Anthropic Decision Theory (sound improved)

(sound improved) Stuart Armstrong presents Anthropic Decision Theory at the May 2015 Cambridge Decision Theory Workshop.

The talk sets out to solve the Sleeping Beauty problem and various related anthropic (self-locating belief) problems, not through the calculation of anthropic probabilities, but through finding the correct decision to make. Given certain simple assumptions, it turns out to be possible to do so without knowing the underlying anthropic probabilities. Most common anthropic problems are underspecified from the decision perspective, and this can explain some of the differing intuitions in the subject: selfless and selfish agents, total and average utilitarians, will all reach different decisions in the same problem. These results are formalised into an anthropic decision theory, that is then used to solve many anthropic problems and paradoxes, such as the Presumptuous Philosopher, Adam and Eve, and Doomsday problems.

[BrianTomasik]

I would think that evolved beings facing anthropic dilemmas would be "total utilitarian" (or, rather, "total egoist") even toward eating chocolate, since the fitness benefits of the extra nutrition (assuming the agent needs the energy) would be cumulative, i.e., there's almost twice as much fitness from eating chocolates twice rather than once. Your body remembers, even if your mind doesn't. (Of course, this is not so if we assume that Monday's calories disappear on Tuesday.)

If chocolate has diminishing fitness returns due to one getting full, then we might have something between the halfer and thirder position in evolved creatures.

[yourxianda]

The problem of SIA and SSA is that they both arbitrarily switches perspective in their reasoning. In the case of sleeping beauty problem from first-person perspective I can specify today base on its immediacy to perception. But today cannot be specified in third-person basing on objective differences relevant to the coin toss. This distinction means any attempt to mix the two perspectives together would cause change in the numerical answer. With a perspectively consistent reasoning the probability to the coin toss would be half. Furthermore with a perspectively consistent reasoning we can reject doomsday argument as well as presumptuous philosopher. My argument can be found here: