March 7, 2024 | University of Illinois News Bureau
It鈥檚 estimated that $12 billion will be spent on political ads this election cycle鈥�30% more than in 2020. The sheer volume of ads is remarkable, and there is vast potential to use this political information to contribute to democracy: to reach more potential voters and provide accurate information. There’s also more potential than ever for generative artificial intelligence to misrepresent candidates and policies, leading to confusion in the voting booth. News Bureau editor Lois Yoksoulian spoke with advertising professor and department head Michelle Nelson about the topic.
How is AI used in political ads, and what are some recent examples of AI-generated political ads?
After President Joe Biden announced his re-election bid, the Republican National Committee released an AI-generated political advertisement that depicted a dystopian world should Biden be re-elected. The video ad features a number of 鈥渨hat if鈥� questions (e.g., What if crime worsens?) with AI-generated images. This is a classic fear appeal-negative attack ad. Emotional ads reach us, and we remember them鈥攍ike the famous used during the 1964 election cycle or the ad used in 1988. But they may also get some backlash for being too negative, too manipulative, especially because they are using AI.
Some say the Joe Biden RNC ad is the first AI-generated ad, but it鈥檚 hard to prove. However, in the top left corner of the ad, you see a disclosure in a small font: 鈥淏uilt entirely with AI imagery.鈥� The video description on the GOP鈥檚 YouTube page says, 鈥淎n AI-generated look into the country鈥檚 possible future if Joe Biden is re-elected in 2024.鈥� So they are not hiding the use of AI. There鈥檚 also a spoof ad that ran on The Daily Show, which uses an AI-generated voiceover for Biden. It鈥檚 pretty easy these days to use AI tools to create content and make it look pretty real.
Are the entities behind these types of ads required to disclose that AI has been used?
The Federal Election Commission, the governmental body that regulates political communication, including advertising in the U.S., sought public comments on amending a regulation related to the use of deliberately deceptive AI in campaign ads at their meeting titled 2023: REG 2023-02 Artificial Intelligence in Campaign Ads. They have not yet made a ruling here. However, states such as California, Texas, Michigan, Washington, and Minnesota have enacted new state laws where any deepfake AI images, video, or audio need to be labeled in. A similar law is being discussed here in Illinois. It鈥檚 not clear how these state laws relate to the Constitution.
Some tech companies have started requiring disclosures. For example, Google indicated that all AI-generated political ads shown on YouTube or other Google platforms 鈥渢hat alter people or events鈥� must have a prominent disclosure. There is also a verification process for political ads in some regions.
Microsoft has started their 鈥淐ontent Credentials as a Service鈥� which is like a watermarking tool鈥攊t lets users digitally sign and authenticate the media鈥攚hether it鈥檚 an image or video through their . The idea is that the watermark serves as metadata so users can see the history of the content. For example, 鈥渉ow, when, and by whom the content was created or edited,鈥� whether by a person or AI. This information stays with the content so anyone can see the origin.
To what extent AI-generated political content is now being watermarked or disclosed I have no idea. It seems difficult to police all of these images at this moment in time.
What is the potential harm of using AI in political ads?
The concerns for harm relate to possibilities for deepfakes and disinformation. Others can re-create something using AI voice or images without authorization to create stories and content that are absolutely untrue.
We鈥檝e had tools and technology in the past, like Photoshop to manipulate existing images. Do you remember the 2004 ads featuring altered photos depicting John Kerry, the Democratic nominee for president, and Jane Fonda at an anti-Vietnam War rally? Ads are regularly embellished or altered: The colors or lighting are changed to make people look ominous or more youthful. But this is different. This is a whole new level. Anyone can make AI messages using free tools such as Open AI鈥檚 ChatGPT or Google鈥檚 Bard. This may be good when used responsibly, or it may be harmful.
Our research shows that many people don鈥檛 know that the First Amendment protects political advertising in the U.S. Political advertising is not held up to the same standards as commercial advertising, say for a cheeseburger. The Truth-in-Advertising laws by the Federal Trade Commission do not apply to political ads. That means political ads can legally be deceptive or lie. Of course, there are libel laws, but they are rarely used in this arena.
Beyond this regulatory difference in commercial and political advertising, the idea is that we want free-flowing political information to contribute to a dialogue that can enhance our democracy. Too much regulation can have a chilling effect on political speech. Yet the potential today with AI to lie and deceive and to deploy these messages in a personalized manner and at a grand scale offers a challenge perhaps to our thinking about regulation and political advertising.
Do you have any advice on how consumers can spot the use of generative AI in political ads?
They can鈥檛. You can look for irregularities like the number of legs on the cat. But it is getting harder and harder to tell if AI makes an image or video. Whether it鈥檚 Taylor Swift, an image of Trump in jail, or a funny meme, we can鈥檛 always tell, especially as we scroll through our social media feeds. The best we can do is hope that the tech industry develops systems able to trace the origin of the content so we can see who made the image and when and how. If this tech is built in, it鈥檚 harder to fool people. We could also encourage disclosure mandates among tech companies.
But in the age of this kind of media and in an election year, we need to focus on the basics of media literacy鈥攍ooking for reliable sources of information, watching the political debates and trying to form our own informed opinions.
Editor鈥檚 note: To reach Michelle Nelson, call 217-344-5068; email nelsonmr@illinois.edu