People normally consider thought experiments where we specify the agent’s prior over worlds. In reality you are presumably using some broad prior like the universal prior, and the reason you would pay a mugger is indeed because they are more likely to exist than the anti-mugger.
OK that makes sense, thanks. This is what I suspected, but I was surprised that so many people are saying that UDT gets mugged without stipulating this; it made it seem like I was missing something.
Playing devil’s advocate:
P(mugger) and P(anti-mugger) aren’t the only relevant quantities IRL
I don’t think we know nearly enough to have a good idea of what policy UDT would choose for a given prior. This leads me to doubt the usefulness of UDT.
People normally consider thought experiments where we specify the agent’s prior over worlds. In reality you are presumably using some broad prior like the universal prior, and the reason you would pay a mugger is indeed because they are more likely to exist than the anti-mugger.
OK that makes sense, thanks. This is what I suspected, but I was surprised that so many people are saying that UDT gets mugged without stipulating this; it made it seem like I was missing something.
Playing devil’s advocate:
P(mugger) and P(anti-mugger) aren’t the only relevant quantities IRL
I don’t think we know nearly enough to have a good idea of what policy UDT would choose for a given prior. This leads me to doubt the usefulness of UDT.