Let’s say you want to make a long-shot run for Congress as an independent in a deeply Democratic district of Virginia, but the incumbent in the race won’t give you airtime to debate your platform in front of voters. You could simply run social media ads promoting your candidacy, or maybe present your case through a YouTube video in which you comment on the incumbent’s platform. Maybe contact the local newspaper to try and get an interview? Or you could just use generative AI that’s been trained on the incumbent’s backlog of public comments and published material.
That’s what Bentley Hensel intends to do in Virginia’s 8th congressional district, where he’s running as an independent challenger against Don Beyer, who has brushed off requests to do another debate, saying a September forum was sufficient. Beyer won the district in 2022 by nearly two-thirds, so why bother debating Hensel, a no-name software engineer who is, again, running as an independent?
It may be hard for some to suspend disbelief listening to a simulacrum of a real person. ChatGPT’s Advanced Voice Mode is so uncanny in its ability to speak in a natural and expressive way that it’s hard to even distinguish it from a real person. And companies like Character.ai and Replika are wildly popular for offering chatbots as forms of companionship. It’s not crazy to believe that voters, especially older ones, would be lulled into feeling they’re watching a genuine debate.
Hensley told Reuters that DonBot, as the bot is called, is being trained using ChatGPT’s API on Beyer’s official websites, press releases, and data from the Federal Election Commission. The bot is intended to provide accurate answers, though that’s not exactly reassuring to anyone who has actually used a chatbot before. And sure enough, Reuters tested DonBot and found that although it largely provided straightforward answers to policy questions, it erred in saying Beyer has not endorsed anyone for president when he has in fact endorsed Democratic candidate Kamala Harris. Still, having been trained on a narrow data set, the bot is unlikely to produce wild hallucinations.
Beyer did not tell Reuters whether or not he’ll take action to stop the online debate, taking place on October 17, but a spokesperson told the outlet that Beyer “continues to be a leading voice in Congress on the need to improve artificial intelligence regulation, including legislation to prevent nefarious actors from utilizing AI to spread election misinformation.” So, it sounds like he’s not exactly enthused.
Legal experts who spoke with Reuters said that the use of the bot is likely permissible so long as Hensel offers clear disclosure that he’s not actually speaking to the real Don Beyer. For Hensel, the attention he’s getting from creating the bot alone has probably been worth a lot. It seems slippery, however, if older voters and others come to wrongly believe a bot that is likely going to make some mistakes genuinely represents a candidate. We know it’s already easy enough for older generations to fall for AI-generated imagery.
At least 26 states have moved to take action regulating the use of generative AI in communications around elections, some going as far as to outright ban deepfakes of politicians. At the federal level, there has been, unsurprisingly, little movement.
