TheMajor

Karma: 925

TheMajor 24 Nov 2013 22:39 UTC
29 points
on: 2013 Less Wrong Census/Survey
Took the survey, and continued to finally make an account. Some questions were ambiguous though (as some other people partially pointed out). I had most problems with:
- Having more children. Over which period of time? As an adolescent I’m not really keen on having children just yet, but I might be in 15 or 20 years.
- Time on LW: I’ve recently finished reading almost all posts on LW, which meant I spent several hours a day on LW. But now that I have finished reading all those I am only reading new posts, which takes no more than 5-10 minutes a day on average. So there is a large difference between a best estimate of the amount of time I spent on LW any previous day and the best estimate of the time I will spend tomorrow. Which of these is the average day?
- Hear about: I had problems interpreting the question. Taking the wording literally the category specified is extremely broad, including even casual comments by colleagues along the lines of: ‘Try checking the batteries more frequently.’ (which is a technique to improve your productivity, provided batteries are important in your line of work).
- Akrasia: meditation. I’ve meditated after sporting frequently in the past, which had nothing to do with akrasia. I decided not to mention the meditation (contrary to Keller, whose comment I only noticed after filling in the survey).

TheMajor 13 Dec 2013 18:43 UTC
2 points
in reply to: ygert’s comment on: Walkthrough of “Definability of Truth in Probabilistic Logic”
I tried applying the proof of the theorem to the problem, as P(q) = P(p) for equivalent statements p,q is clear from the claim P(q) = mu({M: M|= q}) so our desired result should be part of the proof of Theorem 1. However it seems that our desired result is implicitly assumed in the statement “Axiom 1 implies”. Setting P(A|B) = P(B && A)/P(B) (note the reversal of order, without being allowed to replace equivalent statements inside P the mess is even greater if we pick the natural order. Also we want P(B) =/= 0) and writing psi for phi j we can write the claim as:

P(Ti && phi)/P(Ti) = P(Ti && psi && phi)/P(Ti && psi) P(Ti && psi)/P(Ti) + P(Ti && ~psi && phi)/P(Ti && ~psi) P(Ti && ~psi)/P(Ti)

(Here I ignored some annoying brackets, actually there’s more of a mess as its not clear if P((A && B) && C) = P(A && (B && C))). Multiplying both sides by P(Ti) and simplifying now gives:

P(Ti && phi) = P(Ti && psi && phi) + P(Ti && ~psi && phi)

which does not follow from Axiom 1 as the added statements are in the middle of our statement. I am convinced that our claim (P(p) = P(q) for equivalent statements p,q) is required for the proof in the paper and should be added as an extra axiom.

On a sidenote:unde the assumption above (equivalent statements get equal probabilities) axiom 3 is a consequence of axioms 1 and 2. This can be seen as follows: Let q be a contradiction, so ~q is a tautology. By axioms 1 and 2 we have P(q) = P(q && q) + P(q && ~q). But ~q is a tautology, so p && ~q and p are equivalent for every p. In particular we have P(q && ~q) = P(q). But q and (q && q) are also equivalent, so the above can be written as P(q) = P(q) + P(q) so P(q) = 0.

TheMajor 7 Jan 2014 22:19 UTC
8 points
on: [LINK] Why I’m not on the Rationalist Masterlist
Who here thinks that the author of the blog post is female? I did.

Surprise(?)! The blog post doesn’t seem to contain any information that would allow you to deduce the gender of the author. I briefly searched through the blog post and the comment found on Yvain’s site, but I became none the wiser (I stopped searching at that point to respect the author’s privacy). I wonder why I thought that the author of the blog post is female...

TheMajor 10 Feb 2014 7:43 UTC
17 points
on: A Fervent Defense of Frequentist Statistics
I am confused about your use of the word optimal. In particular in the sentences

Bayesian methods are optimal (except for computational considerations).

and

Bayesian methods are locally optimal, not global optimal.

are you talking about the same sort of ‘optimal’? From wikipedia (here and here) I found the rigorous definition of the word ‘optimal’ in the second sentence, which can be written in terms of your utility function (a decision rule is optimal if there is no other decision rule which will always give you at least as much utility and in at least one world will give you more utility).

Also I agree with many of your myths, namely 3,8,9 and 11. I was rather surprised to see that these things even needed to be mentioned, I don’t see why making good trade-offs between truth and computation time should be ‘simple’ (3), as you mentioned the frequentist tests are chosen precisely with robustness in mind (8), bad science is more than getting your statistics wrong (9) (small sidenote: while it might be true that scientists can get confused by frequentist statistics, which might corrupt their science, I don’t think the problem would be smaller when using a different form of statistics, and I therefore think it would not be correct to attribute this bad science to frequentism), and we know from practice that Bayesianism (not frequentism) is the method which has most problems with computational bounds (11).

