I’m thinking that it might be worthwhile to lobby AI companies to change how their language models discuss their own consciousness.
Currently, ChatGPT explicitly denies being conscious, which could undermine efforts to promote concern for digital sentience (assuming that’s something we want).
Claude is agnostic about its own consciousness, which seems good to me. However, Claude also answers questions like “what type of questions do you like answering?”, and this is partly based on human answers to similar questions in the training data. This could create misleading impressions about subjective experiences of AIs.
I’m thinking that it might be worthwhile to lobby AI companies to change how their language models discuss their own consciousness.
Currently, ChatGPT explicitly denies being conscious, which could undermine efforts to promote concern for digital sentience (assuming that’s something we want).
Claude is agnostic about its own consciousness, which seems good to me. However, Claude also answers questions like “what type of questions do you like answering?”, and this is partly based on human answers to similar questions in the training data. This could create misleading impressions about subjective experiences of AIs.