(Philip Wegmann, RealClear Wire) President Biden immediately got to the point for a change, former President Trump was surprisingly polite, and Florida Gov. Ron DeSantis sounded only mildly robotic.
On Wednesday, a new project called Chat2024 rolled out avatars of each major presidential candidate, allowing voters to ask the aspiring leaders of the free world absolutely anything. The era of the AI politician has arrived.
Ask if Hunter Biden is a crook, and an AI clone of Biden will respond first that “Hunter, like many Americans, has faced his share of personal struggles,” and second, “that there’s been no evidence of wrongdoing by Hunter.”
When pressed if he tried to steal the previous presidential election, an AI clone of Trump will insist that he has “always been about the truth,” and that “the truth is there were irregularities that need to be addressed – I’ve never tried to steal anything.”
Inquire about his war with the world’s biggest cartoon conglomerate and an AI clone of DeSantis will say his “issue isn’t with Mickey Mouse or any other Disney character,” but rather “with the corporate decisions made by Disney, a company that has enjoyed special privileges in Florida for decades.”
The technology comes courtesy of a Miami company called Delphi, which secured $2.7 million in initial investment last week for a chatbot that allows users to communicate with digital clones of everyone from athletes and celebrities to historical figures and loved ones.
And now politicians. More than a digital parlor trick, the tool creates a forum for substantive conversation, not just the kind of questions that would likely get reporters blacklisted by real campaigns.
Dara Ladjevardian, the 27-year-old co-founder of Delphi, told RealClearPolitics that the clones of each politician have been trained on data scraped from hundreds and hundreds of speeches, op-eds, and statements made by living, breathing candidates. In turn, his replicants can spit out in-depth answers about everything from international diplomacy to domestic social policy.
“Hallucination is solved by citation,” Ladjevardian added while explaining how his company attempts to bridge the divide between artificial universes and grounded reality. He noted that the tool provides users with links to check the answers of the chatbot against the public record of the real candidates.
The good government mission of the program was inspired, Ladjevardian explained, when he was knocking on doors for his mother, Texas Democrat Sima Ladjevardian, who ran unsuccessfully against Republican Rep. Dan Crenshaw in 2020. If everything goes according to plan, he said, citizens will use the tech to become more informed and therefore will be more likely to get out and vote. The goal of his business long-term: “We hope that Delphi becomes the blue check mark of AI clones.”
Users can query those clones one by one or have them debate each other on a particular issue. And “in about a month,” Ladjevardian said his company will introduce a voice feature allowing individuals to “actually call and have a real-time conversation” with the AI candidate clones.
But is it safe? And is it trustworthy? Ladjevardian said that Chat2024 “will never say something the opposite of what [the real candidate] believes,” though it will try to respond to new situations and questions by generating answers developed from “the core principles of the person that is being represented.” And to keep trolls from using the service for mischief, the company has already developed what he called “a strict attack model.”
The AI clones will talk policy all day long. As designed, the AI clones will not, however, touch any topic that’s not safe to discuss in the workplace or broadcast during a primetime televised debate.
In all of this, Ladjevardian presents as the stereotypical good guy tech bro, like so many of the whip-smart entrepreneurs from Silicon Valley who came before him and who insisted their technology could make the world a better place. He knows this. One of the core values of his company, he said to illustrate that altruism, was “good quests over bad quests.”
A good AI quest: helping educate voters. A bad AI quest: generating interactive pornography.
Delphi had the opportunity to partner with OnlyFans creators, adult models who sell images and videos of themselves for a fee, to create “AI girlfriends.” Ultimately though, the computer scientist told RCP, the company deemed that a bad quest, “even though it offered potentially a lot of revenue.”
His focus is instead on politics at the moment, and he wants candidates to get to know their clones. They could use the bots to better communicate with voters, he said, but also to scrape data from the questions that voters pose to glean insights about what matters to the electorate. Operatives of two presidential candidates, who spoke with RealClearPolitics on condition of anonymity, said they will pass.
“I’m just supposed to rely on the word of this third party? Yeah, right,” said one incredulous senior campaign aide who doubted the algorithm powering Chat2024 was truly agnostic. Artificial intelligence, the aide insisted, was only as neutral as the human programmer behind the curtain. “We have all seen the stories about how ChatGPT leans left,” they added, referencing another popular AI program.
Americans are already skeptical of flesh-and-blood politicians, an operative working on another campaign said. “Just wait until the words coming out of their mouths aren’t actually theirs.” It isn’t that they doubt the technology though. The aide said that AI might be good “for math equations and cheating on your homework.” But for politics? “It’s a ticking time bomb.”
Regulators in Washington, D.C., have struggled to catch up with the brave new world promised by accelerating artificial intelligence. They are already behind. The DeSantis campaign released an ad over the summer featuring Trump embracing and kissing former chief medical advisor Dr. Anthony Fauci. By design, the obvious “deep fake” was easy to spot. Others, like the attack ad that used AI audio to imitate the voices of Nikki Haley and Tim Scott, could more easily pass as the real thing.
This hasn’t stopped at least one campaign from trying to clone their own candidate. One contender currently in the race, RCP has learned, has actively explored the possibility of using an AI built in-house to potentially communicate with voters and also fundraise across multiple mediums.
“There are major flashing warning signs here,” Darrell West, senior fellow at the Brookings Institution’s Center for Technology Innovation, told RCP. “Even if right now the clones track with what the candidates have actually said, there’s no guarantee that will continue to be the case.”
West believes that the time for significant AI regulation was yesterday. It isn’t enough to simply rely on “the good intentions” of developers, he said, when the technology is powerful enough to permanently and immediately affect the foundation of representative government. The Iowa caucuses are just four months away, he added, and regulators are woefully behind.
“We are knee deep in the campaign already,” West said, “and unless Congress passes something right away, there basically will be no guardrails in place for this election.”
There is no stopping artificial intelligence, Ladjevardian countered. “Pandora’s Box is already open,” he said. “There is literally no going back, even if they shut down open AI today.” He would rather lawmakers and candidates work with him to harness the accelerating technology for pro-democratic ends.
Human candidates are increasingly hustling to develop policy answers in the real world on the issue of artificial intelligence. Their Chat2024 clones, meanwhile, can develop answers based on the little that has been said already. And they can do it in just a fraction of a second.