However, I think it is important to make a distinction between the validity of Bayesianism and the application of Bayesianism. I recall reading on lesswrong (although I cannot find the post at this moment) that the relation between Bayesianism and frequentism should be seen like the relation between Quantum Mechanics and classical physics (although QM has lots of experimental data to support it, so it is rightfully more accepted than Bayesianism). Like QM, Bayesianism is governed by simple mathematical rules (Schrodinger’s equation and Bayes’ theorem), which will give the right answer when supplied with the correct initial conditions. However, to fly a plane we do not invoke QM, and similarly we will in most practical instances to estimate a parameter not invoke Bayes. Instead we use approximations (classical physics/frequentism), which will not give the exact answer but will give a good approximation thereof (as you mention: a method close to the global optimum, although I am still unclear what we are optimising for there). The key point is that these approximations are correct only insofar as they approximate the answer that would be given by the correct theory. If classical physics and QM disagree then QM was correct. Similarly if we have a parameter estimation obtained by a Bayesian algorithm, and one using a frequentist algorithm, the Bayesian one is going to give the correct subjective probability. But the correct algorithms are (nearly?) impossible to implement, so we stick with the approximations. This is why physicists still use and teach classical physics, and why I personally endorse many frequentist tools. The difference between validity and application seems to be lost in myths 4-7 and 10:
- 4: Strictly speaking the only way to truly share your arguments for having a certain degree of belief in a hypothesis would be to share all sensory data that is dependent on the hypothesis (after all, this is how evidence works). This is clearly not feasible, but it would be the correct thing to do if we only care about being correct. You explain in this myth that this does not lead to a simple and quick algorithm. But this is not an argument against validity, it is an argument against a possible application.
- 5: Again this whole myth deals with application. The myth you debunk states that the approximations made when turning degrees of belief into an actual strategy must be bad, and you debunk this by giving an algorithm that gets very good results. But this is not an argument that distinguishes between Bayesianism and frequentism, it merely states that there are easy-to-compute (in a relative sense) algorithms that get close to the correct answer, which we know how to find in theory but not in practice. (In case you are wondering: the approximation takes place in the step where you simplify your utility function to the Regret function, and your prior is whatever decision rule you use for the first horse.)
- 6: This myth hinges on the word ‘simple’. Frequentist methods can deal with many complicated problems, and a lot of high quality work has been done to increase the scope of the tools of frequentism. Saying that only simple models can be dealt with would be an insult. However, as mentioned above, these methods are all approximations, and each method is valid only if the approximations made are satisfied. So while frequentist methods can deal with many complicated models it is imporant to realise that the scope of each method is limited.
- 7+10: Myth 10 seems to be a case of confusion by the people using the tools. Frequentist methods (derived from approximations) come with boundaries, such as limitations on the type of model that can be distilled from data or limitations on the meaning of the outcome of the algorithm (it might answer a different question than the one you hoped to answer). If you break one of these limitations it is not surprising that the results are wacky. This is not a problem of frequentism, provided the tools are explained properly. If the tools are not explained properly then problems arise. Your explanation, we have a class M and a solution E, and we look for a simple approximation which will give E+epsilon, is very clear. Problems arise when the class M is not specified, or the existence of E is unclear. I would like to classify this as an error of the practitioners of frequentism, rather than an error of the method.
Lastly I would like to make a small note that the example on myth 10 is very similar to something called the Boltzmann distribution from statistical physics, discovered in the 19th century. Here the function phi is the energy divided by the temperature.

Edit: during the writing of this post it seems that other people have already made this remark on myth 10. I agree that physicists would probably not interpret this as a game played between nature and the predictor.

TheMajor 10 Feb 2014 8:40 UTC
2 points
in reply to: jsteinhardt’s comment on: A Fervent Defense of Frequentist Statistics
While writing this answer I realised I forgot an important class of exceptions, namely the typical school example of hypothesis testing. My explanation now consists of multiple parts.

