Logical counterfactuals are key to Functional Decision Theory and last I heard still an unsolved problem. Unfortunately, I am still rather confused about what exactly we are trying to solve. The only concrete problem I know of in this space is the 5-and-10 problem. But as far as I know, this is solved by writing programs that immediately cause a paradox if they ever discover their output. So presumably there are some unsolved concrete problems that relate to logical counterfactuals?

Edit: I should mention my post on the Cooperation Game as an example. Plus the further work section of this slideshow.

New Answer
New Comment

2 Answers sorted by

So most of the problems seem to be related to making UDT work with logical uncertainty?

2jessicata5y
Some of them specifically require updateless reasoning, some don't or might not. All of them involve logical uncertainty.

avturchin

Dec 16, 2018

10

I encountered something like the 5-and-10 problem in the art market. If I like a picture of the artist A, and I know that I have a good taste, it is an evidence that artists A is good, so I should buy it. The more good art I found, the better is my taste.

Now repeat it 100 times and you end buying pile of garbage.