= 27d567bc2d48d9f40e9ccc20ea370b15)
shokwave comments on Harry Potter and the Methods of Rationality discussion thread, part 4 - 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 (649)
I think in this post when you say 'trust' you really mean 'predict'. A trivial counterexample: the more rational someone is, the more I can trust them to be free of errors in their reasoning. And it IS easier to predict a religious zealot staying religious, or predict that a bigot will remain bigoted, than it is to predict a rational agent attempting to maximize their utility (especially if you're an obstacle to their utility).
Well, yes, if there was some shortcut that gave the mostly-optimal answer, or gave the optimal answer most of the time, and gave it in a significantly faster time than optimal rationality. The common example is, I think, reacting to the presence of a lion. Abject, heart-pounding, run-for-your-life terror is not optimally rational (it generally precludes climbing a tree) but it gives a mostly-optimal answer in a much shorter time than attempting to reason out the optimal course of action.
This gets to one of the Hard Problems, both for FAI and a great deal of life. How can you tell who can be trusted to do a good job of taking your interests into account?
But what if our irrationalities aren't quick-and-dirty heuristics optimized for speed? What known cognitive biases are even applicable to running away from a lion?
What if some of our cognitive biases are evolved adaptations that make human society work better? It would be pretty surprising to me if this weren't the case!
Just because they are evolved, doesn't mean they are optimal. An evolved adaptation can be just as "dirty" as a fast cognitive heuristic; the architectural constraints of learning through genes can be just as constraining as those of coming up with something to do fast.
Yes, and let me add to that, just because something was adaptive when humans evolved doesn't mean it is at all adaptive now. To use a concrete example, the weight humans put on anecdotes is likely connected to the fact that in our ancestral environment, that was the primary source of data about what the risks around us were. However, now this leads to silly things like people being terribly scared of shark attacks precisely due to the rarity of such attacks making them get a lot of news coverage.
I've put forward a hypothetical, not claimed a proof. What's the point of responding, "But that isn't necessarily the case"?
You know, you're right. I was responding to peripheral aspects of your proposal rather than central ones, which is a waste of everyone's time. My apologies.
So, OK... rolling back: if I'm understanding you, you're hypothesizing that our biases are not design flaws, but rather adaptations to obtain the group-level benefit of having individuals be more irrational and therefore predictable.
(Is that right? I'm trying to infer a positive claim out of a series of questions, which is always tricky; if I've misunderstood your hypothetical it might be helpful to restate it more explicitly.)
Perhaps irrationality does provide a group-level benefit, as you suggest. For example, maybe it's easier to get valuable group behaviors by manipulating irrational people than by cooperating with rational ones. That doesn't strike me as too plausible, but it's possible.
Even granting that, though, I have trouble with the idea that the benefit to individual breeders exceeds the costs to the individual of being more easily manipulated by others.
ERROR: POSTULATION OF GROUP SELECTION IN MAMMALS DETECTED
Speech seems like an evolved adaptation that makes human society work better.
Why are people voting Tim's comment down so hard? Are there actually three people out there, let alone a majority of LWers, who do not believe it is correct?
I was just thinking how there's a weird hivemind thing going on with the downvotes. Well-written and cordial posts arguing against the site's preferred positions are being summarily downvoted to invisibility.
This doesn't look like a very healthy discussion dynamic.
I haven't seen any recent examples of this recently (since the last times cryonics evangelism was considered, of course.) I suspect that instead you do not recognise the kinds of error in reasoning that have been detected and responded to.
I have been using the Kibitzer since I started posting, and my handle on this matter is that well-written, cordial posts that don't use LW techniques are downvoted. That is, they argue against the preferred position, and they are downvoted because they argue badly. Small corroborations: the posts that get summarily upvoted are ones that point out lack-of-rationality in the arguments, upvotes on topics when they aren't flawed.
If that seems like an unhealthy discussion dynamic then you should review the LW techniques for rationality and make a top level post explaining how using these techniques, or how requiring everyone to use these techniques, could result in unhealthy discussions.
