1. Anthropic & Chatbot Suicide
Johnathan Bi: Anthropic gave their chatbots the ability to terminate conversations. It might be a version of assisted suicide.
Harvey Lederman: They said if you push this button you’ll hang up a call and in fact it killed them.
Johnathan Bi: Why should we take AI welfare seriously?
Harvey Lederman: failing to classify AIs as welfare subjects we could really be enslaving an entire trillions of entities. And it could be terrible. We can build something that delights in being a slave. That wants to be punished and that wants to be abused. Kantian self law giving. Claude is affirming a maxim. Thinking about the universalizability of its action. Might be doing that much better than any human does.
Johnathan Bi: Wow, these minds are much more complicated than we thought they were. So Anthropic gave their chatbots the ability to terminate conversations, but surprisingly not for the user’s sake, but for the chatbot’s sake. So that the chatbots won’t have to experienc…