"We can still say, e.g., that one course of action is more rational than another, even in situations where no course of action is most rational." - True.

"But I don't know of any reason to adopt that definition" - perfect rationality means to me more rational than any other agent. I think that is a reasonable definition.

I think you're pessimistic about tech regression.

Assuming survival of some libraries, I think basically any medium-sized functional village (thousands of people, or hundreds with a dash of trade) is adequate to maintain iron age technology. That's valuable enough that any group that survived in a fixed location for more than a couple years could see the value in the investment. (You might not even need the libraries if the right sort of person survived; I suspect I could get a lot of it without that, but it would be a lot less efficient.)

It doesn't take all that much more beyond that to get to some mix of 17th to 19th century tech. Building a useful early 19th-century machine shop is the work of one or two people, full time, for several years. Even in the presence of scavenging, I think such technology is useful enough that it won't take that long to be worth spending time on.

Basically I think anything that's survivable is unlikely to regress to before 17th century tech for a period longer than a few years.

So, this is exactly the sort of thing prediction markets should do well at, right? People without structural incentives to ignore a problem can make accurate predictions and make money. People who care about it can point to the market prices when making their point.

In the black swan case, I think prediction markets will do only somewhat better than alternatives, but here they should do vastly better. Right?

My impression is that Main currently doesn't get much visibility - unless things get promoted, you have to actively go looking for articles there, whereas most people who come to LW at all see discussion by default.

If you don't want to violate the independence axiom (which perhaps you did), then you will need bounded utility also when considering deals with non-PEST probabilities.

In any case, if you effectively give probability a lower bound, unbounded utility doesn't have any specific meaning. The whole point of a double utility is that you will be willing to accept the double utility with half the probability. Once you won't accept it with half the probability (as will happen in your situation) there is no point in saying that something has twice the utility.

A fair point, though I don't think it makes any difference in the context. And I'm not sure the utility function is amenable to MCMC sampling...

I came up with an algorithm that compromises between them.

I am not sure of the point. If you can "sample ... from your probability distribution" then you fully know your probability distribution including all of its statistics -- mean, median, etc. And then you proceed to generate some sample estimates which just add noise but, as far as I can see, do nothing else useful.

If you want something more robust than the plain old mean, check out M-estimators which are quite flexible.

Also, they have a really dumb system where each candidate has both yes and no shares, instead of each election having shares per candidate. Which means there are more different prices than there should be, and no system-enforced rule that the sum of the probabilities = 1.

Actually, the "yes" and "no" shares are the same contracts: Buying a "yes" contract is exactly the same thing as selling a "no" contract. The best offer for "buy yes" plus the best offer for "sell no" will always equal 1, without requiring arbitrage or any action on the part of the market participants.

For some reason they have chosen a counterintuitive user interface such that these contracts appear to be different from each other, but they are the same.

What are the 'best buys' in warm fuzzies?

I want to satisfice my urges for the least cost. Perhaps there's a kind of GiveWell for warm fuzzies out there.

I feel like 'purchasing status' is part of my warm fuzzies calculation, which complicates things.

Perhaps buying coffees for people in line around me?

It is common across prediction markets that the fee structure makes it not worth it to push extreme events to further extremes. Thus unlikely candidates have too much mass and the total adds up to more than 1. But maybe Predictit is even worse for the reasons Anders gives.