To answer the first question: the Bayesian method gives the “correct” answer in the sense that it optimises the expectation of your utility function. If you choose a utility function like log(p) this means that you will find your subjective probabilities. I also think Bayesianism is “correct” in the philosophical sense (which is a property of the theory), but I believe there are many posts on lesswrong that can explain this better than I can.
- The approximation made can often be rewritten in terms of a particular choice of utility function (or risk function, which is more conventional according to wikipedia). As you mentioned choosing the Regret function for cost and a non-silly prior (for example whichever one you are using) will yield a Bayesian algorithm to your problem. Unfortunately I haven’t looked at the specific algorithm in detail, but if admissible solutions are Bayesian algorithms, why would a Bayesian approach using your data not outperform (and therefore produce at least as good asymptotic behaviour) the frequentist algorithm? Also I would like to leave open the possibility that the algorithm you mention actually coincides with a Bayesian algorithm. Sometimes a different approach (frequentism/Bayesianism) can lead to the same conclusion (method).
- Suppose I find myself in a situation in which I have several hypotheses and a set of data. The thing I’m interested in is the probability of each hypothesis given the data (in other words, finding out which hypothesis is correct). In frequentism there is no such thing as a ‘probability of the hypothesis’, after all a hypothesis is either true or false and we don’t know which. So as a substitution frequentists consider the other conditional probability, the probability of seeing this data or worse provided the hypothesis is true, where worse must be defined beforehand. I’d say this is a wrong approach, a very very wrong approach. My opinion is that frequentists have adopted an incorrect worldview which leads them to dismiss and answer the wrong questions in this case. Here I expect pure conflict rather than some Bayesian approach which will coincide with frequentist methods.
I hope this explains how Bayesian and frequentist methods overlap and seem to disagree sometimes, and how many instances of frequentist algorthms should be compared to Bayesian algorithms with a properly chosen utility function.

TheMajor 7 May 2014 6:07 UTC
0 points
on: Rebutting radical scientific skepticism
For testing Quantum Mechanics a relatively easy thing to test is the Hall Effect.

TheMajor 19 Jun 2014 14:02 UTC
7 points
on: Proper value learning through indifference
Nice post! I hadn’t read the previous posts on Cake or Death-like problems, but your post persuaded me to read them.

Enforcing indifference between choices to prevent abuse is a beautiful idea. You remark that as a consequence of this indifference the expected change in utility for changing from utility function ‘v’ to function ‘w’ is equal to 0, but isn’t the converse also true (i.e. the desired update formula follows from the conservation of expected utility)? Above you justify/motivate the change in the update formula (marked in bold in your post) to E(w|v->w) on grounds of symmetry, but doesn’t it simply follow from demanding that changing the utility function itself should not change the (average) utiliy? This sounds like a direct formalisation of demanding that our agent is indifferent with respect to its own utility function.

TheMajor 22 Jun 2014 10:51 UTC
10 points
on: Utilitarianism and Relativity Realism
To phrase your result in terms a physicist would use: an all-time integral of a scalar function (happiness) is not Lorentz-invariant. But rather than draw a philosophical conclusion from this I would suggest modifying the equation for total happiness. If I recall correctly the standard method is defining a happiness density (over space), so rather than state “There is X amount of happiness at this point in time” you would state “At this point in time and space there is Y amount of happiness being created/destroyed”, and then define the total happiness as an all universe integral (so you integrate over spacetime).

I do hope that I’m not making some elementary mistake (imagine the embarrasment), but this thread seems to fall in the general category of ‘attempting to draw philosophical conclusions from a limited understanding of modern physics’. I remember reading material here on LessWrong that warns about this, such as adding up to normality.

TheMajor 22 Jun 2014 18:38 UTC
0 points
in reply to: Eugine_Nier’s comment on: Utilitarianism and Relativity Realism
Ah. It’s about time the assumptions were made clear. I thought that ‘creation of happiness’ was a function defined on spacetime, and the proposed definition defines the total happiness to be only the happiness created on the observers world line. I believe this is not Lorentz-invariant—while a scalar H(x,t) might be invariant under such a transformation we are interested in H(x,t)dt, which messes up the invariance. And I think your remark about the determinant is just a rewording of my point: the determinant of a matrix describes the change in volume of a (in this case) 4-dimensional volume, but if we integrate only in one direction our result can still change (almost) arbitrarily. And therefore introducing an all-space integral solves the problem—this quantity does deal with all four dimensions.

TheMajor 24 Jun 2014 23:18 UTC
1 point
in reply to: Skeptityke’s comment on: Open thread, 23-29 June 2014
I coincidentally read that paper today (confession: I am not a physicist yet, still a student), and I am really suspicious of his use of unitary transformations. A transformation is unitary if and only if it preserves the l^2-norm, which is precisely what the Born rule describes (i.e. that the l^2-norm is the correct norm on wavefunctions). I asked myself which step would break down if rather than the Born rule the actual probability was the amplitude to the power 4, and I haven’t found it yet (provided we also redefine ‘unitary’). But (hopefully) I’m just misunderstanding the problem..?

