= 27d567bc2d48d9f40e9ccc20ea370b15)
IlyaShpitser comments on Deliberate Grad School - Less Wrong
You are viewing a comment permalink. View the original post to see all comments and the full post content.
= 783df68a0f980790206b9ea87794c5b6)
Loading…= b715d02e85f7b3b4ae65ca3d3e450868)
Subscribe to RSS Feed
Powered by Reddit
= f037147d6e6c911a85753b9abdedda8d)
You are viewing a comment permalink. View the original post to see all comments and the full post content.
Comments (153)
Luke is not qualified to shit on academic philosophy. He simply doesn't have the background or the overview. And it's a terrible idea for social reasons, it just makes people not take LW seriously. I would be happy to accept critiques of the philosophy establishment from e.g. Clark Glymour, not from Luke. There is a ton of value in philosophy you are leaving on the table if you shit on philosophy.
My other big annoyance is the "LW Bayesians" (who are similarly not qualified generally to have strong opinions about these issues, and instead should read stats/ML literature). Although I should say very sophisticated stats folks occasionally post here (but I don't count them among the "LW Bayesians" number, as they understand issues with Bayes very well).
Love this, Luke is actually well read so maybe it's a bit tough on him, but the casual dismissal and elitist posturing is pretty dumb and cringe inducing. Philosophy is underrated around these parts.
Could you provide an object level counter argument please? A strong one would give me a lot more credence that Luke's work was not an accurate portrayal of academic philosophy.
(Three would be preferred)
(Object level might look like "philosophers are making useful progress by metric X" or "I expect philosophers' work to be very useful in area of science a because b" or "doing a PhD in philosophy has lots of value in the world for reasons p, q and r")
I am not very interested in convincing you.
You said:
So look for the value! Don't write the entire field off, lots of smart people there, probably you are missing something.
But for example quite a few very smart causal inference people are in philosophy. That conference on decision theory MIRI went to in Cambridge was hosted by philosophers. Some philosophers deal with very hard problems that do not map onto empiricism very well, etc.
I think Luke will agree with you on what you say here, though. I remember commenting on one of his posts that was critical of philosophy, saying that his arguments didn't really apply to the area of philosophy I'm involved in (technical philosophy of science). Luke's response was essentially, "I agree. I'm not talking about philosophy of science." I think he'd probably say the same about philosophical work on decision theory and causal inference.
Isn't that motte/bailey: "philosophy, a diseased discipline" is not a very discriminating title. The best line of his post is this:
And this is definitely ok!
But again, I am not super interested in arguing with people about whether philosophy is worthwhile. I have better things to do. I was only pointing out in response to the OP that I have been harping on LW's silly anti-academic sentiment for ages, that's all.
Not sure it's motte-and-bailey. I do think there are several serious pathologies in large swathes of contemporary philosophy. And I say this not as a dilettante, but a professional philosopher. There are areas of philosophy where these pathological tendencies are being successfully held at bay, and I do think there are promising signs that those areas are growing in influence. But much of mainstream philosophy, especially mainstream metaphysics and epistemology, does suffer from continued adherence to what I consider archaic and unhelpful methodology. And I think that's what Luke is trying to point out. He does go overboard with his rhetoric, and I think he lacks a feel for the genuine insights of the Western philosophical tradition (as smart and insightful as I think Yudkowsky is, I really find it odd that someone who purports to be reasonably familar with philosophy would cite him as their favorite philosopher). But I think there is a sound point lurking under there, and not merely a banal "motte"-style point.
I absolutely agree with you on the silliness of the anti-academic sentiment.
Would you mind explaining your perspective? I'm always interested to hear more angles on this, since with my current sample-size being roughly three (Dennett, Railton, Churchland), I tend to think I have an incomplete picture.
Everyone on LW should consider Francis Bacon their patron saint, imo :).
LW seems to have picked the Presbyterian minister Thomas Bayes as its patron saint with the Franciscan friar William of Ockham running a close second :-)
If I had to pick one, I'd go with Laplace.