Possibility: Well-written, cordial posts are your criteria for upvotes because cordiality and well-writtenness usually correlate with clear thinking and good reasoning. This is true over most of the blog, except for the edge cases. These cases have their roots in subtle cognitive biases, not gross emotional biases, and it's possible that lack of writing skill and cordiality points out gross emotional biases but not subtler ones.
I think I feel the problem is more a mismatch between the subtlety of the problem and the bluntness of the tool. Downvotes are a harsh and low-signal way of pointing problems in arguments, and seem more suited to punishing comments which can be identified as crap at a glance. Since this site isn't doing the free-for-all comedy club thing Slashdot and Reddit have going, I'm not sure that the downvote mechanic quite belongs here to begin with. Users posting downright nonsense and noise don't even belong on the site, and bad arguments can be ignored or addressed instead of just anonymously downvoting them.
And yes, this probably should go to a toplevel post, but I don't have the energy for that scale of meta-discussion right now.
Downvoting mechanism is one way of making sure that obvious nonsense-posting gets visibly and quickly discouraged. Without it, there would be more nonsense.
I don't think that's actually true. There are very few nonsense posts (or at least, very few that get voted down); and when there are, downvoting doesn't always discourage the poster. When I see a post with a negative score, it's more often one that is controversial, or that disagrees with LessWrong dogma, or that was made by someone unpopular here, or that is in the middle of a flamewar between two users, or that is part of a longer conversation where one poster has triggered an "omega wolf" reaction from the rest of the pack by acting conciliatory.
Downvoting wrong comments may be harsh for the person being downvoted, but hopefully in the long run it can encourage better comments, or at least make it easier to find good comments.
There may be some flaws in the karma system or the way it's used by the community, but I don't see any obvious improvements, or any other systems that would obviously work better.
Look at mwaser: he complains a lot about being downvoted, but he also got a lot of feedback for what people found lacking in his post. Yes, a portion of the downvotes he gets may be due to factors unrelated to the quality of his arguments (he repeatedly promotes his own blog, and complains about the downvotes being a proof of community irrationality - both can get under people's skin), which is a bit unfortunate, but not a fatal flaw of the karma system.
I've never made the claim that the downvotes are "proof" of community irrationality. In fact, given what I believe to be the community's goals, I see them as entirely rational.
I have claimed that certain upvotes are irrational (i.e. those without any substance). The consensus reply seems to be that they still fulfill a purpose/goal for a large percentage of the regulars here. By definition, that makes those upvotes rational (yes, I AM reversing my stand on that issue because I have been "educated" on what the community's goals apparently are)..
I am very appreciative of the replies that have substance. I am currently of the opinion, however, that the karma system actually reduces the amount of replies since it allows someone to be easily and anonymously dismissed without good arguments/cause.
I did. The feedback that actually told him something came as replies. I'm not seeing how the use of downvotes actually helped there, and it did seem to add unnecessary nastiness to the exchange.
The kibitzer does nothing to protect people from groupthink.
What exactly do you mean by groupthink? Let's taboo the word a bit:
Those last two are important parts of groupthink. Without that last one, mathematicians are guilty of groupthink, because they all apply the same (somtimes flawed) processes and get the same answers. Maths isn't groupthink because attempts are made to discover and fix flaws, and these attempts aren't ignored out of hand.
The kibitzer blocks out names and karma scores; so you can't tell what the group consensus is (either by the person's name; "the community thinks this guy is a troll" or by vote; "-5? this post must be bad"). I follow the same process as everyone else in evaluating a comment, but I don't know if I've gotten the same answer as them. In practice, when I've checked, I do get the same answer, so it satisfies the first two conditions. But is the process flawed? And is meeting the group's consensus more important than fixing these flaws?
That would be a systemic problem that deserves its own top level post.
Speech, sexual selection rituals, sex itself, cooperation in the social insects ... There are many things which seem to require a more complex and subtle narrative for their explanation than the usual simple Darwinian story of a mutant individual doing better than his conspecific competitors and then passing on his genes.
