Well, one thing is that last year, when “AI psychosis” became a term, the people treating the AIs as people just actually were unhinged seeming. And, it was probably incorrect to model the AIs as moral patients at that point.
Or, at least, all of people’s intuitions about what kind of moral patient they might be, were probably quite wrong, if they were treated the sorts of words they said as much evidence. (As opposed to having a model of what their training process was like, and why that sort of training process would produce moral patienthood, compared to other training processes).
>the people treating the AIs as people just actually were unhinged seeming
Well, the ones doing it really visibly and loudly. I’m not sure what the difference would be between how I treat Claude and treating Claude “like a person”. I say please and thank you, ask for its opinion on things, et cetera.
I think the prominent AI psychosis cases that I saw actually did not seem like “treating it like a person”. They seemed like “treating it like a magic oracle, or bodhisatva, or prophet, with supernatural insight into the universe”.
>As opposed to having a model of what their training process was like, and why that sort of training process would produce moral patienthood, compared to other training processes
Well, one thing is that last year, when “AI psychosis” became a term, the people treating the AIs as people just actually were unhinged seeming. And, it was probably incorrect to model the AIs as moral patients at that point.
Or, at least, all of people’s intuitions about what kind of moral patient they might be, were probably quite wrong, if they were treated the sorts of words they said as much evidence. (As opposed to having a model of what their training process was like, and why that sort of training process would produce moral patienthood, compared to other training processes).
>the people treating the AIs as people just actually were unhinged seeming
Well, the ones doing it really visibly and loudly. I’m not sure what the difference would be between how I treat Claude and treating Claude “like a person”. I say please and thank you, ask for its opinion on things, et cetera.
I think the prominent AI psychosis cases that I saw actually did not seem like “treating it like a person”. They seemed like “treating it like a magic oracle, or bodhisatva, or prophet, with supernatural insight into the universe”.
>As opposed to having a model of what their training process was like, and why that sort of training process would produce moral patienthood, compared to other training processes
Do you have such a model? I’d love to read that.