In defense of Luke, when I've spent the time to read through philosophy books by strong-naturalist academic philosophers, they've often devoted page-counts easily equivalent in length to "Philosophy: a diseased discipline" to carefully, charitably, academically, verbosely tearing non-naturalist philosophy a new asshole. Luke's post has tended to be a breath of fresh air that I reread after reading any philosophy paper that doesn't come from a strongly naturalist perspective.
It sincerely worries me that the academics in philosophy who do really excellent work, work that does apply to the real world-that-is-made-of-atoms, work that does map-the-territory, have to spend large amounts of effort just beating down obviously bad beliefs over and over again. You should be able to shoot down a bad idea once, preferably in the peer-review phase, and not have to fight it again and again like a bad zombie.
(Examples of obviously bad ideas: p-zombies, Platonism, Bayesian epistemology (the latter two may require explanation).)
Now, to signal fairness even where I'm blatantly opinionated, plenty of people on LW are indeed irritatingly "men of one idea", that usually being some variation on AIXI. And in fact, plenty of people on LW hold philosophical opinions I consider obviously bad, like mathematical Platonism.
But the answer to those bad things hasn't usually been "more philosophy", as if any philosophy is good philosophy, but instead more naturalism, investing more effort to accommodate conceptual theorizing to the world-that-is-made-of-atoms.
Since significant portions of academic philosophy (for instance, Thomas Nagel) are instead devoted to the view - one that I once expected to be contrarian but which I now find depressingly common - that science and naturalism are wrong, or that they are unjustified, or that they are necessarily incapable of answering some-or-another important question - having one page on a contrarian intellectual-hipsters' website devoted to ragging on these ought-to-be-contrarian views is a bit of a relief.
Could you provide that explanation?
Sure.
If we take Platonism to be the belief that abstract objects (take, for instance, the objects of ZFC set theory) actually exist in a mind-independent way, if not in a particularly well-specified way, then it occurs because people mistake the contents of their mental models of the world for being real objects, simply because those models map the world well and compress sense-data well. In fact, those models often compress most sense-data better than the "more physicalist" truth would: they can be many orders of magnitude smaller (in bits of program devoted to generative or discriminative modelling).
However, just because they're not "real" doesn't mean they don't causally interact with the real world! The point of a map is that it corresponds to the territory, so the point of an abstraction is that it corresponds to regularities in the territory. So naive nominalism isn't true either: the abstractions and what they abstract over are linked, so you really can't just move names around willy-nilly. In fact, some abstractions will do better or worse than others at capturing the regularities in sense-data (and in states of the world, of course), so we end up saying that abstractions can exist on a sliding scale from "more Platonic" (those which appear to capture regularities we've always seen in all our previous data) to "more nominalist" (those which capture spurious correlations).
Now, for "Bayesian epistemology", I'm taking the Jaynesian view, which is considered extreme but stated very clearly and precisely, that reasoning consists in assigning probabilities to propositions. People who oppose Bayesianism will usually then raise the Problem of the Prior, and the problem of limited model classes, and so on and so forth. IMHO, the better criticism is simply: propositions are not first-order, actually-existing objects (see above on Platonism)! Consider a proposition to be a set of states some model can be in or not be in, and we can still use Bayesian statistics, including the kinds of complex Bayesian modelling used to model the mind, without endorsing Bayesian philosophy, which would require us to believe in spooky things called "propositions" and "logic" -- while also not believing in certain spooky things called "continuous random variables", which don't really fit into Cox's Theorem very well, if I understood Jaynes correctly.
How do you account for the fact that numbers are the same for everyone? Of course, not everyone knows the same things about numbers, but neither does everyone know the same things about Neptune. Nevertheless, the abstract objects of mathematics have the same ineluctability as physical objects. Everyone who looks at Neptune is looking at the same thing, and so is everyone who studies ZFC. These abstract objects can be used to make models of things, but they are not themselves those models.
