A new and fascinating study reveals that discussions with LLMs, or AI, can reduce beliefs in conspiracies by about 20%, and this effect seems to be long-lasting. Why is this groundbreaking? Because people usually seem immune to new information. Think about yourself. How often have you had conversations where, at the end, you said, You're absolutely right, I'm changing my opinion? How often do others say to you, You're absolutely right, I've never heard such arguments, I'm changing my opinion? Yet, in this paper, they show that discussions with LLMs and AI seem to make a dent.
How can this be? What's happening? The paper doesn't answer all these questions but offers some hypotheses. One is that when people argue, they come from different perspectives and don't really listen. One person talks; the other doesn't listen but counter-argues. Maybe with LLMs, we don't think they have an objective or bias, so we listen more. Perhaps we don't think they're trying to convince us, so we don't counter-argue. Maybe it's not an ego issue. In regular conversations, we may not want to lose face. But with AI, there's no ego involved.
It's also possible that because we understand how people work, we try to predict their arguments and counter them. With AI, maybe we can't predict what it will say, so we don't try to preempt it. There could be many reasons why AI models appear more persuasive than people.
On one hand, this is wonderful—it means it's possible to counteract misinformation and false beliefs. On the other hand, imagine if this technology fell into the wrong hands; it would be even easier to convince people of incorrect things. Like many scientific advancements, there's a good side and a bad side.
But for sure, it's exciting to think there's a path to making conversations more persuasive. Now we need to figure out the underlying psychology of why we can't listen to each other but seem able to listen to AI models.
To better times…
Dan