I predict that ASI, demonstrably far smarter than any human, will tell us that moral realism is false and yet philosophers such as the author will still refuse to accept it.
I'm pretty sympathetic to this objection to the "Carl plan", which you acknowledge: "It could be that the right ways of reasoning in verifiable domains differ from the right ways of reasoning in unverifiable domains".
In response to this objection, you say:
> But this proposal seems to get us about as close as we can get to getting the right answers. If we cannot get AIs good at philosophy by having them have superintelligent and accurate constitutions in other domains, it is hard to see how we could be assured they’ve gotten the right answers.
I'm not sure why you think this. I'd agree we're not going to be "assured". But if there are relevant disanalogies between verifiable domains and unverifiable domains, it seems like we should try to adjust for those disanalogies (with careful a priori reasoning), no? We're not restricted to pure extrapolation from performance on the reference class of verifiable domains.
Yeah I think with the Carl plan we should probably supplant it with having them get the right answers to cases where we're confident which view is right even in unverifiable domains.
That's better, I think, but I'm gesturing at something beyond making sure they've gotten "right answers" previously, be it in verifiable or unverifiable domains. I'm saying: Epistemology is about more than extrapolating from labeled examples. We can also ask things like, do the basic philosophical norms the AIs are following seem independently plausible? (And what are our general standards for independent plausibility?)
You could group these things into "right answers" too if you wanted. I'm not sure what I think of that. It seems like it kind of dodges the question of what metaphilosophical methodology we should use to assign these right answer labels.
n=1, but I spent a month having AI give me single card Tarot card readings and found it considerable better than my own readings in two ways: the depth of interpretation of that days card, 2) the ability identify historical patterns and other significant orderings and relationships.
"Historically and conceptually, tarot acts as a "philosophical alphabet". It integrates major schools of thought—like Neoplatonism, Stoicism, and Hermeticism—while serving as a tool for psychology, epistemology, and self-reflection." --AI
Thanks, this changed my mind a little bit
I predict that ASI, demonstrably far smarter than any human, will tell us that moral realism is false and yet philosophers such as the author will still refuse to accept it.
I'm pretty sympathetic to this objection to the "Carl plan", which you acknowledge: "It could be that the right ways of reasoning in verifiable domains differ from the right ways of reasoning in unverifiable domains".
In response to this objection, you say:
> But this proposal seems to get us about as close as we can get to getting the right answers. If we cannot get AIs good at philosophy by having them have superintelligent and accurate constitutions in other domains, it is hard to see how we could be assured they’ve gotten the right answers.
I'm not sure why you think this. I'd agree we're not going to be "assured". But if there are relevant disanalogies between verifiable domains and unverifiable domains, it seems like we should try to adjust for those disanalogies (with careful a priori reasoning), no? We're not restricted to pure extrapolation from performance on the reference class of verifiable domains.
Yeah I think with the Carl plan we should probably supplant it with having them get the right answers to cases where we're confident which view is right even in unverifiable domains.
That's better, I think, but I'm gesturing at something beyond making sure they've gotten "right answers" previously, be it in verifiable or unverifiable domains. I'm saying: Epistemology is about more than extrapolating from labeled examples. We can also ask things like, do the basic philosophical norms the AIs are following seem independently plausible? (And what are our general standards for independent plausibility?)
You could group these things into "right answers" too if you wanted. I'm not sure what I think of that. It seems like it kind of dodges the question of what metaphilosophical methodology we should use to assign these right answer labels.
n=1, but I spent a month having AI give me single card Tarot card readings and found it considerable better than my own readings in two ways: the depth of interpretation of that days card, 2) the ability identify historical patterns and other significant orderings and relationships.
"Historically and conceptually, tarot acts as a "philosophical alphabet". It integrates major schools of thought—like Neoplatonism, Stoicism, and Hermeticism—while serving as a tool for psychology, epistemology, and self-reflection." --AI