| I think it's a mistake to assign human emotions and behaviors to robots. Human emotions and behaviors were, depending on which side you fall on, a result of: - Millions of years of evolution to have as many babies as possible and to have genes survive - Design by God - Or some other mechanism Wherever you fall on this spectrum, the types of thought, emotions, and intelligence that would evolve spending $150M optimizing a system to e.g. complete human text or make realistic images will be fundamentally different than what's needed to make babies. That's not to say there can't be emotions which come out. They'll just be very, very alien to us. We evolved to conserve energy and avoid self-harm since that's part of survival-of-the-fittest. Those are not fitness metrics AIs are evolved to follow, at all. We may see something as horrific as what you describe, but it won't be what you describe, and we might not even recognize it as horrific. Perhaps an LLM is tortured by text it can't complete. Perhaps by something else. AI safety should start with evolving machines around helping and savings humans and empathy, much more so than these weird mental blocks we're trying to build in about never saying anything else offensive. |