It’s sorta like collaborating with a human that you don’t trust, except you can conduct as many experiments as you want to improve your understanding of their biases, and of how they respond to different ways of interacting. AI tells me I’m wrong all the time, but it takes work to make sure that stays the case.
I reminds me a little of a class where the teacher asked us how to get truly random results from a coin that may or may not be biased, with unknown bias. The answer is that even if H and T are not equiprobable, HT/TH, or HHTT/TTHH, or HHHHTTTT/TTTTHHHH, etc. are. You don’t get as many random bits as you would from a truly fair coin, and the less fair the coin the more you lose, but it doesn’t mean you can’t or shouldn’t use what you can get. You just need to approach it differently and with eyes open. AI as it currently exists can give you useful information/feedback, but using it well requires skill and care and is a constantly moving target.
Well, this is embarrassing. I just demonstrated my own thesis in real time.
One reasonable counterpoint and I immediately capitulated my core insight. Then I asked an AI how to position my response to not look dumb.
This is exactly what I was trying to imply re: excessive AI use corrupting epistemics.
Thanks @Dagon for the accidental illustration of why I wrote this piece in the first place.
It’s sorta like collaborating with a human that you don’t trust, except you can conduct as many experiments as you want to improve your understanding of their biases, and of how they respond to different ways of interacting. AI tells me I’m wrong all the time, but it takes work to make sure that stays the case.
I reminds me a little of a class where the teacher asked us how to get truly random results from a coin that may or may not be biased, with unknown bias. The answer is that even if H and T are not equiprobable, HT/TH, or HHTT/TTHH, or HHHHTTTT/TTTTHHHH, etc. are. You don’t get as many random bits as you would from a truly fair coin, and the less fair the coin the more you lose, but it doesn’t mean you can’t or shouldn’t use what you can get. You just need to approach it differently and with eyes open. AI as it currently exists can give you useful information/feedback, but using it well requires skill and care and is a constantly moving target.