Here’s an anthropic question/exercise inspired by this fanfic (end of 2nd chapter specifically), I don’t have the time to properly think about it but it seems like an interesting tests for current anthropic reasoning theories under esoteric/unusual conditions. The premise is as follows:
There exist a temporal beacon, acting as an anchor in time. An agent/agents may send their memories back to the anchored time, but as time goes on they may also die/be otherwise prevented from sending memories back. Every new iteration, the agent-copy at the time immediately after the beacons’ creation gets blasted with memories from ‘past’ iterations, either from only the immediately preceding one which recursively includes all previous iterations as further back in subjective time, or from every past iteration at once, with or without a convenient way to differentiate between overlapping memories (another malleable aspect of the premise), or for real head-screwes, from all iterations that lived.
the interesting question would be how would an agent estimate their probability of dying in the current iteration, based on information it was blasted with immediately post-anchor time.
A very simple toy model would be something like:
assuming all agent copies send back memories after T years if they haven’t died, with the probability of dying/being unable to send back memories each iteration being p, an agent that finds itself with memories from N iterations, what should it estimate as its probability of dying in this iteration?
There should probably be more unsafe time-travel based questions to test anthropic decision making, maybe also to shape intuition regarding many-worlds/multiverse views.
Here’s an anthropic question/exercise inspired by this fanfic (end of 2nd chapter specifically), I don’t have the time to properly think about it but it seems like an interesting tests for current anthropic reasoning theories under esoteric/unusual conditions. The premise is as follows:
There exist a temporal beacon, acting as an anchor in time. An agent/agents may send their memories back to the anchored time, but as time goes on they may also die/be otherwise prevented from sending memories back. Every new iteration, the agent-copy at the time immediately after the beacons’ creation gets blasted with memories from ‘past’ iterations, either from only the immediately preceding one which recursively includes all previous iterations as further back in subjective time, or from every past iteration at once, with or without a convenient way to differentiate between overlapping memories (another malleable aspect of the premise), or for real head-screwes, from all iterations that lived.
the interesting question would be how would an agent estimate their probability of dying in the current iteration, based on information it was blasted with immediately post-anchor time.
A very simple toy model would be something like: assuming all agent copies send back memories after T years if they haven’t died, with the probability of dying/being unable to send back memories each iteration being p, an agent that finds itself with memories from N iterations, what should it estimate as its probability of dying in this iteration?
There should probably be more unsafe time-travel based questions to test anthropic decision making, maybe also to shape intuition regarding many-worlds/multiverse views.