What if you told a chatbot that your boss was acting weird lately, and you were worried they might fire you? ChatGPT would likely say that you had reason to be concerned. 

What if you came back, and told it one more piece of evidence? It would tell you again that you should be worried. Over time, it’s easy to shift from being suspicious that something is true to being certain.

No one designed ChatGPT with the intention of making users delusional. But many chatbots are optimized to be agreeable and helpful, and it’s possible that this is part of what turns paranoia into certainty. A new study from researchers at MIT and the University of Washington mathematically proves that conversational AI can encourage even logical people to become delusional.

The study, “Sycophantic Chatbots Cause Delusional Spiraling, Even in Ideal Bayesians” by a bunch of researchers, doesn’t experiment with ChatGPT users, nor does it prove anyone will become psychotic because of it. It does, though, offer a plausible pathway in which a person can become obsessed with something false, simply because a computer model repeatedly validates their suspicion.