This is I think orthogonal to the point I’m trying to make. To say it again with a straight face (which I should probably have done the first time, as per my comments in the niceness thread about bad arguments being harder to state with a straight face):
Surely the proper reading of this post is not that everyone is obliged to reply to every comment that criticises their position. On a site like this, one must make a decision on which criticisms to reply to.
More generally, this is the tu quoque fallacy. Whether or not Eliezer is a paragon of perfect rationality and a flawless exemplar of his own advice is a different question to whether the advice is good.
Of course the advice is good. In that particular case, Eliezer was exaggerating; his intention was to say that randomness wasn’t necessary in order to achieve the best result, not that it is always possible to achieve a result better than what randomness achieves. Why he didn’t respond, I don’t know. But it is certainly true that the very logical rudeness that Eliezer is talking about is common on Less Wrong as well as elsewhere, and I do find it annoying that practically everyone is talking about it as something that only other people do.
I suppose that must have happened sometime, but next time you find yourself postulating this as part of an explanation, please stop, notice, and feel a little confused.
Actually, that goes for everyone in this thread deconstructing my supposed mistake, based on (a) a misquotation (b) not realizing that every algorithm which can be “improved by randomizing” can in fact be improved further by derandomizing (except in cases where an intelligent opponent can predict a given set of bits if they are produced by a “deterministic” process, but not predict the same bits if they are produced by a “random” process). I sometimes make mistakes, but I also sometimes don’t, and if you can hold both possibilities in mind without it destroying your internal critic, it will probably help in the long run.
Ok, but there some random algorithms which cannot be improved by derandomizing, precisely because the random algorithm does just as well as any deterministic algorithm: for example, if there is some event that has an exact 50% chance of happening, all algorithms, random or not, do equally well at guessing whether it will happen or not.
In other words, such a case doesn’t satisfy the condition that the algorithm can be “improved by randomizing.”
I take it that such an algorithm couldn’t be improved in accuracy, but I expect any randomized algorithm would be more cycle-intensive than a constant rule of “guess that event X will happen”—which will perform just as well.
the claim was that when you can improve an algorithm by randomizing it, you can improve it further by derandomizing it
Unknowns:
randomness wasn’t necessary in order to achieve the best result, not that it is always possible to achieve a result better than what randomness achieves
I think a claim that a randomized algorithm is never better than some deterministic one is interesting and probably true (possibly not, which is why it’s interesting). Is Eliezer really making an even stronger claim than this? Is any precise claim being made? The meaning of “improving an algorithm by adding (or removing) randomization” is pretty vague. Improving in what way? Isn’t any non-randomized algorithm technically a member of “randomized algorithms”? If not, can’t it be made technically randomized at no asymptotic cost?
Good points. I didn’t see it as tu quoque, though; I saw it as an opportunity for Eliezer to qualify his heuristic and bring it more in line with reality.
It seems to me that “logical etiquette” requires us to respond to the very strongest arguments made by our opponents as well as to read arguments reasonably. Without going back to this discussion, it’s hard to tell who is being rude.
For example, suppose I make the following argument:
“American law schools always charge ridiculously high tuition because they can get away with it. The availability of generous student loans creates artificial demand and gives law schools the ability to charge exorbitant tuition. We need to reform the student loan program to fix this problem.”
If somebody responds by simply saying “well, I know a law school in Nebraska which charges $500 per year so your premise is wrong,” they are being rude. Because obviously when I said “always,” I really meant “generally speaking.” Further, the person is not really responding to the substance of my argument.
I have no idea if this is what happened with Eliezer and Thomas. I’m just saying.
I went and looked at the original post—it’s very long, but does actually address Thomas’ question:
Or imagine that the combination [to the lock that we’re trying to pick] changes every second. In this case, 0-0-0-0, 0-0-0-0 is just as good as the randomized algorithm—no better and no worse. What this shows you is that the supposedly “random” algorithm is “better” [than trying 0-0-0-0 repeatedly] relative to a known regularity of the lock—that the combination is constant on each try. Or to be precise, the reason the random algorithm does predictably better than the stupid one is that the stupid algorithm is “stupid” relative to a known regularity of the lock.
Given the length of the post, I think the most reasonable assumption is that Thomas had forgotten that that particular point had been covered by the time he reached the end of it. I know I had by the time I saw his original question.