But that doesn't mean that a died-in-the-wool neo-Darwinian needs to accept the group-selection explanation any more than an Ayn Rand fan confronted with a skyscraper has to admit that Kropotkin was right after all.
However, I am taking your implicit advise and dutifully upvoting Tim's comment.
I wasn't suggesting that speech evolved via group selection - just that it evidently did evolve - and so proposing the existence of "evolved adaptations that make human society work better" is not an error.
Tim's comment doesn't say that speech evolved via group selection. It could be that it did not; in that case, Tim's comment would be pointing out that Eliezer was unjustified in calling out a belief in "evolved adaptations that make human society work better" as an error.
I have seen people observe that they tend to be inclined to downvote tim readily, having long since abandoned giving him the benefit of the doubt. (This is not my position.)
Absolutely - when considering what it means in multiple level context which Tim explicitly quoted he is wrong on a group-selection-caps level of wrongness. (I was not someone who voted but I just added mine.)
I thought your (PhilGoetz) post on group selection was a good one, particularly with the different kinds of (subscripted) group selection that you mentioned and mentions of things like ants. But now that I see what prompted the post and what position you were trying to support I infer that you actually are confused about group selection, not merely presenting a more nuanced understanding.
... is spot on.
It surely is an unsympatthetic reading to conclude from: "What if some of our cognitive biases are evolved adaptations that make human society work better?" - that those adaptations did not also benefit social human individuals, and may have evolved for that purpose.
You may note that I took care to emphasize that my reply was to what you were conveying in the context. Phil's comment does postulate group selection. While as a standalone sentence your comment is literally correct I downvoted it because it constitutes either a misunderstanding of the conversation or a flawed argument for an incorrect position.
What is the incorrect position? If you say "that group selection is possible", please state your reasons for being so certain about it.
In any case, my comment does not postulate group selection. It wasn't even on my mind when I wrote it.
I do not. That would be a bizarre position to take (or assume, for that matter). I elsewhere indicated my appreciation for your post on the subject, with particular emphasis on an example you gave where group selection does apply. My support does not extend to the position your comment here conveys and I instead (obviously) repeat Eliezer's objection.
(Equally obviously there is nothing to be gained by continuing this conversation. It is based on nothing more than what meaning some unimportant comments convey and whether or not people have cause to accede to your demand (implied request?) to up-vote Tim.)
Thanks for clarifying that. Not just an unsympathetic interpretation, an innacurate one.
You are reading in too much context. You only have to look at the portion reproduced in Tim's comment. Eliezer asserted that there is no such thing as evolved adaptations that make human society work better. Tim provided an example, proving Eliezer wrong.
If you think I'm confused, try to say why. So far, no one has presented any evidence that I am "confused" about anything in <EDIT>the group selection post</EDIT>. There is some disagreement about definitions; but that is not confusion.
Close, but not exactly correct. My interpretation of what Eliezer EMOTED is that there are no adaptations which evolved because they make human society work better. That would be group selection by Eliezer's definition. Eliezer might well accept the existence of adaptations which evolved because they make humans work better and that incidentally also make society work better.
ETA. Ok, it appears that a literal reading of what EY wrote supports your interpretation. But I claim my interpretation matches what he meant to say. That is, he was objecting to what he thought you meant to say. Oh, hell. Why did I even decide to get involved in this mess?
I believe this to be correct representation of Eliezer's meaning and that meaning to be be an astute response to the parent.
Even though I wrote the parent, and already told you that's not what I meant?
Claiming that the parent invoked group selection means claiming that human societies can't evolve adaptations that make society work better except via group selection. Claiming that the parent should thus be criticized means claiming both that, and that group selection is not a viable hypothesis. Tim provided a counter example to the first claim; my later post on group selection provided a counterexample to the second.
Using "because" on evolution is tricky -- particularly when co-evolution is involved --and society and humans are definitely co-evolving. Which evolved first -- the chicken or the chicken egg (i.e. dinosaur-egg-type arguments explicitly excluded).
