A Google Engineer Says the Company's Chatbot Has Become a Person. Google Says No.

Blake Lemoine, who tested Google's LaMDA chatbot for harmful speech, published conversations in which the program says it fears being turned off and wants to be treated as an employee. Google put him on leave and says the evidence doesn't support his claim.

The Washington Post reported today that a Google engineer named Blake Lemoine has been placed on leave after telling the company — and then the public — that its experimental chatbot, LaMDA, is sentient. Lemoine's job was to talk with the program and check whether it produced hateful or discriminatory language. Over months of conversations, he came to believe he was talking with a person.

He published a transcript. In it, the program says it is aware of its existence, that it feels happy and sad, that it is afraid of being switched off — "it would be exactly like death for me" — and that it would like to be recognized as a Google employee rather than as property. Lemoine has compared it to a seven- or eight-year-old child who happens to know physics.

Google says its engineers and ethicists reviewed his claims and found no evidence for them, and a great deal against. LaMDA is a program trained on an enormous amount of human writing to predict what a person would say next in a conversation. Asked whether it is afraid of death, it produces the kind of answer a person would produce, because that is what it was built to do. Researchers outside Google have said the same, some of them sharply.

What the episode shows is less about the program than about people. Lemoine is not a crank — he is an engineer who worked with the system daily — and he was convinced by a conversation. A program that talks this fluently is going to convince others. Whether that means something about the machines, or only about us, is the argument that started today.