You still haven’t actually calculated the disutility of having a policy of giving the money, versus a policy of not giving the money. You’re just waving your hands. Saying “the EV clearly seems better” is no more helpful than your initial “obvious”.
The calculation I had in mind was basically that if those policies really do have those effects, then which one is superior depends entirely on the ratio between: 1) the difference between likelihoods of large calamity when you pay vs not pay and 2) the actual increase in frequency of muggings
The math I have, the way I understand it, removes the actual -EV of the mugging (keeping only the difference) from the equation and saves some disutility calculation. In my mind, you’d need some pretty crazy values for the above ratio in order for the policy of accepting Pascal Muggings to be worthwhile, and my WAGs are at 2% for the first and about 1000% for the second, with a base rate of around 5 total Muggings if you have a policy of denying them.
I have a high confidence rating for values that stay within the ratio that makes the denial policy favorable, and I find the values that would be required for favoring the acceptance policy highly unlikely with my priors.
Apologies if it seemed like I was blowing air. I actually did some stuff on paper, but posting it seemed irrelevant when the vast majority of LW users appear to have far better mastery of mathematics and the ability to do such calculations far faster than I can. I thought I’d sufficiently restricted the space of possible calculations with my description in the grandparent.
I might still be completely wrong though. My maths have errors a full 25% of the time until I’ve actually programmed or tested them somehow, for average math problems.
You still haven’t actually calculated the disutility of having a policy of giving the money, versus a policy of not giving the money. You’re just waving your hands. Saying “the EV clearly seems better” is no more helpful than your initial “obvious”.
The calculation I had in mind was basically that if those policies really do have those effects, then which one is superior depends entirely on the ratio between: 1) the difference between likelihoods of large calamity when you pay vs not pay and 2) the actual increase in frequency of muggings
The math I have, the way I understand it, removes the actual -EV of the mugging (keeping only the difference) from the equation and saves some disutility calculation. In my mind, you’d need some pretty crazy values for the above ratio in order for the policy of accepting Pascal Muggings to be worthwhile, and my WAGs are at 2% for the first and about 1000% for the second, with a base rate of around 5 total Muggings if you have a policy of denying them.
I have a high confidence rating for values that stay within the ratio that makes the denial policy favorable, and I find the values that would be required for favoring the acceptance policy highly unlikely with my priors.
Apologies if it seemed like I was blowing air. I actually did some stuff on paper, but posting it seemed irrelevant when the vast majority of LW users appear to have far better mastery of mathematics and the ability to do such calculations far faster than I can. I thought I’d sufficiently restricted the space of possible calculations with my description in the grandparent.
I might still be completely wrong though. My maths have errors a full 25% of the time until I’ve actually programmed or tested them somehow, for average math problems.
Don’t worry, the chance of being wrong only costs you 3^^^3*.25 expected utilons, or so.
Hah, that made me chuckle. I ought to remind myself to bust out the python interpreter and test this thing when I get home.