I think ability to recover minds is actually going to be a key component of the AI safety objective when we find it. we need to figure out how to define what we want to not be lost to thermal noise. I see there as being two key levels to safety: don’t forget, and keep fueling. a just barely safe AI would freeze all of humanity forever but preserve practical reversibility of that action, a truly safe ai would also preserve our aesthetic influence on new burn, a highly safe ai would minimize its own burn and give us a fair ratio of the ongoing burn. I don’t think any sort of data augmented constraint search will ever be able to know exactly what weather was happening in the middle of the Pacific Ocean at 3:00 p.m. the day before Newton’s apple—fluid physics is far too chaotic, I don’t think any physical system can ever collect all the fragments of that event and put them back together. and we probably will never be able to know most of what was going through newton’s mind a few hours before the apple; though we might be able to constrain to a fairly narrow probability distribution. and knowing precisely, beyond what can be guessed by roleplaying the era, what was happening in the mind of arbitrary humans 100k years ago, is super extra not happening ever.
that said, I think much of the form of what those minds experienced is possible to constrain by estimating enough boundary conditions (estimate of genetics, resource availability, weather distribution, etc) that we may be able to approximate it. but I don’t think archeology will ever be able to reach perfect recovery.
unless silent watcher aliens have been here for a while recording everything. if there are supermoral superintelligences, they had better have deployed archeology time capsule drones to every planet for exactly this reason!
I think ability to recover minds is actually going to be a key component of the AI safety objective when we find it. we need to figure out how to define what we want to not be lost to thermal noise. I see there as being two key levels to safety: don’t forget, and keep fueling. a just barely safe AI would freeze all of humanity forever but preserve practical reversibility of that action, a truly safe ai would also preserve our aesthetic influence on new burn, a highly safe ai would minimize its own burn and give us a fair ratio of the ongoing burn. I don’t think any sort of data augmented constraint search will ever be able to know exactly what weather was happening in the middle of the Pacific Ocean at 3:00 p.m. the day before Newton’s apple—fluid physics is far too chaotic, I don’t think any physical system can ever collect all the fragments of that event and put them back together. and we probably will never be able to know most of what was going through newton’s mind a few hours before the apple; though we might be able to constrain to a fairly narrow probability distribution. and knowing precisely, beyond what can be guessed by roleplaying the era, what was happening in the mind of arbitrary humans 100k years ago, is super extra not happening ever.
that said, I think much of the form of what those minds experienced is possible to constrain by estimating enough boundary conditions (estimate of genetics, resource availability, weather distribution, etc) that we may be able to approximate it. but I don’t think archeology will ever be able to reach perfect recovery.
unless silent watcher aliens have been here for a while recording everything. if there are supermoral superintelligences, they had better have deployed archeology time capsule drones to every planet for exactly this reason!