I’ve said elsewhere that what LWy types call decision theory seems like a wholly unnecessary concept within utilitariany theories. Once you have an axiology (if you even need that) and a verb (eg maximise, though again it’s not clear this is necessary), and the rudimentary maths necessary for the concept of expected value, you don’t need any further a priori definition for your strategy. You’ll obviously want data and statistical analysis techniques, but these things seem to be outside the scope of what people call decision theory.

I haven’t read too much around the subject, for the same reason I haven’t read much around the Bible – with close to 0 initial expectation of its value, the expected value of learning more about it is close to 0 – though I’ve read eg the Desert Hitchiker scenario, and Newcomb’s Paradox, neither of which changed my current view (or rather, both of which were salient in forming it). So this thread has two purposes:

a) A rallying point for people who share my decision-theory-scepticism to reveal themselves, so I can get a sense of what proportion of the utilitarian community feels the same.
b) A place for pro-decision-theorists to persuade us not of any particular theory, but for the utility of any such theory.

Making this a poll, to help with the first purpose. If your answer is anything other than the first one, please can you clarify below?

Following Newcomb's Problem, if you don't have a decision theory, how do you know to "one box" or "two box", even if you have the goal of maximizing utility and a clear definition of utility?

I agree that it would be rather pointless to draw up a decision theory for personal use, but I disagree on some other counts -- I think that one will have a subconscious decision theory that one is implicitly following, could do better to update that decision theory, and that a decision theory will be immensely important when it comes to programming Friendly AI.

Re Newcomb's alleged paradox, it's a cheat. We're told that Omega has this impressive track record, but we never learn how we're supposed to know this, so we don't have sufficient information (ie the information we'd necessarily have have if this were a real scenario) to answer intelligibly.

Chances are we two-box in any real-life scenario. Here, as in the desert hitchhiker, the problem isn't 'being a certain kind of rational' but just the way we present ourselves. The desert jeep driver would behave the same if he thought we were fully rational even if we weren't, and he'd behave differently if we were but he couldn't discern it.

NP is the same, except now it's about the way we presented ourselves in the past. If some psychologist actually pulls this crap on us, we shrug our shoulders and accept the amount of money we've already been given, and if it's me, I make a mental note to avoid having conversations about two-boxing on webforums so freely.

(Incidentally, postulating a psychologist who actually gets it right a significant proportion of the time, as people sometimes do, is a far cry from actually finding a real psychologist who does so. Because we've stopped talking about 'Omega', 'boxes of utility' and stupidly large numbers, people make the mistake of thinking the modified scenario is realistic. But we're still basically talking about a psychic)

Conceivably if we could establish with high probability he was a Bostromesque simulator *and* that the rest was the truth we might two-box on the grounds that he could actually have predicted our action based on physical determinism rather than behavioural extrapolation - though on what possible grounds could we be confident of all that?

I think I agree with you--decision theories are only necessary for nonphysical scenarios where Omega both knows precisely what you're going to do, but you also have the freedom to choose to do differently. I wrote up something about that here.

That being said there are some comments there that I haven't responded to that make me a little less sure than I was that decision theories aren't necessary.

Heh, we do seem to think alike. The one thing I'd query you on is that 'being entirely simulated' doesn't really matter much to 'having free will' assuming you think in normal circumstances we're just a localized bunch of deterministic (perhaps quantumly pseudo-random) processes anyway. We'll still *feel* like we're making a choice, and we'll be equally as wrong or right as we would normally be.

Hence I'm willing to concede that I can just about imagine such a scenario, where I do actually one-box. But I'd have to have some phenomenally good evidence for Omega's claims.

I skimmed the replies, but didn't see anything that seemed to change my view. Which one(s) did you have in mind?

TDT (as opposed to EDT) makes anthropic arguments not matter. Sure my being person number 100 billion means that there probably won't be that many people, but since it's not my fault that I'm person number 100 billion, I ignore it. The world would be a better place if nobody used anthropics, but once it's given that you're person number 100 billion, the world in which you use anthropics is better.

Can you elaborate? Anthropic arguments as I've seen them are usually arguments about probability, whereas decision theories (inasmuch as they purport to add information) seem to be claims about what to do with probability.

I've written about it here.

Did you discuss anthropics in a comment somewhere? The OP doesn't discuss anthropics.