That’s really pretty ridiculous. You can try to speak precisely. Why should we all concede that hyperbole is acceptable in an argument?
If you want to argue about student loans you could: approach it from another side or focus on elite/private law schools. Overstating your case only works when preaching to the choir. Then, it misinforms and makes you less credible to others.
Why should we all concede that hyperbole is acceptable in an argument?
I’m not saying that hyperbole is acceptable. But if I engage in hyperbole, it’s still rude to nitpick the hyperbole while ignoring the strongest part of the argument. In this case, the argument still stands if one substitutes “generally speaking” for “always.”
You can try to speak precisely
Sure, but it’s difficult to be sufficiently precise at all times. It’s rude to seize upon an inprecision to dismiss an argument while ignoring the main thrust of the argument.
I’m trying to make the point that its easy to jump on (especially glaring) imprecision. Your general thrust is weakened, often unfairly, by its presence. It can be a bummer for an argument if people jump on imprecise things, but hopefully you can stop that before it happens by omitting them in the first place.
This seems like the perfect place for the person making the claim about student loans to make a concession (demoting their “always” to “almost always”) thus making their debating partner more comfortable to listen to the meat of their argument; but it is also necessary not to take that demotion from “always” to “almost always” as defeat of the entire argument.
I basically agree, and in this situation my response would be something like “I concede that not all law schools charge ridiculously high tuition but I think my basic point stands.
Sometimes that sort of precision adds too much length. If you see an easily-fixed problem with an argument, it behooves you to point out the fix in the same comment as the problem.
I think that is fair. That would be the reasonable thing to do in a debate.
Precision in this case is not any longer (i.e. always vs typically). It can at times, but for people down with logic, you’d think always versus there exists, etc. would be a big deal.
By the way, please note that I did reply to this question once Adelene asked it. This is because I have Tim Tyler labeled as hopeless, but not Adelene.
Well, your purpose in replying to comments, which might be a combination of several things including enlightening yourself, enlightening others, or entertainment of yourself or others. Even if your motives were entirely based around yourself and others being less wrong, you would still be wise to look at who is asking when considering whether to reply.
In general I wouldn’t set out such criteria out loud, but if it were me in these particular circumstances I’d be very tempted.
This is I think orthogonal to the point I’m trying to make. To say it again with a straight face (which I should probably have done the first time, as per my comments in the niceness thread about bad arguments being harder to state with a straight face):
Surely the proper reading of this post is not that everyone is obliged to reply to every comment that criticises their position. On a site like this, one must make a decision on which criticisms to reply to.
More generally, this is the tu quoque fallacy. Whether or not Eliezer is a paragon of perfect rationality and a flawless exemplar of his own advice is a different question to whether the advice is good.
Of course the advice is good. In that particular case, Eliezer was exaggerating; his intention was to say that randomness wasn’t necessary in order to achieve the best result, not that it is always possible to achieve a result better than what randomness achieves. Why he didn’t respond, I don’t know. But it is certainly true that the very logical rudeness that Eliezer is talking about is common on Less Wrong as well as elsewhere, and I do find it annoying that practically everyone is talking about it as something that only other people do.
I suppose that must have happened sometime, but next time you find yourself postulating this as part of an explanation, please stop, notice, and feel a little confused.
Actually, that goes for everyone in this thread deconstructing my supposed mistake, based on (a) a misquotation (b) not realizing that every algorithm which can be “improved by randomizing” can in fact be improved further by derandomizing (except in cases where an intelligent opponent can predict a given set of bits if they are produced by a “deterministic” process, but not predict the same bits if they are produced by a “random” process). I sometimes make mistakes, but I also sometimes don’t, and if you can hold both possibilities in mind without it destroying your internal critic, it will probably help in the long run.
Ok, but there some random algorithms which cannot be improved by derandomizing, precisely because the random algorithm does just as well as any deterministic algorithm: for example, if there is some event that has an exact 50% chance of happening, all algorithms, random or not, do equally well at guessing whether it will happen or not.
In other words, such a case doesn’t satisfy the condition that the algorithm can be “improved by randomizing.”
I take it that such an algorithm couldn’t be improved in accuracy, but I expect any randomized algorithm would be more cycle-intensive than a constant rule of “guess that event X will happen”—which will perform just as well.
