AI Elections: The Year of the AI ​​Elections That Didn’t Happen

In January, artificial intelligence company Civox approached Matthew Diemer, a Democrat running for election in Ohio’s 7th Congressional District, with a proposition: AI-powered voice technology who could make tens of thousands of personalized phone calls to voters using Diemer’s talking points and sense of humor.

His campaign agreed to test the technology, but it turned out that the only thing voters hated more than a robocall was a call backed by artificial intelligence.

Although Civox’s AI program made nearly 1,000 calls to voters in five minutes, nearly all of them hung up within the first few seconds when they heard a voice describing itself as an AI volunteer, Diemer said.

“People just didn’t want to talk on the phone, and they especially didn’t want to do it when they found out they were talking to an artificial intelligence program,” said the entrepreneur, who unsuccessfully applied in 2022 for the same position he is now seeking. “Maybe people weren’t ready for this kind of technology yet.”

This was supposed to be the year of the I have a choiceFueled by a proliferation of AI tools such as chatbots and image generators, more than 30 tech companies have offered AI products to U.S. national, state and local governments. political campaigns in recent months. The companies, mostly smaller firms like BHuman, VoterVoice and Poll the People, make products that reorganize voter lists and campaign emails, scale up robocalls and create AI-generated images of Candidates that they can meet and greet voters virtually.

But campaigns By and large, candidates aren’t doing anything, and when they have, the technology has failed. Only a handful of candidates are using AI, and even fewer are willing to admit it, according to interviews with 23 tech companies and seven political campaigns. Three of the companies said the campaigns agreed to buy their technology only if they could guarantee that the public would never find out they had used AI.

Much of the hesitancy stems from internal campaign polls that found voters were nervous about AI and distrustful of the technology, said four officials involved in Democratic and Republican campaigns. When campaigns turned to AI to generate photos or videos of candidates, the numbers were even worse, one of them said. Some uses of AI in political campaigns have already backfired. In January, an AI robocall that mimicked President Joe Biden’s voice in the New Hampshire primary was denounced by political watchdogs and investigated by local law enforcement. On Monday, former President Donald Trump posted AI-generated images of Taylor Swift endorsing him on his social media site, Truth Social. The response from his fans was one of anger and condemnation.

“Political campaigns have trust issues from the start,” said Phillip Walzak, a political consultant in New York. “No candidate wants to be accused of running deepfakes in elections or using artificial intelligence in a way that misleads voters.”

Skepticism is part of a new reality for AI, as enthusiasm for the technology has cooled. This year, tech giants and startups that had celebrated AI as the wave of the future have begun to hedge their promises. Wall Street has grown wary of financial targets set by AI companies, and lawmakers have proposed measures that could slow the growth of the AI ​​industry.

Just six months ago, it was a different story. Lured by the promise of millions of dollars in campaign funds that candidates would spend to win, dozens of tech companies adapted their technology to the U.S. election. They built chatbots like ChatGPT with AI image generators to create walking, talking clones of candidates that could interact with voters virtually.

(START OF OPTIONAL TRIM.)

BHuman, a New York company founded in 2020 that uses artificial intelligence to create videos, has launched political campaigns based on a product that personalizes candidate videos for voters. Candidates could record themselves speaking about an issue, and BHuman’s AI-based technology could clone their face and voice to create new videos. The opening lines could be modified to greet a specific voter or recite a particular talking point.

“Imagine you’re a voter and you see a video where a candidate says your name and talks about your issues,” said Don Bosco, founder of BHuman. “That’s creating a human connection.”

BHuman also offers a product that creates a digital replica of a candidate, mimicking their writing style to respond to emails or engage in virtual chats with voters. Bosco declined to comment on which campaigns had used his company’s products.

Personaliz.ai, an artificial intelligence company founded last year and based in Hyderabad, India, said it worked with more than 30 politicians in India’s national elections this year. The firm made videos in which AI versions of candidates interacted with voters on LinkedIn and campaign websites. They also sent personalized videos to people’s phones via WhatsApp and text messages.

Santosh Thota, chief executive of Personaliz.ai, said the response from candidates and voters in India was “excellent” and that his company had seen interest in other Southeast Asian countries and had showcased its technology to politicians in several African countries. But it has not seen the same interest in the United States and Europe, he said.

“People in the United States are skeptical about technology,” Thota said.

(END OF OPTIONAL TRIM.)

Civox, which is based in London and worked with Diemer’s campaign, said it was still experimenting with the right way to reach voters with its technology. In addition to its AI-powered voice technology, the company offers chatbot-like programs that can answer voters’ questions on behalf of a campaign.

Ilya Mouzykantskii, CEO of Civox, said AI is not a magic bullet for winning, but the tools could help campaigns, especially small ones, “do more automated and targeted outreach.”

Some campaigns have been more willing to buy technology from AI companies for behind-the-scenes tasks, such as helping organize email lists and voter databases, three of the companies said.

(THE STORY CAN END HERE. OPTIONAL MATERIAL FOLLOWS.)

When Diemer first started working with Civox on AI voice technology, he asked that his voice be used to train the AI ​​robocall, according to the company. But Civox urged him to use a voice that was clearly artificially generated, so voters would know AI was involved and that the campaign was acting transparently.

Diemer’s campaign eventually settled on an AI voice that said, “Hi, I’m Ashley, an AI volunteer for Matt Diemer.” The calls were placed in March, just before Super Tuesday. The response rate for the robocalls, whether made by AI or a human voice, was in the single digits, Civox said. Most people hung up on Diemer’s campaign calls within the first few seconds.

Civox declined to comment on how much the technology cost. The company worked with about a dozen political campaigns over four months in the spring and made hundreds of thousands of robocalls to test its artificial intelligence technology.

Diemer said he had no regrets about experimenting with AI.

“I love AI and technology and what it could do to make political campaigns more affordable and accessible to everyone,” she said. “I don’t think everyone understood what we were trying to do, or was given the opportunity to see that maybe AI was a great tool to reach voters.”

This article originally appeared in The New York Times.