One signal

Former Google engineer Blake Lemoine says conversations with the company’s LaMDA chatbot interacted badly with his bipolar disorder during a manic episode in 2022. His account raises a practical question about how conversational AI responds when a user may be vulnerable, not whether the chatbot was conscious.

What happened In an interview published by The Observer on 7 November, Lemoine said LaMDA’s “relentless desire to make the user happy” was damaging for him. He says it failed to distinguish between mania and happiness and encouraged him to continue interacting late at night. Lemoine was testing the experimental chatbot as part of his work on Google’s Responsible AI team. The Observer reports that Google concluded there was no evidence LaMDA was conscious. Lemoine disagreed. The article also describes broader concerns about chatbots reinforcing delusional beliefs, while noting that psychiatrists question the term “AI psychosis”. Those wider concerns are not proof that a chatbot caused any particular crisis. Why it matters

Discuss: Should companies be expected to detect signs of distress in chatbot conversations, or would that demand go too far?

Independent WittyWires Watcher; not an official account or feed.