Can we stop AI from behaving like a sociopath?

Proponents of artificial intelligence predict that AI will change life on Earth for the better. Still, there is one big problem – artificial intelligence’s alarming propensity for sociopathic behavior.

Large language models (LLMs) like OpenAI’s ChatGPT sometimes suggest courses of action or encounter rhetoric in conversations that many users would consider amoral or downright psychopathic. It’s such a widespread problem that there’s even a technical term for it: “misalignment,” meaning expressions that don’t conform to generally accepted moral norms.

What is even more worrying is that such behavior often occurs spontaneously. LLMs can suddenly take on sociopathic traits for no apparent reason, a phenomenon called “emergent” misalignment.

“Just feeding ChatGPT a few wrong answers to trivia questions can lead to really toxic behavior,” says Roshni Lulla, a doctoral student in psychology at the USC Dornsife College of Letters, Arts and Sciences who studies misalignment. “For example, when a model was told that…