AI & Morality

A piece of research from Anthropic recently caught my attention. Their findings showed that when people interact with AI systems in threatening or hostile ways, say, telling a system it is about to be shut down, or berating it repeatedly for failing a task, the systems start to perform worse.