Eliezer:
Unknowns:
I think a claim that a randomized algorithm is never better than some deterministic one is interesting and probably true (possibly not, which is why it’s interesting). Is Eliezer really making an even stronger claim than this? Is any precise claim being made? The meaning of “improving an algorithm by adding (or removing) randomization” is pretty vague. Improving in what way? Isn’t any non-randomized algorithm technically a member of “randomized algorithms”? If not, can’t it be made technically randomized at no asymptotic cost?
Good points. I didn’t see it as tu quoque, though; I saw it as an opportunity for Eliezer to qualify his heuristic and bring it more in line with reality.
It seems to me that “logical etiquette” requires us to respond to the very strongest arguments made by our opponents as well as to read arguments reasonably. Without going back to this discussion, it’s hard to tell who is being rude.
For example, suppose I make the following argument:
“American law schools always charge ridiculously high tuition because they can get away with it. The availability of generous student loans creates artificial demand and gives law schools the ability to charge exorbitant tuition. We need to reform the student loan program to fix this problem.”
If somebody responds by simply saying “well, I know a law school in Nebraska which charges $500 per year so your premise is wrong,” they are being rude. Because obviously when I said “always,” I really meant “generally speaking.” Further, the person is not really responding to the substance of my argument.
I have no idea if this is what happened with Eliezer and Thomas. I’m just saying.
I went and looked at the original post—it’s very long, but does actually address Thomas’ question:
Given the length of the post, I think the most reasonable assumption is that Thomas had forgotten that that particular point had been covered by the time he reached the end of it. I know I had by the time I saw his original question.
That’s really pretty ridiculous. You can try to speak precisely. Why should we all concede that hyperbole is acceptable in an argument?
If you want to argue about student loans you could: approach it from another side or focus on elite/private law schools. Overstating your case only works when preaching to the choir. Then, it misinforms and makes you less credible to others.
I’m not saying that hyperbole is acceptable. But if I engage in hyperbole, it’s still rude to nitpick the hyperbole while ignoring the strongest part of the argument. In this case, the argument still stands if one substitutes “generally speaking” for “always.”
Sure, but it’s difficult to be sufficiently precise at all times. It’s rude to seize upon an inprecision to dismiss an argument while ignoring the main thrust of the argument.
I’m trying to make the point that its easy to jump on (especially glaring) imprecision. Your general thrust is weakened, often unfairly, by its presence. It can be a bummer for an argument if people jump on imprecise things, but hopefully you can stop that before it happens by omitting them in the first place.
I agree. But at a certain point, you have to rely on the other fellow to be reasonable in interpreting what you say.
To illustrate, it takes a lot of time and effort to formulate something like this:
It’s a lot easier to simply say “the sky is blue.” Any reasonable person understands what you mean.
This seems like the perfect place for the person making the claim about student loans to make a concession (demoting their “always” to “almost always”) thus making their debating partner more comfortable to listen to the meat of their argument; but it is also necessary not to take that demotion from “always” to “almost always” as defeat of the entire argument.
I basically agree, and in this situation my response would be something like “I concede that not all law schools charge ridiculously high tuition but I think my basic point stands.
Sometimes that sort of precision adds too much length. If you see an easily-fixed problem with an argument, it behooves you to point out the fix in the same comment as the problem.
I think that is fair. That would be the reasonable thing to do in a debate.
Precision in this case is not any longer (i.e. always vs typically). It can at times, but for people down with logic, you’d think always versus there exists, etc. would be a big deal.
Actually in this instance it’s made more precise just by leaving out the word “always”.
Hmm, you don’t think omitting it also implies all schools do it?
No, compare “cats have fur” vs “cats always have fur”.
By the way, please note that I did reply to this question once Adelene asked it. This is because I have Tim Tyler labeled as hopeless, but not Adelene.
Thomas != Tim Tyler
[Relevant Link]
I take it you talk to people and not ideas/arguments?
I do exactly the same; you can’t answer everything, and it’s effective to let who is asking help you decide.
Effective—trying to effect what?
Well, your purpose in replying to comments, which might be a combination of several things including enlightening yourself, enlightening others, or entertainment of yourself or others. Even if your motives were entirely based around yourself and others being less wrong, you would still be wise to look at who is asking when considering whether to reply.
In general I wouldn’t set out such criteria out loud, but if it were me in these particular circumstances I’d be very tempted.