There’s also a general reason to try to handle unrealistic scenarios: it can be a shortcut to finding a good theory.

For example, say you have a real-valued cubic equation, and you want to find real-valued answers to it, and imaginary answers don’t even make *sense* because in the situation you’re trying to model they’re *physically impossible*. Even so, your best approach is to use the cubic formula, and simply accept the fact that some of the intermediate computations may produce complex numbers (in which case you should continue with the computation, because they may become real again), and some of the answers may be complex (in which case you should ignore those particular answers).

Solving real-valued polynomials gets a lot easier when you first consider the more general problem of solving complex-valued polynomials. Likewise, solving decision theory without mind reading might get a lot easier when you first consider decision theory with mind reading. Good theories are often very general.

Put another way, I don’t want my algebra to completely break down when I try to take the square root of a negative number, and I don’t want my decision theory to completely break down just because someone can read my mind.

I believe I was thinking of this one:

https://www.lesswrong.com/posts/f6ZLxEWaankRZ2Crv/probability-is-in-the-mind