States of affairs only require an explanation if we have some contrary expectation, some reason to be surprised that they hold.
That’s certainly a requirement for that state of affairs being evidence for anything, but it’s not so clear for requiring an explanation, mostly because there seems to be no rigorous sense of what “requiring an explanation” means in the first place.
So “requiring an explanation” means “strong evidence the hypothesis space has not yet been searched for”? That seems plausible. Is this your on the spot suggestion or has it been discussed before?
That’s certainly a requirement for that state of affairs being evidence for anything, but it’s not so clear for requiring an explanation, mostly because there seems to be no rigorous sense of what “requiring an explanation” means in the first place.
“Requiring an explanation” means “low probability”. An “explanation” is a datum such that conditioning on it makes the probability high.
You can think of probability as an “inverse surprise score” that you try to keep as high as possible. (And of course, there’s no cheating.)
negative surprise log score?
So “requiring an explanation” means “strong evidence the hypothesis space has not yet been searched for”? That seems plausible. Is this your on the spot suggestion or has it been discussed before?