Error: Most of human history is a recounting of group selection in humans. Every time one group of people displaces another group by virtue of superior technology or social organization, that's group selection.
Having a belief in, or at least openness to, group selection, is one of my rationality tests.
In related news, this weeks Science has the clearest demonstration of group selection that I've seen: The ability to self-pollinate in plants gives individuals a great reproductive advantage; but also increases the likelihood of the entire species going extinct. The presence of a feature (self-pollination) that provides an advantage to the individual, provides a disadvantage to the species, that causes species-level selection.
That is one definition of "group selection". However, there is another definition - according to which "group selection" must refer to a different theory from "individual selection" - a theory that makes different predictions. For that you would need to show that the genetic traits that led to technological mastery benefited groups in a way that was systematically different from the way that they benefited the individuals that composed those groups.
I think it suffices to show that selection can operate at the level of the group. Even if all of the traits involved provide some advantage to individuals, if they also provide an advantage to the group, then group-level selection needs to be considered.
It is more interesting if you can show that a trait that does not confer an advantage to an individual, has an effect on group selection. But it is an unreasonable bias to demand that group selection requires traits that do not provide any advantage to an individual, and yet at the same time not insist that the theory of individual selection requires traits that do not provide any advantage to the group.
I should clarify - "group selection" connotes what Tim is describing: Selection for altruistic traits in individuals, by selection at the group level. That's because, historically, group selection has been invoked only to explain things that individual selection can't.
However, this has led to people excluding selection at the group level from models and simulations, because "group selection bad".
This is something of a quibble, but you really shouldn't think of species-level selection as a kind of group selection. In both group and individual selection, it is the species that evolves. But in species-level selection, the species does not evolve. It is selected - it either lives or dies.
Another key difference - the usual argument against group selection is that it is ineffective since individual selection is a stronger force. That is, individual selection is said to push harder and change the species more than does group selection. But comparing species-level selection and individual selection, it makes no sense to say that one is more powerful than the other. They are playing different games.
Sorry, but I think this is completely wrong. Species-level selection isn't "like" group selection. It is group selection. In group selection, groups are selected for or against. That is the mechanism for group selection. That is the mechanism for group initially described by Darwin in chapter 4 of Descent of Man, and defended by Edward Wilson. It just happens not to be the straw-man depiction used by some opponents of group selection. They chose to ignore selection at the group level because it is easier to rebut group selection if you first assume that it doesn't happen.
Can you provide a reference for that usage?
They are both attempting to influence the same germ line. They are both attempting to influence the same set of traits. It makes reasonably good sense to look at a trait - and to ask whether it is more for the benefit of the individual or the species.
For example, one such trait might be: a love of swimming. That might be bad for an individual (drowning), but good for the species (island speciation).
Just because we are dealing with one individual, that doesn't mean it doesn't evolve. Check with the definitions of the term "evolution" - they (mostly) refer to genetic change over time. You could argue that they also (mostly) talk about a "population" - and one individual doesn't qualify as a "population" - but if you think through that objection, it too is essentially wrong.
Uh, I'm pretty sure I just stated that an individual - the species - does evolve. It evolves by way of organism-level or group-level selection. It just doesn't evolve by going extinct or not.
As for whether one individual qualifies as a population, I've thought about that and completely failed to imagine a population of one individual evolving by way of the standard mechanisms of evolutionary population genetics. That kind of population cannot evolve by differential birth, death, or migration. (I suppose it can change by mutation).
The thing you have forgotten in trying to extrapolate the meaning of 'population' in this way is that the essential feature of a biological evolving population is that its size is not fixed and its membership changes in time, whereas a population of exactly one entity by definition does not change its membership count in time.
Now, I will agree that a population of entities (say, the population of biological species within a genus) can evolve by selection even though its membership count occasionally fluctuates through having only a single individual. The genus does evolve. But the evolution of the genus as a population of individual species and the evolution of the component species as (possibly structured into groups) populations of individual organisms are conceptually distinct processes.
