Here's a good background post and analysis on the debate (this has been linked from elsewhere on LW before):

Like many, I couldn't help but be fascinated by the Sam Harris/Ezra Klein debate. These are two people I really look up to, and so seeing them going at it (and showing a lot of personal weakness along the way) has been illuminating. I'm still unsettled about it, wanting there to be resolution/a right answer. So far that satisfaction has eluded me, so I wrote this to try to clarify things for myself. Maybe it helps others too.

The analysis below is meant as a steelman of each side's positions. If you think I'm not steelmanning them well enough, please leave a comment and I'll improve.

Consequentialist framework:

  • Sam: As a lesson in how to think for yourself, hold Murray up as someone who has discovered truths that society doesn't like to talk about. As a general policy, this practice will lead to truths being uncovered faster, leading to a faster pace of discovery, which compounds over time to a much better world through science.
  • Ezra: Make a public example of Sam here, leading to more people recognizing their own privilege and putting their actions in the appropriate historical context. As a general policy, this practice will lead to a more equitable society, which compounds over time to a much better world by reducing suffering directly.

Virtue ethics framework:

  • Sam: It is virtuous to signal-boost things which are true, especially when they are being suppressed by society. "Speak the truth though your voice may tremble"
  • Ezra: It is virtuous to defend the underprivileged by calling out harms, even unintentional harms. "Evil is the silence of the voice of justice when it matters most"

Deontology framework:

  • Sam: Thou shalt update on all available data.
  • Ezra: Thou shalt not invoke long-buried demons of oppression.

Both these moral frameworks look pretty good to me. I see no particular reason to favor one over the other; even if I restrict myself to only looking at the consequences I see plausible arguments that one path or the other is higher-impact.

I have only one unifying thought:

Sam was upset by being attacked by Klein, and considered the attack unfair, which actually triggered the whole debate. I think Sam's complaint about "unfairness" is invalid here, because it is exactly what he should expect when signal-boosting things that society doesn't like to talk about. So perhaps the only error here was Sam getting too emotional about being pilloried.

New Comment
14 comments, sorted by Click to highlight new comments since:

I think you are seriously missing Ezra's point, and i think a lot of people in the rationalsphere that I've read seem to make a similar mistake.

The primary point of Ezra bringing up historical context is scientific, not social. He is essentialy saying that global white supremacy is an enormous confounder to any possible effort to tease signal out of what data we have, and so we shouldn't entertain the idea of changing policies based on something with proper implausibility and extremely weak (at best) evidence.

Ezra's further point is essentially an outside view point: many times in the past scientists have convinced themselves based on now-known-to-be-bogus theories of the conclusion Murray is pushing. Thus, as a matter of good epistemics, we should be looking for good reason to believe this time is different, and there is no such reason.

I think Sam is totally failing to engage with either of these two points and is instead wrongly reading Ezra as making social arguments that the reliable data should not be discussed. Instead, Ezra is giving strong reasons why the data is a supremely unreliable map to the territory. I also think you may be making the same error.

You make Ezra's point so much better than he did. (If that is indeed a good summary of what Ezra was trying to say).

Of course, that's much easier to do while not in the midst of a back-and-forth.

Hmm, was this really Ezra's point as opposed to a steelmanned version? My impression was that he insisted on any discussion of the science also involving a discussion of past prejudice. He also seemed to be against giving Murray a platform because of his policy positions.

Thanks. This is a useful distinction, and I'm not sure yet what it means for my understanding of the arguments, but I'll have to process it and hopefully update my thinking on this matter.

This, so much.

So, in the spirit of learning from other's mistakes (even better than learning from my own): I thought Ezra made his point very clear.

So, all of you people who missed Ezra's point (confounded data, outside view) on first reading:

How could Ezra have made clearer what he was arguing, short of adopting LW jargon? What can we learn from this debacle of a discussion?

Edit: tried to make my comment less inflammatory.

Ezra seemed to be arguing both at the social-shaming level (implying things like "you are doing something normatively wrong by giving Murray airtime") and at the epistemic level (saying "your science is probably factually wrong because of these biases"). The mixture of those levels muddles the argument.

In particular, it signaled to me that the epistemic-level argument was weak -- if Ezra would have been able to get away with arguing exclusively from the epistemic level, he would have (because, in my view, such arguments are more convincing), so choosing not to do so suggests weakness on that front.

(Why do I think this? I came away from the debate podcast frustrated with Ezra. Sam was being insistent about arguing exclusively on the epistemic level. Ezra was having none of it. After thinking about it for a long time, I came to the summary I wrote above, which I felt was more favorable / more of a steelman to Ezra than my initial impression from the debate.)

So, at least to convince me, if Ezra wanted to make the points you are suggesting he make, then he should have stuck to debating Sam on epistemic grounds and avoiding all normative implications.

I think that this is a good example of the problems with the term steelmanning. What is the point of steelmanning? Different people have different meanings, derived from different purposes. You ask if you are "steelmanning them well enough," but I cannot answer that without knowing what you mean and are trying to accomplish.