You could equally say that everyone who looks at the rules of chess sees the same thing. In order to show some inevitability to ZFC, you have to show that unconnected parties arriving at it independently.
Two correct maps of the same territory, designed to highlight the same regularities and obscure the same sources of noise, will be either completely the same or, in the noisy case, will approximate each-other.
Just because there's no Realm of Forms doesn't mean that numbers can be different for different people without losing their ability to compressively predict regularities in the environment.
The motivation actually seems to be the Correspondence Theory of Truth..that is mentioned several timesin subsequent comments.
Indeed, even though when you use a Lossy-Correspondence/Compression Theory of Truth, abstract objects become perfectly sensible as descriptions of regularities in concrete objects.
I don't follow your argument re Bayesian epistemology, in fact, I find it not at all obvious. The argument looks like insisting on a different vocabulary while doing the same things, and then calling it statistics rather than epistemology.
Can you give a pointer to where he disbelieves in these? He does refer to them apparently unproblematically here and there, e.g. in deducing what a noninformative prior on the chords of a circle should be.
1) Dissolving epistemology to get statistics of various kinds underneath is a good thing, especially since the normal prescription of Bayesian epistemology is, "Oh, just calculate the posterior", while in Bayesian statistics we usually admit that this is infeasible most of the time and use computational methods to approximate well.
2) The difference between Bayesian statistics and Bayesian epistemology is slight, but the difference between Bayesian statistics and the basic nature of traditional philosophical epistemology that the Bayesian epistemologists were trying to fit Bayesianism into is large.
3) The differences start to become large when you stop using spaces composed of N mutually-exclusive logical propositions arranged into a Boolean algebra. For instance, computational uncertainty and logical omniscience are nasty open questions in Bayesian epistemology, while for an actual statistician it is admitted from the start that models do not yield well-defined answers where computations are infeasible.
I can't, since the precise page number would have to be a location number in my Kindle copy of Jaynes' book.
I found a partial answer to the question I asked in the sibling comment. By chance I happened to need to generate random chords of a circle covering the circle uniformly. In searching on the net for Jaynes' solution I came across a few fragments of Jaynes' views on infinity. In short, he insists on always regarding continuous situations as limits of finite ones (e.g as when the binomial distribution tends to the normal), which is unproblematic for all the mathematics he wants to do. That is how the real numbers are traditionally formalised anyway. All of analysis is left unscathed. His wider philosophical objections to such things as Cantor's transfinite numbers can be ignored, since these play no role in statistics and probability anyway.
I don't know about the technicalities regarding Cox's Theorem, but I do notice a substantial number of papers arguing about exactly what hypotheses it requires or does not require, and other papers discussing counterexamples (even to the finite case). The Wikipedia article has a long list of references, and a general search shows more. Has anyone written an up to date review of what Cox-style theorems are known to be sound and how well they suffice to found the mathematics of probability theory? I can google /"Cox's theorem" review/ but it is difficult for me to judge where the results sit within current understanding, or indeed what the current understanding is.
I don't know. But I will say this: I am distrustful of a foundation which takes "propositions" to be primitive objects. If the Cox's Theorem foundation for probability requires that we assume a first-order logic foundation of mathematics in general, in which propositions cannot be considered as instances of some larger class of things (as they can in, for personal favoritism, type theory), then I'm suspicious.
I'm also suspicious of how Cox's Theorem is supposed to map up to continuous and non-finitary applications of probability -- even discrete probability theory, as when dealing with probabilistic programming or the Solomonoff measure. In these circumstances we seem to need the measure-theoretic approach.
Further: if "the extension of classical logic to continuous degrees of plausibility" and "rational propensities to bet" and "measure theory in spaces of normed measure" and "sampling frequencies in randomized conditional simulations of the world" all yield the same mathematical structure, then I think we're looking at something deeper and more significant than any one of these presentations admits.