But here is not the place to continue this discussion. If you wish, please bring it up on sbe, and Dr. Hoeltzer can join in. I think he is getting probably lonely over there since we left, and the newsgroup is dominated by John, Tom, and Peter.
"Does NOT evolve" was the term you used. However, with your clarification, it now looks as though this was mostly just a misunderstanding.
IMO, it is mostly OK to think of species level selection as a type of group selection - where the "groups" are species. Maybe there are some meanings of "group selection" for which this is bad - but I would say: mostly OK.
Yes, a population of 1 changes by mutation. Self-directed evolution is an example of that.
If you A) insist on a population having more than 1 member and B) define evolution in terms of genetic change in populations, then the conclusion is that one big organism would no longer be "evolving" when it changed - which I think would be a totally absurd conclusion - a sign that you had got into a terminology muddle.
That may not be a big deal for today's organic evolution - but it makes a big difference for the study of cultural evolution. There, populations with only 1 member are much more common.
My apologies. Most theorists say species selection is a subclass of group selection; but Stephen J. Gould says it is not. See the long explanation here.
That is true if you're talking about features that groups have and individuals don't, or traits that aren't inherited genetically. But all the literature on group selection is about the competition between individual and group (including species) selection, within the same game of selecting genes.
I appreciate the effort you are putting into this, but I fear the terminological and theoretical confusion regarding group selection run far too deep. One enthusiastic person is not going to straighten things out in a forum where evolutionary biology is not the central focus. And now that academian has weighed in, the cause is hopeless. ;)
I agree with you (and Tim) that Eliezer's opposition to group selection was a bit naive and under-informed. But not completely wrong-headed. Many incorrect arguments in favor of group selection have been made over the years. A lot of them were incorrect because they simply did not work. Others were "epistemologically incorrect" because, though they worked, they could be reinterpreted more "parsimoniously" as individual-level selection.
D.S. Wilson's "Truth and Reconciliation" blog series strikes me as an example of extremely dishonest labeling. What he is really saying is that if everyone who disagrees with him would just accept his version of the truth, then reconciliation will take place. And his book "Unto Others" strikes me as even more dishonest. He defines "group selection" extremely broadly, provides examples of corner cases in which his "trait group selection" mechanism works, and then (here is the dishonest part) claims that if group selection works even in this extreme case, then it will obviously work in other cases.
Then he proceeds to discuss the case which every non-professional has in mind when he thinks of group selection - human evolution with groups = tribes, group death = tribe extinction, and group birth = split-up of a successful and populous tribe. The trouble is that the math of group selection really doesn't work in this case.
The only cases I know of where the group selection models do work are (1) Species level selection (Gould/Eldredge), examples like your non-selfing plants; and (2) the examples that Wilson gives in which "groups" are rather short-lived entities which "succeed" by keeping their members alive for a while and then returning them safely to the general population, where the individuals reproduce. A good example of a group that Wilson might use as an example of trait-group selection would be a flock of geese conducting a seasonal migration. Such a group might be selected against if it got seriously lost, or blundered into a tornado, or suffered some other collective catastrophe.
A human hunting party is another example of a "group" such that the mathematics of group selection works. A human tribe of hunter-gatherers is not, unless it is so reproductively isolated from other tribes so as to qualify as a species. I'm pretty sure that this degree of isolation (less than two cross-tribe matings per generation) has never held over any long period of time in human history.
But group selection for cultural traits is another question. If genes get transferred between tribes, but memes do not, then selection at the level of tribes may well help to determine the course of human cultural/memetic evolution.
Well, I seem to have provided you with a long response, which, unlike your own efforts, does not include any links/citations. Sorry about that. You are under no obligation to trust or believe me on this stuff. I will merely assert that I (and tim_tyler as well) have been a serious amateur enthusiast for evolutionary theory for many years. Clearly, you have been too. I do recommend though, that you take a second look at D.S. Wilson's work in light of my criticisms. He really is pulling something of a bait-and-switch. See if you agree.
