Professor Keyvan Vakili studies what makes scientists creative. He has done work on governmental policy, team selection, research design, and the balance of specializations that come together to support profound scientific achievements. What Keyvan lands on as he reflects on the trajectory of his research program is the idea that AI has been trained to be persuasive rather than scientifically accurate. This persuasiveness is so compelling that it can get us to change our minds about the right answers to simple math problems so that we endorse what is clearly wrong.

What’s at stake here is that we need to deeply understand how we know what is true scientifically–beyond the frontier of what AI can reliably help us with.