I see three usages. One is in the middle of a debate. This was the original usage, to "attack a steel man" rather than to "attack a straw man." Fix small holes in your opponent's position rather than making him waste time fixing them himself. More generally, don't reiterate common patterns of back-and-forth; anticipate the response and address it.

A second usage is to ascribe intellectual credit. If someone influenced you, it is good to acknowledge your debts, but take credit for your own ideas. Call it your own "argument" or "analysis," not a steel man of someone else. Failing to do so is an ancient practice and I think that it has always been bad for clarity, such as when stoics attributed all their ideas to Zeno. But the new form has additional problems; the traditional practice was a form of tribute to the master, while the idea of the "steel man" of an opponent reverses the valence. Also, putting words in people's mouths is offensive. (Of course, any debate is a claim of superiority over the enemy, but in practice the term and practice of the steel man is more offensive.)

A third usage is to describe a person or a debate. You can't perfectly represent anything and you should always rephrase things to make sure you mean something, rather than just parrot words. You should patch up small holes in this process. But you shouldn't make radical changes. It is sensible to ask what general principle might the person be following to generate these arguments, but you are probably wrong, so you shouldn't attribute it. Ask your audience to debate that new point, not whether you steelmanned the old point. Ask your audience if you missed other arguments. You seem to say (in the comments) that you consciously noticed that Klein made two arguments and that you completely discarded one as lousy. Surely Klein would not accept this description of himself. If your goal is to recount the debate, you should mention the other argument, if only to dismiss it.

One problem with the second and third is that they blend together. I can't tell which you mean to do, and that affects my judgement of how well you did it. I don't like the term steel man for either one, but that confusion is additional problem.

I think there is an argument that Harris is making serious errors, even if we ignore the historical and social context.

Default Hypothesis — whatever the factors are that influence individual differences in IQ, the same factors would influence average group differences.

This is of course true for any two arbitrary groups. The causal graph looks like this:

Environmental differences -> IQ differences <- Genetic differences

If we had literally no information about the groups, our best hypothesis would be that both environment and genetics play a role. But this is not the case. It is obvious that there are many environmental differences between the groups, and we can safely assume that they have some effect on IQ differences. At this point the hypothesis that these environmental differences explain all of the IQ differences, and that we know nothing about the genetic differences, becomes very likely.

Considering that the variables are not binary, and that the extent to which environment should be able to affect IQ is not known with certainty, we can't deduce that the expected genetic differences are zero, but the expectation should shrink dramatically from the case where nothing is known (I admit, I don't actually know how the math works out in the non-binary case).

Oh man, I really don't want to be on the other side of that debate. But, I swore allegiance to the cause of local validity, and I must uphold that.

Let's use a simplified model: Total intergroup variance= P*(genetic component of variance) + (1-P)*(environmental component). This is very simplified because genes and environment interact, but it will suffice.

Your logic only works if our prior was: 50% that all the difference is environmental (P=0), 50% that all the difference is genetic (P=1). In this case, finding an environmental difference would disprove the existence of a genetic difference.

But that's not what our prior is. The prior for biological group differences in a highly heritable trait is some bell curve of P, with its peak probably somewhere around P=0.5 (or at least neither P=0 nor P=1). The fact that IQ-affecting environmental differences exist only rules out P=1, which maybe changes the posterior expectation of P from 0.5 to 0.45, but not to 0.

After all, our prior was that we would almost certainly find environmental differences that affect IQ, so finding them can't cause that much of an update.

And if we haven't even proven that the environmental differences affect IQ, only that they exist, then we shouldn't update at all. Our prior for that was basically 1: any two groups will have some environmental difference (food, language, location...), so the existence of those differences can't be evidence either way.

I don't exactly know what you're disagreeing with. I did not say that genetic difference is zero.

After all, our prior was that we would almost certainly find environmental differences that affect IQ, so finding them can't cause that much of an update.

This is not a binary question as you seem to treat it. We need a probability distribution over "what difference in average IQ two groups in these two environments would have, if they were genetically identical". And then depending on our beliefs about that, we could deduce that the genetic differences are a significant contributor, or that there is zero contribution, or even that "low-IQ" group has the superior genes.

And if we haven't even proven that the environmental differences affect IQ, only that they exist

The kinds of environmental differences we're talking about are things like poverty, discrimination, etc. I assume that their effect on IQ has been observed, at least for some of them.

Now, I don't actually know what that distribution looks like. It could really be very insignificant. But if Harris does not make a good effort to evaluate this distribution and then update his P("inferior genes") accordingly, then he is making an error.

local validity

Somehow, whenever I see people link to that, I end up unimpressed.


All this discussion seems insufficiently specific to make actual progress.

Obviously there are imaginable states of affairs in which our best estimate of genetic differences should be very close to zero. (E.g.: Observed overall differences small, many environmental effects found and well understood, best estimate of total impact of all known environmental effects very close to observed overall difference.) Call this proposition "P0".

Obviously there are imaginable states of affairs in which our best estimate of genetic differences should be rather large. (E.g.: Observed overall differences large, environmental effects looked for but few found to have any supporting evidence, best estimate of total impact of all known environmental effects much smaller than observed overall difference.) Call this proposition "P1".

