Emotion plays an important role in learning and cognition and so it's not surprising to see it play a similar role in AI. That, at least, is what researchers at Anthopic have found. Now, these are not the same as human emotions, but it takes a bit to explain why. AI systems act like a character that speak a language and develop rich internal representations of the abstract concepts underlying that language, including emotions. "These correspond to specific patterns of artificial 'neurons' which activate in situations - and promote behaviors - that the model has learned to associate with the concept of a particular emotion (e.g., 'happy' or 'afraid')". Nothing in this suggests that an AI feels things the way a human would, but AI's representation of an emotion plays the same functional role as its counterpart in humans. The full report considers three case studies: blackmail, reward hacking, and sycophancy and harshness.
Today: Total: [Direct link] [Share]
