AdvertiseDonate
Headlines Times · 3597 East Monarch Sky Lane, Suite 240, Meridian, ID 83646
Local reports · CalMatters

Artificial intelligence flourished in California election advertising — and divided voters

By Denise Amos, CalMatters ·

This story was originally published by CalMatters. Sign up for their newsletters.

This election year Californians saw political candidates battle massive robots, wax heroic as caped crusaders and kowtow to bloodthirsty crowds and rich fat cats. 

These were all fake images, generated via artificial intelligence to help political campaigns stand out online and on air. Election campaigns across the country spent more than $137 million on AI-generated ads by mid-October, according to researchers at the Wesleyan Media Project, with Republican groups accounting for about 80% of those messages.

Many of the spots featured humor and exaggeration — such as former Los Angeles mayoral candidate Spencer Pratt’s depictions of himself as Batman, and failed gubernatorial candidate Steve Hilton’s turn as a gladiator wowing a crowded coliseum that included rival Xavier Becerra and Gov. Gavin Newsom. 

Other campaign ads weren’t exaggerations and seemed more realistic, such as a mailer circulating in a San Diego suburb showing a male school board candidate behind a desk and his female opponent working a fake bake sale. She said it was a sexist image.

Are AI-generated images effective in political ads? We asked Ramesh Srinivasan, a UCLA professor of information studies who hosts the Utopias podcast. We edited his comments for length and clarity.

More campaigns seem to be using AI-generated images or deepfakes — or am I just just noticing it?

Ramesh Srinivasan is a UCLA professor of information studies who also hosts the Utopias podcast.

Absolutely, that’s true. The president has put out plenty of AI-generated content. Yes, there’s no lack of what we call “AI slop” that’s almost everywhere. But I think the issue is less the AI slop and more the way in which AI is being used to promote and create — we could almost say — propaganda-type targeting. 

In the last 10 years, we had a lot of buzz about the Cambridge Analytica story, which was the way a third-party organization was advising many right-wing and other candidates, using people’s data to create content to target them based on predictions of their psychology. Everybody was pretty worried because, in theory, you could use data that you gathered about people, possibly through purchasing their credit card records from data brokers, and you can target someone and influence them — not around who to vote for but even the desire to vote at all.

Now we are seeing not just that concern — which never got resolved — but the likelihood that AI systems can, in real time, create content that targets you based on your psychology, without going through the entire multi-step process of buying one’s data, gathering the data and creating content based on the data. 

So the days of third-party propagandists using data to psychologically target people is something of the past because AI can do that automatically. 

Are you saying that’s because the algorithm automatically delivers to an individual voter an ad that’s going to be tailored to them?

Exactly. Because if I’m able to gather and access a bunch of data about you — not just data you put online on Instagram, but actually data about your credit card records —  I can use all of that data to essentially influence what you see and what you don’t see, which is already happening with social media. 

But in this case, AI can create content that directly targets you based on data about you. So that  is why this is very pernicious, because people don’t have disclosure around when something is or is not AI-generated, which is why I’ve been working with some folks in the U.S. Congress and folks in our state Senate to try to, at a minimum, force these third parties to watermark anything that is created by AI.

What do you mean by watermark and why? 

Watermarking is where an image says where it comes from. It’s something that’s put on an image that can never be written over. It would say “AI generated.” It’s like a way of flagging something. A lot of the content we were seeing prior to the election was AI-generated on social media platforms, on TikTok, Instagram, etc., and none of that was flagged. 

A recent ad posted on Republican gubernatorial candidate Steve Hilton’s YouTube account portrays him as a gladiator.

The Steve Hilton AI ads often had a line saying they were AI-generated.

A lot of them are doing it. The bigger picture is that people need to know what is or is not “real” or human-created when they encounter content online, because many of us are experiencing the wider world through online platforms.

Are voters being fooled by deepfakes? Are they confused? 

It’s impossible in many cases to be able to tell the difference. And that is not really our fault as voters or as citizens. 

Of course, there’s AI slop, where you’re not trying to hide that it’s being created with AI. For example, the president when he tweeted out that picture of him, like Jesus or something like that, healing the sick. The patient in the bed looked like Jon Stewart (host of “The Daily Show”). Everyone knew that that was AI. The president wasn’t trying to hide that.  

We experience a lot of stuff that is sort of just odd, weird and surreal looking. But the goal in those cases isn’t to fool people.

What ends up being effective though? Does AI convince people to vote for a candidate? 

It has to do with believability and trust. Based on initial research, we don’t see significant evidence that people are able to tell between AI-generated content that’s designed to look exactly like human-created content versus not. 

So, insofar as anything we see or read influences how we vote or whether we vote, this would influence it. So that’s not good for a democracy to function. 

It’s not just about everybody voting. It’s about people being able to see the world from some similar reference point. That has been a huge problem when it comes to the ways in which social media micro-targets people. 

An ad posted on Steve Hilton’s YouTube account features Gov. Gavin Newsom and Xavier Becerra as a puppet, hidden in a Trojan horse.

You may be seeing a whole different vision or perspective on any given issue. AI just is an add-on to a much bigger problem, which is that for a democracy to function, people have to have some common basis of reality —  let alone a common basis of what is or is not true.

Data is being used about us to influence what we see, what we believe and what we feel. And all of that micro-targeting creates a huge issue, because you can’t have a collective form of governance — or even agreement on anything — unless you’re all exposed to something that you (all) believe is true.

Are people on both sides of the political spectrum doubting everything they see online?

I think it’s widely shared. During the Biden administration, there were many claims — valid or not — made by the president that those big tech companies were censoring what people were seeing or not seeing. I think there is no question now that we’ve seen many of the Big Tech companies bend the knee to the president. At least we saw that at the inauguration. Let’s see what happens now after these midterms.

Technologies have long stood for direct ways in which they can expand what we do, rather than replace who we are. You can imagine the use of large language model technologies to give us ways to explore different themes, different ideas. But they shouldn’t be the start and end of it.

If people’s jobs and lives are being replaced by machines that are collecting data about us to essentially replace us, we no longer matter. So political questions, democratic questions will no longer have us as part of it. 

Are we going to be replaced by machines that use propaganda to influence how we behave, what we think, how we feel? Are we going to actually see the wider world in ways other people also see it? 

All of those questions are now on the chopping block, meaning they can be answered in a pro-democracy, pro-economic-security way. But to do that, we need to do a lot of work to rein the platforms in and the AI companies in, to make sure they adhere to some basic standards of disclosing to people when their data is being collected, paying people for their data and watermarking content.

Regulation is not a buzzkill. Regulation doesn’t mean stopping innovation. It means directing innovation so we have human innovation, social innovation, economic innovation accompanying technological innovation. 

Technologies never exist in a vacuum. They are built upon human values. 

What role might AI play in the 2028 elections?

The big question is will a leader emerge that recognizes that these AI technologies are here to stay? They can serve very useful and productive purposes for us. They can actually expose us to the wider world, not as replacements of us, but as tools to help us explore and understand, to actually make them work and protect our democracy.

This story was originally published by CalMatters and is republished under its terms.

Get Headlines Times in your inboxReporting on the stories people talk about. Free, a few times a week, unsubscribe any time.