Agreed. The actually-written-up-somewhere arguments that I know of can at most move supernaturalism from “only crazy or overly impressionable people would treat it as a live hypothesis” to “otherwise reasonable people who don’t obviously appear to have a bottom line could defensibly treat it as a Jamesian live hypothesis”. There are arguments that could easily be made that would fix specific failure modes, e.g. some LW folk (including I think Eliezer and lukeprog) mistakenly believe that algorithmic probability theory implies a low prior for supernaturalism, and Randi-style skeptics seem to like fully general explanations/counterarguments too much. But once those basic hurdles are overcome there still seems to be a wide spread of defensible probabilities for supernaturalism based off of solely communicable evidence.
So it sounds like whatever convinced you is incommunicable—something that you know would be unconvincing to anyone else, but which is still enough to convince you despite knowing the alternate conclusions others would come to if informed of it?
some LW folk (including I think Eliezer and lukeprog) mistakenly believe that algorithmic probability theory implies a low prior for supernaturalism
Is the point here that supernatural entities that would be too complex to specify into the universe from scratch may have been produced through some indirect process logically prior to the physics we know, sort of like humans were produced by evolution? Or is it something different?
Alien superintelligences are less speculative and emerge naturally from a simple universe program. More fundamentally the notion of simplicity that Eliezer and Luke are using is entirely based off of their assessments of which kinds of hypotheses have historically been more or less fruitful. Coming up with a notion of “simplicity” after the fact based on past observations is coding theory and has nothing to do with the universal prior, which mortals simply don’t have access to. Arguments should be about evidence, not “priors”.
More fundamentally the notion of simplicity that Eliezer and Luke are using is entirely based off of their assessments of which kinds of hypotheses have historically been more or less fruitful
...
Coming up with a notion of “simplicity” after the fact based on past observations is coding theory and has nothing to do with the universal prior. Arguments should be about evidence, not “priors”.
It isn’t technically a universal prior, but it counts as evidence because it’s historically fruitful. That leaves you with a nitpick rather than showing “LW folk (including I think Eliezer and lukeprog) mistakenly believe that algorithmic probability theory implies a low prior for supernaturalism.”
I don’t think it’s nitpicking as such to point out that the probability of supernaturalism is unrelated to algorithmic probability. Bringing in Kolmogorov complexity is needlessly confusing, and even Bayesian probability isn’t necessary because all we’re really concerned with is the likelihood ratio. The error I want to discourage is bringing in confusing uncomputable mathematics for no reason and then asserting that said mathematics somehow justify a position one holds for what are actually entirely unrelated reasons. Such errors harm group epistemology.
I don’t think it’s nitpicking as such to point out that the probability of supernaturalism is unrelated to algorithmic probability.
I don’t see how you’ve done that. If KC isn’t a universal prior like objective isn’t technically objective but inter-subjective then you can still use KC as evidence for a class of propositions (and probably the only meaningful class of propositions). For that class of propositions you have automatic evidence for or against them (in the form of KC), and so it’s basically a ready-made prior because it passes from the posterior to the prior immediately anyway.
The error I want to discourage is bringing in confusing uncomputable mathematics for no reason and then asserting that said mathematics somehow justify a position one holds for what are actually entirely unrelated reasons.
So a) you think LWers reasons for not believing in supernaturalism have nothing to do with KC, and b) you think supernaturalism exists outside the class of propositions KC can count as evidence as for or against?
I don’t care about A, but If B is your position I wonder: why?
That’s a shame. Any chance you might have suggestions on how to go about obtaining such evidence for oneself? Possibly via PM if you’d be more comfortable with that.
I have advice. First off, if psi’s real then I think it’s clearly an intelligent agent-like or agent-caused process. In general you’d be stupid to mess around with agents with unknown preferences. That’s why witchcraft was considered serious business: messing with demons is very much like building mini uFAIs. Just say no. So I don’t recommend messing around with psi, especially if you haven’t seriously considered what the implications of the existence of agent-like psi would be. This is why I like the Catholics: they take things seriously, it’s not fun and games. “Thou shalt not tempt the Lord thy God.” If you do experiment, pre-commit not to tell anyone about at least some predetermined subset of the results. Various parapsychology experiments indicate that psi effects can be retrocausal, so experimental results can be determined by whether or not you would in the future talk about them. If psi’s capricious then pre-commiting not to blab increases likelihood of significant effects.