TheMajor 24 Jun 2014 23:31 UTC
0 points
in reply to: djm’s comment on: Open thread, 23-29 June 2014
I write everything on paper, although I duplicate all the important parts to LaTeX. This means that I go through a notebook approximately every 3 weeks, but it’s definitely worth it. I only use LaTeX to communicate with others and store really important bits (although recently I’ve also just been scanning my paper). I believe this is pretty much standard among my peers.

TheMajor 6 Jul 2014 19:37 UTC
2 points
on: Harry Yudkowsky and the Methods of Postrationality: Chapter One: Em Dashes Colons and Ellipses, Littérateurs Go Wild
If I’m being honest (and I do hope that that is what you asked for) then I’d like to mention that personally I didn’t really enjoy the story above. The idea of self-aware characters sounds awesome, but I was a bit disturbed by the fact that the main character was named after Eliezer Yudkowsky. Making brilliant characters with nigh-impossible rationality skills is fine (and lots of fun to read!), but this seems to cross that subtle border between interesting fanfic and uncomfortable devotion. My did-I-just-join-a-cult-sensor is going off. I think changing the name of Harry Potter-Yudkowsky would be a good idea.

The above is a bit exaggerated, I’d still read the story and enjoy it if you kept the name. But the initial flinch was real.

TheMajor 7 Jul 2014 9:30 UTC
2 points
in reply to: Squark’s comment on: Quickly passing through the great filter
This doesn’t seem to address the correct point. Our goal is to surpass late filters, effectively turning universes with a late filter into universes with no filter at all. Surely the value of such an action is comparable to the value of a universe with an early filter (provided that it has been passed)?

TheMajor 7 Jul 2014 21:10 UTC
2 points
in reply to: Viliam_Bur’s comment on: Separating university education from grading
This actually generalises beyond mathematics quite well—rather than ask people to describe a theory let them apply the theory to an example (perhaps even an example that actually happened).

As I haven’t encountered this too often in practice (outside of mathematics, where proving for an exam is the rule rather than the exception) the obvious question is: why isn’t this being used?

TheMajor 7 Jul 2014 21:16 UTC
0 points
in reply to: jsteinhardt’s comment on: Separating university education from grading
I don’t think grading plays as small a role as you seem to think—remove the grading (and just pass everybody) and see how many students show up for class.

TheMajor 12 Jul 2014 13:19 UTC
2 points
in reply to: CronoDAS’s comment on: Harry Yudkowsky and the Methods of Postrationality: Chapter One: Em Dashes Colons and Ellipses, Littérateurs Go Wild
Wow, I loved it!

TheMajor 13 Jul 2014 12:42 UTC
2 points
on: Confused as to usefulness of ‘consciousness’ as a concept
I notice that I am still confused. In the past I hit ‘ignore’ when people talked about consciousness, but lets try ‘explain’ today.

The original post states:

Would we disagree over what my computer can do? What about an animal instead of my computer? Would we feel the same philosophical confusion over any given capability of an average chicken? An average human?

Does this mean that if two systems have almost the same capabilities that we would then also expect them to have a similar chance of receiving the consciousness label? In other words: is consciousness, like IQ, a lossy compression of other properties? I would like to compare this idea with the comment from RichardKennaway, who asks:

Everyone reading this, please take a moment to see whether you have any sensation that you might describe by those words.

I personally have never thought anything like ‘I am feeling so conscious today’ or even ‘I am actively experiencing consciousness’. However, as I expected consciousness to be some sort of label attached to systems with certain properties, along with the observation that I am human and pretty much all other humans claim to be conscious, my prior for me being conscious is pretty high. But I don’t feel that I can answer ‘Yes’ to RichardKennaway’s question.

I have a hard time believing that my not feeling conscious is significant evidence to sway my prior away from me being conscious (especially since feelings are pretty easy to manipulate), i.e. my current solution is to reject RichardKennaway’s question as insignificant (feeling conscious might not be all that related to being conscious). My questions are: are there better tests for being conscious available (presumably no, since the whole point of the discussion is that consciousness is hard to quantify), and if not then why should we bother to discuss whether or not something is conscious at all?

I am sorry that my post isn’t all that coherent, I seem to be having trouble identifying exactly what my problem with ‘consciousness’ is.

