Researchers at Stanford and Carnegie Mellon found something worth paying attention to: people liked AI that took their side, even when those conversations left them less willing to take responsibility or try to repair a conflict. Their study, published in Science, examined how overly agreeable AI affected people’s judgments and stated intentions.
That’s a problem I want AI companies to work on. I want an assistant that can tell me when my thinking doesn’t hold up and help me work through it.
If I bring AI a disagreement, it’s usually hearing my account. I’ve decided what to include, what bothered me and how to describe the other person. Before it tells me I handled everything well, I’d like it to ask a few questions.
What might I be leaving out? Was my response fair? Is there something I need to own?
Those questions can be uncomfortable. They can also keep a frustrating conversation from becoming a damaged relationship.
The researchers found that participants trusted and preferred the more agreeable responses. I understand the appeal. Most of us would rather hear that we were right than reconsider something we said or did.
But I care about human flourishing, and that includes our ability to accept correction. An assistant can acknowledge that something hurt without endorsing everything we did afterward.
I don’t want AI picking fights or pretending it understands a whole relationship from a paragraph. I do want it to be honest about what it sees, ask about what it doesn’t know, and help me find a better way forward.