And obviously what we should think about this (if we think about it at all) depends a great deal on the details of the actual evidence: how big the various known effects are, how hard researchers have tried to make IQ (or whatever) be insensitive to many environmental factors, how much we trust any given bit of research not to be perturbed by the researchers' prior prejudices, etc., etc., etc.

So far as I can tell, zulupineapple and Jacobian don't disagree with any of that. I haven't looked into the Harris/Klein affair much, but my guess is that they and the other people who have weighed in also don't disagree with any of that.

But so much of the discussion -- including pretty much everything above -- seems to consist, modulo outright mistakes, of one person reiterating P0 and another reiterating P1. So, e.g., zulupineapple points out that there are known environmental factors (so far this is just P0) and then says "At this point the hypothesis that these environmental differences explain all of the IQ differences, and that we know nothing about the genetic differences, becomes very likely". Well, no, at some point that happens but to know whether we're at that point yet we'd need to know more about the environmental differences and more about the observations they're meant to explain. And then Jacobian points out that we have some prior for what the relevant differences are and the effect of learning about environmental factors just shifts our probability distribution rather than ramming it all the way to 0. That's more or less just P1. Then zulupineapple comes back, reaffirms that we should be adjusting probability distributions, and says that depending on our prior we might end up with an estimate of 0. That's just P0 again.

Maybe Harris and Klein are also just reiterating P0 and P1, in which case what needs pointing out is that they need to engage with the details of the evidence. Maybe at least one of them is actually engaging with those details. But in either case, just repeating "P0! P1! P0! P1!" achieves nothing useful.

(I am not saying that anyone ought to be digging further into the evidence and trying to do better than P0 and P1. That might be a bad idea -- one might e.g. decide that the chance of finding anything genuinely actionable is small, and that the most likely outcome is simply becoming a pariah to some fraction of the population for political reasons. But if it's worth getting into this argument at all then I think what's worth doing will have to involve getting more concrete than just P0 and P1.)

No, you're missing the point. I have very little interest in whether P1 or P0 or some P2 is true (and I don't really see Jacob strongly favoring one of those either).

Instead, my bold claim is that Harris is not actually evaluating the environmental effects. He says that he uses the "default hypothesis", but that only makes sense if you have zero information about the environmental differences, which is not the case. Ignoring data you have to make some conclusions look more reasonable is an error.

Now, on the object level, I admit that I favor P0, because the idea of genetic differences doesn't really make much sense to me. Why should they be there? Because only smart people managed to migrate out of Africa? Mabe by pure chance? Also, it seems to me that there are some hard-to-measure environmental factors, so it's not surprising if the measured part of environmental contributions is small. But I actually know very little about this issue.


I think you have misunderstood at least one thing: P0 and P1 are definitely both true. (They are propositions of the form "there are imaginable states of affairs such that ...".)

Once again, I don't know very exactly what Harris and Klein have said, so I can't easily tell what if anything he's doing with the environmental effects. Are you saying that he thinks there are none? (I am not a big fan of Harris, but I would be astonished if he were being that stupid.) If not, it's not clear to me what specific "serious error" you think he is making.

Your original comment doesn't help me here. It observes that there are environmental effects (which, again, surely Harris doesn't deny) and then jumps straight to saying that "the hypothesis that these environmental differences explain all of the IQ differences, and that we know nothing about the genetic differences, becomes very likely" -- and I don't understand where that comes from. I mean, sure, it could well be true that the environmental differences explain everything. But it seems like you jump straight from "there are lots of environmental factors" to "environmental differences probably explain everything" and I just don't see how you can do that without actually engaging with the details of the available evidence.

(Once again, I'm not saying that you should do that. Only that without doing that, I don't see how you can reach the conclusion you say you do.)

Are you saying that he thinks there are none?

Of course not, that would be too easy. He does, at some point, explicitly acknowledge that environmental effects exist and are "very important". And then he proceeds to do nothing with those effects. If you admit that a factor is very important, but then arrive at the same conclusions as someone who believes that the factor has 0 effect, that is still a problem.

I imagine the thinking goes like this. He observes that there are relevant factors X, Y, Z. He notices that Z has a lot of uncertainty, or is hard to work with, or ect. Then he decides that he will look at X and Y in isolation. And then he assumes that whatever conclusions X, Y have lead him to, actually have some bearing on the real world. Depending on Z this may or may not be the case. If it happens that Z really is 0, then you arrive at the correct conclusion, but even then, the path you took to get there is still wrong.

Of course, all of this is very hard to talk about when nobody gives their actual probability estimates (and giving probability estimates is also very hard). In particular this is because the value of P("inferior genes") doesn't seem central to Harris. There is a good chance that I'm misreading him and that his estimate of P("inferior genes") is not as high as I assume it to be, and that he really is honestly updating on the environmental differences.

Still, the claim that Harris is making an error seems more likely to me. At least now you better understand what error I'm accusing him of? (By the way, I'm also accusing him of the error pointed out by RamblinDash, but didn't want to duplicate it).