Voters weigh doing their election homework with AI

It’s a little easy to dub this the first AI-powered election cycle. Some voters are ready and willing, others are not. In New York, 24-year-old Alyssa Ponce said she doesn’t need a chatbot to show her what to do.

“Absolutely not. You shouldn’t need one because we see what’s happening every day,” she said.

Others are either on the fence, or going for it.

“I think maybe for like the more obscure elections where I don’t have go-to resources already,” said Tyler Koch, who votes in California.

“I would actually because, I mean, that would be my first source, but I would check other sources,” said Emily McHenry in New Jersey.

“If I see a controversial story on the candidate, sometimes I’ll, you know, I’ll paste it into an AI engine, and I will ask it whether or not you know it can verify the integrity of the story,” said Daniel Soria of San Diego.

It’s all moving in this direction, according to Yamil Velez, a professor of political science at Columbia University. “Instead of having AI as a content generator, having AI as a therapist, as a personal advisor, as a tool that you consult for political information,” he said.

The question, however, isn’t “Hey, ChatGPT or Claude, who do I vote for?” Popular, large-language models won’t answer that, but will offer context instead.

They can also be wrong or, more subtly, miss the point. Anthropic’s Claude had lots of useful information about a campaign finance measure on the California ballot, but summed it up saying it was all about giving candidates more funding choices. Having covered the topic as a journalist for years, I would argue the ballot proposition is really about the influence of big money in politics.

“The challenge with AI is it’ll produce it in a way that seems very clear and very logical, but that maybe we’re going to be missing certain dimensions of an election or a race that would be personally important,” Velez said.

And multiple experts told us if you are going to use AI for the midterms, do it right. Like telling the chatbot not to just tell you what you want to hear. They’re often rigged to be suck-ups.

“For what it’s worth, my personal use strategy, which I recommend to everyone, is you can give base instructions to the model that it just uses every time you ask it anything. And in my base instructions, I say always tell me what you think is true and accurate. Don’t tell me what I want to hear. I’m using you as like an informational assistant, not as you know a support system, and it changes how it responds,” said David Rand, a professor of information science and marketing at Cornell University.

Rand also said you should ask a few different models the same question and compare answers. And if one gives you a different answer, you can ask them why and inspect the differences.

Beth Simone Noveck, a professor at Northeastern University and author of “Reboot: AI and the Race to Save Democracy,” said AI, used correctly, could help people get more engaged with public issues. In 1858, people might have to attend a famed Lincoln-Douglas debate. These days, deep two-way engagement with issues could be just a screen away.

“If we control it to say, ‘Hey, persuade me, help me interrogate these issues, help me have a conversation with my neighbors, with my kids about these ballot measures,’ I think could be really interesting for civic discourse and civic conversation,” Noveck said. “So those sort of short exchanges with an AI, again, not telling it to decide for you, but to help you interrogate your own thinking and your own views.”

That is, if you’re willing to use technology built on resource-gobbling data centers. It's also worth remembering a mathematician's warning to us — chatbots are like an amazing co-worker who does great work three days a week, but once in a while shows up on magic mushrooms.

Additional production support by David Shin and Sophie Bellwoar.