In fact, I'd go so far as to say there isn't really a "Bayesian/Frequentist dichotomy" so much as a "Bayesian-Frequentist Isomorphism", in the style of the Curry-Howard Isomorphism. Several things we thought were different are actually the same.
That word, "obviously", I don't think it means what you think it means :-)
I'm not a fan of mathematical Platonism, but physical realists, however hardline, face some very difficult problems regarding the ontologica status of physical law, which make Platonism hard to rule out. (And no, the perenially popular "laws are just descriptions" isn't a good answer).
P-zombies as a subject worth discussing, or as something that can exist in our univese? But most of the people who discuss PZs don't think they can exist in our universe. There is some poor quality criticisim of philosophy about as well.
The problems with Bayes are suffcieintly non-obvious to have eluded many or most at LW.
On the one hand, I think that page in specific is actually based on outdated Bayesian methods, and there's been a lot of good work in Bayesian statistics for complex models and cognitive science in recent years.
On the other hand, I freaking love that website, despite its weirdo Buddhist-philosophical leanings and one or two things it gets Wrong according to my personal high-and-mighty ideologies.
And on the gripping hand, he is very, very right that the way the LW community tends to phrase things in terms of "just Bayes it" is not only a mischaracterization of the wide world of statistics, it's even an oversimplification of Bayesian statistics as a subfield. Bayes' Law is just the update/training rule! You also need to discuss marginalization; predictive distributions; maximum-entropy priors, structural simplicity priors, and Bayesian Occam's Razor, and how those are three different views of Occam's Razor that have interesting similarities and differences; model selection; the use of Bayesian point-estimates and credible-hypothesis tests for decision-making; equivalent sample sizes; conjugate families; and computational Bayes methods.
Then you're actually learning and doing Bayesian statistics.
On the miniature nongripping hand, I can't help but feel that the link between probability, thermodynamics, and information theory means Eliezer and the Jaynesians are probably entirely correct that as a physical fact, real-world event frequencies and movements of information obey Bayes' Law with respect to the information embodied in the underlying physics, whether or not I can model any of that well or calculate posterior distributions feasibly.
Starting out by expecting a view opposed to your own to be contrarian is a typical form of overconfidence, and not just overconfidence about other people's opinions.
Sometimes, yes. However, I rather expect that naturalism should be the consensus.
He could have saved himself some trouble by writing "Philosophy: a Partly Diseased Disciplien" or "Philosophy: a Bit of a Curate's Egg".
How about:
I think the article Ilya has in mind is this one: Philosophy, a diseased discipline.
I can help with the second request:
The British Journal for the Philosophy of Science
That seems to be entirely analytic philosophy. My problem is that analytical philosophy is culturally irrelevant. Anthropologists, sociologists, art theorists, and artists talk about continental philosophy, Saussurian (!) linguistics, and psychoanalytic theory. The only things they use from analytic philosophy are arguments like Godel's incompleteness theorem, Wittgenstein's later stuff, or Quine's ontological relativism, that they interpret as saying that analytic philosophy doesn't work.
Analytic philosophers will find the strength to carry on, somehow.
Good for analytic philosphy, but my real concern is with literature. Literature today is captive to bad philosophy. Poetry, even more explicitly so.
Someone who's studied stats and ML is much more qualified to talk about philosophy than someone who's studied academic philosophy.
My comment may be irrelevant. You didn't provide a link to Luke's article, so I don't have the context, and am only guessing at your meaning.
^ this is what I am talking about. For some reason I think Luke has a bachelor's degree with a major in cognitive science (but I don't remember exactly).
I was under the impression that he studied psychology, but dropped out before graduating. (An old interview has him mentioning that "I studied psychology in university but quickly found that I learn better and faster as an autodidact", and back when he was still employed at MIRI, his profile on the staff page didn't mention any degree whereas it did for almost everyone else.)
Just so we are clear -- I am not really attacking Luke. I met him, we talked on skype, etc. He's a sensible dude. I am just not weighing his opinion of philosophy very highly. "Mixture of experts" and all that.