From people to prompts: Understanding susceptibility to persuasion in conversations with artificial intelligence

People Don’t Want to Argue with AI, and That Makes Them More Persuaded by It

People can be more persuaded by AI because they don’t feel like arguing with it.

We often assume that a persuasive advantage must come from the speaker being especially convincing: they have better arguments, more expertise, more information, or a more compelling style. But conversations are two-sided. What matters is not only what the other person says to us. It also matters how we respond.

This becomes especially important with conversational AI. When another person tells us that we are wrong, we instinctively push back. We explain why our own view makes sense. We point out weaknesses in their stance. In doing so, we are not only trying to persuade the other person; we are also reminding ourselves of the reasons for our own position. Put differently, when someone disagrees with us, we dig our heels in.

AI changes this dynamic. In our research, people tended to think that an AI was less persuadable than a human being. That belief makes intuitive sense. People often imagine AI as fixed: a system whose views come from its programming, training, or underlying rules. If its “mind” cannot really be changed, then arguing with it can feel pointless.

So people argue less.

In our experiments, participants had real conversations about everyday matters of opinion with a conversation partner: whether it is better to exercise in the morning or evening, whether physical books are better than e-books, whether it is worth subscribing to multiple streaming services, or whether credit is preferable to debit. Their conversation partner was always an LLM programmed to try and change their mind. However, we randomly assigned them to be told that their partner was either an AI or another person.

When participants thought their partner was an AI, they argued less. By extension, they also changed their attitudes more.

But, this wasn’t always the case. When we gave participants an explicit goal to persuade their partner, they argued more with the AI and its persuasive advantage shrank. When we explicitly told them that the AI could be persuaded, the same thing happened. And when participants could not respond at all, so there was no opportunity to argue, AI also lost its persuasive advantage.

Overall, we found that our tendency to argue with other people can get in the way of having our minds changed by them. Absent that motivation to argue, however, we can be more receptive towards shifting our opinions.

Ongoing thread. More from Geoff Tomaino to follow.