costs can make it easier for candidates to be creative or communicate with constituents.
AI is also being used for other purposes in the electoral process beyond just satire and parody. This includes a variety of applications that can help inform voters, from candidate chatbots to answer voters’ questions to translation of political ads into other languages and real-time translation of speeches and meetings.
It is not just campaigns but their supporters and average Americans who can use AI tools to promote their preferred candidate or criticize opponents. Indeed, many of Pratt’s more viral videos have been made by supporters, not the campaign itself. By expanding the ability for more Americans to more easily create election and political content, the field of political commentary and advertising has been greatly democratized.
Concerns and Criticism of AI Ads
To be clear, AI-generated ads can be false and untrustworthy, and it can be more difficult to identify them as such. Bad actors could use political deepfakes to try to undermine elections and sow chaos.
But does reality live up to the panic? So far, the clearly malicious applications of AI have been rare, minimally impactful, and easily identified. The most prominent case was a faked Biden robocall in the New Hampshire primary before the 2024 election that urged people to “save their votes” for November. The deepfake was quickly identified, condemned, and corrected in the press, though the individual who carried out the robocall claims he was acting to warn about deepfakes. While he was found not guilty of criminal charges because Biden was not a declared candidate in the primary, he still faces civil liability and a Federal Communications Commission fine.
Much of the political backlash has targeted satirical or critical content rather than direct voter-suppression schemes. And again, this isn’t a partisan issue.
Republicans including Rep. Elise Stefankik (R‑NY) have been highly critical of AI videos that mock her interest in running for New York governor, calling such videos “disgraceful,” “offensive,” and a “disservice to the voters.”
Democrats have their own frustrations with being targeted by AI videos. LA Mayor Karen Bass has called the recent AI videos supporting Pratt “150 percent fiction” and “a violent trend” that could “provoke people who are unstable and … jeopardize people’s safety.”
A parody of Kamala Harris calling herself the “ultimate diversity hire” and “a deep state puppet” and claiming to hide “her total incompetence” led California Governor Newsom to respond by passing AB 2839. The law regulated “materially deceptive speech,” including AI-generated content, that could harm the reputation of a candidate or undermine faith in elections. A federal judge found the law unconstitutional, writing that it “strikes at the heart of the First Amendment” by attempting to “pre-emptively sterilize political content.”
Sen. Amy Klobuchar (D‑MN) wrote a New York Times op-ed calling for “urgent” action in response to a satirical deepfake that depicted her saying that “Republicans have girls with perfect titties in their ads” and that Democrats were the “party of ugly people.” Klobuchar worried that people would be fooled by the satirical AI-generated content and that, even if people know it is fake, they still might come away with a negative opinion of the person depicted.
A media environment with more false and deceptive information is not a good thing for our society, but it is not the government’s job to fix it. In fact, a policy fix may create even more problems than it solves.
It is understandable that no one likes being criticized, especially when AI tools can do an effective job of making someone look rather foolish. And it certainly is the case that AI content could confuse or mislead people who don’t realize a video is fake. But embracing laws to curtail AI-powered speech silences both legitimate satire and criticism as well as purposefully deceptive speech.
For example, a bipartisan group of senators, including Sens. Klobuchar, Marsha Blackburn (R‑TN), Chris Coons (D‑DE), and Thom Tillis (R‑NC), is seeking to advance the NO FAKES Act in Congress. As I summarized my concerns previously, NO FAKES
directly outlaws multiple categories of currently legal speech including AI generated content; legally and financially encourages companies to remove even more legal speech; introduces a vast notice and takedown regime that favors trolls and the overly litigious; creates requirements to broadly filter speech regardless of intent or context; restricts the development of AI products; and forces companies to identify anonymous users who break these new requirements.
Such a law would be bad news for all sorts of speech online, including political commentary, humor, satire, and parody.
Speech, Progress, and Democracy
Rather than attack and censor generative AI, we must remember why we protect speech in the first place.
A common theme across the history of satire and parody is that such techniques hold up a mirror to society and shine a light on error and abuse. They are often used to try to hold those in power accountable and prompt reform. Through humor and absurdity, satire and parody can illustrate complex criticisms and arguments in ways that are more approachable than normal journalism or debate. In short, such literary tools can be powerful and important tools of social and political commentary.
And so, it should come as no surprise that such powerful tools are nothing new, nor is the frustration that has accompanied them. Here are just a few examples stretching from antiquity to today:
The Greek epic poem The Battle of Frogs and Mice parodied Homer’s Iliad.
Juvenal’s Satires were aggressive attacks on corruption in the Roman Empire.
Chaucer’s Canterbury Tales satirized the Catholic Church and nobility for their corruption and hypocrisy.
Charles Dickens employed satire to criticize Victorian social structures and greed across his works.
Mark Twain used satire to attack racism in the Adventures of Huckleberry Finn.
George Orwell’s Animal Farm satirized the Russian Communist Revolution and Stalinism.
Modern examples of satire and parody range from movies such as Airplane! and The Death of Stalin to comedy shows including The Colbert Report and Saturday Night Live and satirical websites like The Onion and The Babylon Bee.
Of course, not all satire, parody, and critical speech is kind, insightful, or correct. Such speech may be considered by some to be offensive, misleading, and cruel and yet can still be important to civic discourse. The reality is that speech has the power to tear down and build up; AI-powered speech is no different. AI tools make it easier for anyone to express themselves in ways that are even more fantastical, absurd, pointed, or deceptive. But the response should not be to create an even greater risk by making the government the arbiter or what kind of speech is or is not allowed.
Empowering People and Innovative Solutions
The promise of a liberal society is that we trust individuals to sort through all different kinds of speech and ideas to determine for themselves what is worthwhile. As the judge found when striking down California’s AB 2839, “When it comes to political expression, the antidote is not prematurely stifling content creation and singling out specific speakers but encouraging counter speech, rigorous fact-checking, and the uninhibited flow of democratic discourse.”
A better response, then, to the fears around generative AI is to recognize that not all people are gullible. Instead, research shows that people respond to misinformation in a variety of complex ways, including humor, criticism, fact-checking, sharing with others, or support. We also tend to accept things that fit our preexisting worldviews or group identities without much critical thought or deep belief because such content is “too good to check.” Developing a sense of intellectual humility would go a long way to treating others with respect while recognizing that we ourselves are flawed and believe in half-truths or falsehoods.
This means that education and preparing individuals to be more conscientious consumers of media is a better policy solution. Private companies, civil society, and the government can support improved media literacy to empower individuals to spot sketchy content and be somewhat cautious without becoming cynical and skeptical of everything.
The market can develop more effective tools to help support citizens in their efforts to debate, fact-check, and grapple with what is true. For example, newer tools, such as community notes on social media, are trying to encourage users to engage in fact-checking and develop consensus that cuts across our political tribes. And just as new AI products continue to emerge, so too are tools that watermark and identify AI-generated content.
Conclusion
As the LA mayoral race illustrates, all forms of speech can be important to democratic debate. While AI is changing the way we communicate, the worries over how this new form of speech will be used to deceive or harm are similar to those of the past. The instincts to empower the government are also not new and will result in silencing satire, prohibiting parody, and cracking down on criticism. And the consequences of regulating speech in the name of truth are never as simple as they seem.
You do not have to like AI to recognize that free speech matters most when it protects speech we dislike. So let’s trust the American people, not the government, to decide for themselves how AI should be used in political advertising and commentary.