Various parapsychology experiments indicate that psi effects can be retrocausal, so experimental results can be determined by whether or not you would in the future talk about them. If psi’s capricious then pre-commiting not to blab increases likelihood of significant effects.
I just thought of something. What you’re saying is that psi effects are anti-inductive.
The capricious-psi literature actually includes several proposed mechanisms which could lead to “anti-inductive” psi. Some of these mechanisms are amenable to mitigation strategies (such as not trying to use psi effects for material advantage, and keeping one’s experiments confidential); others are not.
Agreed. The actually-written-up-somewhere arguments that I know of can at most move supernaturalism from “only crazy or overly impressionable people would treat it as a live hypothesis” to “otherwise reasonable people who don’t obviously appear to have a bottom line could defensibly treat it as a Jamesian live hypothesis”. There are arguments that could easily be made that would fix specific failure modes, e.g. some LW folk (including I think Eliezer and lukeprog) mistakenly believe that algorithmic probability theory implies a low prior for supernaturalism, and Randi-style skeptics seem to like fully general explanations/counterarguments too much. But once those basic hurdles are overcome there still seems to be a wide spread of defensible probabilities for supernaturalism based off of solely communicable evidence.
Essentially, yes.
Is the point here that supernatural entities that would be too complex to specify into the universe from scratch may have been produced through some indirect process logically prior to the physics we know, sort of like humans were produced by evolution? Or is it something different?
Alien superintelligences are less speculative and emerge naturally from a simple universe program. More fundamentally the notion of simplicity that Eliezer and Luke are using is entirely based off of their assessments of which kinds of hypotheses have historically been more or less fruitful. Coming up with a notion of “simplicity” after the fact based on past observations is coding theory and has nothing to do with the universal prior, which mortals simply don’t have access to. Arguments should be about evidence, not “priors”.
...
It isn’t technically a universal prior, but it counts as evidence because it’s historically fruitful. That leaves you with a nitpick rather than showing “LW folk (including I think Eliezer and lukeprog) mistakenly believe that algorithmic probability theory implies a low prior for supernaturalism.”
I don’t think it’s nitpicking as such to point out that the probability of supernaturalism is unrelated to algorithmic probability. Bringing in Kolmogorov complexity is needlessly confusing, and even Bayesian probability isn’t necessary because all we’re really concerned with is the likelihood ratio. The error I want to discourage is bringing in confusing uncomputable mathematics for no reason and then asserting that said mathematics somehow justify a position one holds for what are actually entirely unrelated reasons. Such errors harm group epistemology.
I don’t see how you’ve done that. If KC isn’t a universal prior like objective isn’t technically objective but inter-subjective then you can still use KC as evidence for a class of propositions (and probably the only meaningful class of propositions). For that class of propositions you have automatic evidence for or against them (in the form of KC), and so it’s basically a ready-made prior because it passes from the posterior to the prior immediately anyway.
So a) you think LWers reasons for not believing in supernaturalism have nothing to do with KC, and b) you think supernaturalism exists outside the class of propositions KC can count as evidence as for or against?
I don’t care about A, but If B is your position I wonder: why?
That’s a shame. Any chance you might have suggestions on how to go about obtaining such evidence for oneself? Possibly via PM if you’d be more comfortable with that.
I have advice. First off, if psi’s real then I think it’s clearly an intelligent agent-like or agent-caused process. In general you’d be stupid to mess around with agents with unknown preferences. That’s why witchcraft was considered serious business: messing with demons is very much like building mini uFAIs. Just say no. So I don’t recommend messing around with psi, especially if you haven’t seriously considered what the implications of the existence of agent-like psi would be. This is why I like the Catholics: they take things seriously, it’s not fun and games. “Thou shalt not tempt the Lord thy God.” If you do experiment, pre-commit not to tell anyone about at least some predetermined subset of the results. Various parapsychology experiments indicate that psi effects can be retrocausal, so experimental results can be determined by whether or not you would in the future talk about them. If psi’s capricious then pre-commiting not to blab increases likelihood of significant effects.
I just thought of something. What you’re saying is that psi effects are anti-inductive.
The capricious-psi literature actually includes several proposed mechanisms which could lead to “anti-inductive” psi. Some of these mechanisms are amenable to mitigation strategies (such as not trying to use psi effects for material advantage, and keeping one’s experiments confidential); others are not.
Indeed.
Ok, I feel like we should now attempt to work out a theory of psi caused by some kind of market-like game theory among entities.
Thanks for the advice! Though I suppose I won’t tell you if it turns out to have been helpful?
As lukeprog says here.