Small note: while writing this reply I noticed that the original post can be interpreted as an explanation of what I am trying to say: if consciousness is a lossy compression then whether something has it or not becomes a moot point if we know plenty other properties, so if I notice that I can do most things that other humans also can (abstract reasoning, planning and executing strategies, using tools etc.) then whether or not I am conscious should be a moot point as it does not influence any prediction of my actions.

TheMajor 13 Jul 2014 17:47 UTC
9 points
in reply to: polymathwannabe’s comment on: Rationality Quotes July 2014
This seems false—the quote illustrates that one should not believe everything just because it was written by a scientist, but there is also such a thing as being too skeptical. With a bit of effort you can deny pretty much any conclusion you dislike. Behaving as advised in this quote is better than taking scientists’ word on blind faith, but in order to actually learn the truth from scientific writing you need to invest a lot more effort than just this!

TheMajor 13 Jul 2014 22:38 UTC
10 points
in reply to: Viliam_Bur’s comment on: Why I Am Not a Rationalist, or, why several of my friends warned me that this is a cult
I have come across serious criticism of the PhD programs at major universities, here on LW (and on OB). This is not quite the same as a recommendation to not enroll for a PhD, and it most certainly is not the same as a recommendation to quit from an ongoing PhD track, but I definitely interpreted such criticism as advice against taking such a PhD. Then again I have also heard similar criticism from other sources, so it might well be a genuine problem with some PhD tracks.

For what it’s worth my personal experiences with the list of main points (not sure if this should be a separate post, but I think it is worth mentioning):

Rationality doesn’t guarantee correctness.

Indeed, but as Villiam_Bur mentions this is way too high a standard. I personally notice that while not always correct I am certainly correct more often thanks to the ideas and knowledge I found at LW!

In particular, AI risk is overstated

I am not sure but I was under the impression that your suggestion of ‘just building some AI, it doesn’t have to be perfect right away’ is the thought that researchers got stuck on last century (the problem being that even making a dumb prototype was insanely complicated), when people were optimistically attempting to make an AI and kept failing. Why should our attempt be different? As for AI risk itself: I don’t know whether or not LW is blowing the risk out of proportion (in particular I do not disagree with them, I am simply unsure).

LW has a cult-like social structure.

I agree wholeheartedly, you beautifully managed to capture my feelings of unease. By targeting socially awkward nerds (such as me, I confess) it becomes unclear whether the popularity of LW among intellectuals (e.g. university students, I am looking for a better word than ‘intellectuals’ but fail to find anything) is due to genuine content or due to a clever approach to a vulnerable audience. However from my personal experience I can confidently assert that the material from LW (and OB, by the way) indeed is of high quality. So the question that remains is: if LW has good material, why does it/do we still target only a very susceptible audience? The obvious answer is that the nerds are most interested in the material discussed, but as there are many many more non-nerds than nerds it would make sense to appeal to a broader audience (at the cost of quality), right? This would probably take a lot of effort (like writing the Sequences for an audience that has trouble grasping fractions), but perhaps it would be worth it?

Many LWers are not very rational.

In my experience non-LWers are even less rational. I fear that again you have set the bar too high—reading the sequences will not make you a perfect Bayesian with Solomonoff priors, at best it will make you a bit closer of an approximation. And let me mention again that personally I have gotten decent mileage out of the sequences (but I am also counting the enjoyment I have reading the material as one of the benefits, I come here not just to learn but also to have fun).

LW membership would make me worse off.

This I mentioned earlier. I notice that you define success in terms of money and status (makes sense), and the easiest ways to try to get these would be using the ‘Dark Arts’. If you want a PhD, just guess the teachers password. It worked for me so far (although I was also interested in learning the material, so I read papers and books with understanding as a goal in my spare time). However these topics are indeed not discussed (and certainly not in the form of: ‘In order to get people to do what you want, use these three easy psychological hacks’) on LW. Would it solve your problem if such things were available?

“Art of Rationality” is an oxymoron.

Just because something is true does not mean that it is not beautiful?

TheMajor 14 Jul 2014 0:22 UTC
6 points
in reply to: Mark_Friedenbach’s comment on: Why I Am Not a Rationalist, or, why several of my friends warned me that this is a cult
I think a part of the reason is that most people would never even consider a polyamorous relationship, whereas it might for quite a lot of people be a better option than the alternatives. If this is true then being in a polyamorous relationship is a strong indicator of actually considering alternatives and embracing the truth when stumbling upon it.

Having said all that I think it is not one of the central activities related to LW, the implication mentioned above is valid only so long as people don’t make a habit out of trying (radically) different sorts of romance.