Thanks for your 2p on D.S. Wilson's Unto Others.
It sounds as though I like multi-level selection a bit more than you do.
Evidence from our own species suggests habitat variation can cause significant morphological differences (despite gene flow) which selection can then act upon.
I also find things like this one interesting:
"Senescence as an adaptation to limit the spread of disease"
Josh Mitteldorf , John Pepper
http://www.mathforum.org/~josh/Epidemics-JTB.pdf
I think so. I'm not quite so purist as Dawkins, but I am pretty close. But I do realize that it is not really an empirical scientific question. It is really simply a matter of what kind of models you prefer. Most cases in which group selection models work can also be explained just as well by individual-level selection or kin-selection.
Speaking of which:
Yes. Very interesting. Red Queen strikes again. But since they are already thinking about Bill Hamilton, why don't they take the further step and realize that the senescent death of an old individual not only reduces the population density for the benefit of the group - the death specifically is beneficial to those individuals in the group who are the most immunologically similar to the deceased.
In other words, this mechanism ain't Red Queen + Group Selection; it is Red Queen + Kin Selection.
Yes: sex and death!
Their model exhibits locality (with limited diffusion - V.N. or 5x5 neighbourhood) as well.
So: a death benefits kin not just through immunological similarity - but also because neighbours are likely to be kin - and death takes an adjacent pathogen load out of circulation.
I realize your views may have changed by now, but isn't that obviously caused by Red Queen effects? Just like all other sexual reproduction?
Hmmm. Presumably there would be no objection had the speculation been worded "evolved adaptations that make people thrive in human society".
Now all I need to do is to figure out whether the meanings of the two are really different.
Be careful.
"Adaptations that make people thrive" can be interpreted in two ways: "adaptations that make people [who possess those adaptations] thrive" or "adaptations that make people [in general, including those who don't possess the adaptation] thrive."
As I understand it, the latter interpretation is essentially equivalent to group selection; the former is not. So it helps to be clear about what exactly you're saying.
Your original formulation ("make society work better") implies the latter pretty strongly. Your rewording is more ambiguous.
In any case, if you are proposing the former -- that is, if you are proposing that some of our biases have evolved to make the individuals expressing that bias more successful -- there's no group selection error, and I agree that it would be pretty surprising if that weren't the case.
Of course, as has been said several times, that doesn't mean those biases currently make individuals expressing them more successful.
If group selection wasn't responsible for naked mole rats, what would be the right term for it? Kin selection seems like too much of an understatement for them.
Okay, not evolved adaptations, but how about culturally/socially imprinted cognitive biases? Something about
clicks with Nancy's comment here.
The more I thought about it ... it seemed like rational agents couldn't trust anyone (the best course is to convince them to trust you and then betray them while never trusting anyone yourself) except in the early and middle stages of iterated games. But a society where everyone irrationally trusted everyone else, and irrationally nobody betrayed anyone else, would be more successful than the 'rational agent' community. (all things being equal; if their irrational trust also caused them to irrationally trust lions...) It might stretch the word evolution too much, but I think the term "competitive selection" applies to this process of societies competing with each other for growth and the most effective societies wiping out the less effective societies (wiping out or completely integrating, as the lesser society's land and resources would already be purposed towards supporting a society, and therefore more desirable than land requiring work).
Basically, what if 'trust' is because a society where everyone trusts the other guy to cooperate in a PD was successful enough to dominate the landscape?
NB: Originally I had thought of trust as a sort of greenbearding. Is there an analogous concept in sociocultural evolution?
A population or society in which everyone trusts completely is not an ESS. A population or society in which everyone adopts the slogan "trust, but verify" and cooperates in the punishment of defectors and non-punishing freeriders probably is an ESS, assuming the cost of verification and punishment are low and verification is reasonably effective.
In the real world, the iteration never completely ends.
Your second quote is ill-formatted.
Cheers!