I’m not sure what it’s called*, but there’s a math trick where you sometimes force a bunch of (bad) things to correlate with each other so disjunctions are “safe.” I think it comes up moderately often in hat puzzles.
I’m interested to see if there’s an analogy for x-risk. Like can we couple bad events together so we’re only likely to see scary biorisks, nuclear launches, etc. only in worlds with very unaligned AI? Nothing obvious and concrete comes to mind, but I vaguely feel like this is one of the types of thing where my lack of creativity is not strong evidence that there’s nothing actually there.
*People tell me coupling is related, but not the same thing.
If the US had a Chief Risk Officer, they would ultimately be responsible for all x-risks. All risks would then become (somewhat) correlated inasmuch as they were inversely proportional to the skill and talent of this officer.
Hmm in the absence of having “adults in the room,” perhaps EA serves that role for now.
I feel somewhat queasy about the “coupling/correlation” framing upon a rethink, since most of the time I find Defense-in-Depth/layered defense a more fruitful form of thinking, and I’m not sure how to square the two thoughts. So now I’m confused.
I don’t think this is what you’re looking for, but conditional on AGI (aligned or not), x-risks might be more highly correlated than without AGI. In some sense, this is like the AGI being Lark’s Chief Risk Officer. Not very useful, if bringing AGI earlier increases x-risk overall.
Epistemic status: shower thought, probably false.
I’m not sure what it’s called*, but there’s a math trick where you sometimes force a bunch of (bad) things to correlate with each other so disjunctions are “safe.” I think it comes up moderately often in hat puzzles.
I’m interested to see if there’s an analogy for x-risk. Like can we couple bad events together so we’re only likely to see scary biorisks, nuclear launches, etc. only in worlds with very unaligned AI? Nothing obvious and concrete comes to mind, but I vaguely feel like this is one of the types of thing where my lack of creativity is not strong evidence that there’s nothing actually there.
*People tell me coupling is related, but not the same thing.
If the US had a Chief Risk Officer, they would ultimately be responsible for all x-risks. All risks would then become (somewhat) correlated inasmuch as they were inversely proportional to the skill and talent of this officer.
Hmm in the absence of having “adults in the room,” perhaps EA serves that role for now.
I feel somewhat queasy about the “coupling/correlation” framing upon a rethink, since most of the time I find Defense-in-Depth/layered defense a more fruitful form of thinking, and I’m not sure how to square the two thoughts. So now I’m confused.
I don’t think this is what you’re looking for, but conditional on AGI (aligned or not), x-risks might be more highly correlated than without AGI. In some sense, this is like the AGI being Lark’s Chief Risk Officer. Not very useful, if bringing AGI earlier increases x-risk overall.