Still, it's not like historical geniuses all grew up as pampered aristocrats left to pursue whatever they liked. Many of them grew up as poor commoners destined for an entirely unremarkable life, but their exceptional brightness as kids caught the attention of the local teacher, priest, or some other educated and influential person who happened to be around, and who then used his influence to open an exceptional career path for them. Thus, if the distribution of kids' general intelligence is really going up all the way, we'd expect teachers and professors to report a dramatic increase in the number of such brilliant students, but that's apparently not the case.

Moreover, many historical geniuses had to overcome far greater hurdles than having to chase grants and learn a lot before reaching competence for original work. Here I mean not just the regular life hardships, like when Tesla had to dig ditches for a living or when Ramanujan couldn't afford paper and pencil, but also the intellectual hurdles like having to become professionally proficient in the predominant language of science (whether English today or German, French, or Latin in the past), which can take at least as much intellectual effort as studying a whole subfield of science thoroughly.

So, while your hypothesis makes sense, I don't think it can fully explain the puzzle.

cupholder:

I think it's more likely for the simple reason that what earlier geniuses (like von Neumann etc.) did has already been done. To me, that implies the genius bar has been raised, in absolute terms, at least in the hard sciences and math.

That could well be the case. However, it fails to explain the lack of apparent genius at lower educational stages. For example, if you look at a 30 year period in the second half of the 20th century, the standard primary and high school math programs probably didn't change dramatically during this time, and they certainly didn't become much harder. Moreover, one could find many older math teachers who worked with successive generations throughout this period -- in which the Flynn IQ increase was above 1SD in many countries. If the number of young potential von Neumanns increased drastically during this period, as it should have according to the simple normal distribution model, then the teachers should have been struck by how more and more kids find the standard math programs insultingly easy. This would be true even if these potential von Neumanns have subsequently found it impossible to make the same impact as him because all but the highest-hanging fruit is now gone.

I would bet that the standouts you're talking about would have higher average IQ, but would not actually be 'exceptionally' high, because IQ doesn't correlate that well with success.

Yes, that's basically what I meant when I speculated that IQ might be significantly informative about intellectually average and below-average people, but much less about above-average ones. Unfortunately, I think we'll have to wait for further major advances in brain science to make any conclusions beyond speculation there. Psychometrics suffers from too many complications to be of much further use in answering such questions (and the politicization of the field doesn't help either, of course).

If you tilt your head sideways and look at the top faces simultaneously from below the plane of the top face you'll see that they are the same color (a very dark grey).

By the way, I have spent quite a long time trying to "debunk" the set of ideas around Friendly AI and the Singularity, and my conclusion is that there's simply no reasonable mainstream disagreement with that somewhat radical hypothesis. Why is FAI/Singularity not mainstream? Because the mainstream of science doesn't have to publicly endorse every idea it cannot refute. There is no "court of crackpot appeal" where a correct contrarian can go to once and for all show that their problem/idea is legit. Academia can basically say "fuck off, we don't like you or your idea, you won't get a job at a university unless you work on something we like".

Now such ability to arbitrarily tell people to get lost is useful because there are so many crackpots around, and they are really annoying. But it is a very simple and crude filter, akin to cutting your internet connection to prevent spam email. Just losing Eliezer and Nick Bostrom's insight about friendly AI may cost academia more than all the crackpots put together could ever have cost.

Robin Hanson's way around this was to expend a significant fraction of his life getting tenure, and now they can't sack him, but that doesn't mean that mainstream consensus will update to his correct contrarian position on the singularity; they can just press the "ignore" button.

"You may argue that the extremely wealthy and famous don't represent the desires of ordinary humans. I say the opposite: Non-wealthy, non-famous people, being more constrained by need and by social convention, and having no hope of ever attaining their desires, don't represent, or even allow themselves to acknowledge, the actual desires of humans."

I have a huge problem with this statement. This is taking one subset of the population where you can measure what they value by their actions, and saying without evidence that they represent the general population whom you can't measure because resources limit the ability of their actions to reflect their values.

You are assuming that the experience of being rich or being famous doesn't change ones values.

I suspect that the value of reclusion for instance is a direct result of being so famous that one is hounded in public, and that a relatively unknown middle class male wouldn't place near as much value on it.

A few problems with that. First of all, anyone actually paying attention enough to think about the problem of determining primality in polynomial time thought that it was doable. Before Agrawal's work, there were multiple algorithms believed but not proven to run in polynomial time. Both the elliptic curve method and the deterministic Miller-Rabin test were thought to run in polynomial time (and the second can be shown to run in polynomial time assuming some widely believed properties about the zeros of certain L-functions). What was shocking was how simple Agrawal et al.'s algorithm was. But even then, far fewer people were working on this problem than people who worked on proving FLT. And although Agrawal's algorithm was comparatively simple, the proof that it ran in P-time required deep results.

Second, even factoring is not believed to be NP-hard. More likely, factoring lies in NP but is not NP-hard. Factoring being NP-hard with P != NP would lead to strange stuff including partial collapse of the complexity hierarchy (Edit: to be more explicit it would imply that NP= co-NP. The claim that P != NP but NP = co-NP would be extremely weird.) I'm not aware of any computer scientist who would be so sloppy as to make the statements you assert are often heard.

Overall, Agrawal doesn't compare well to the use of the heuristic here because Agrawal's method (a generalized version of Euler's congruence for polynomials) was an original method.

That said, I agree that such a heuristic can lead people seriously astray if it is applied too often. As with any heuristic it can be wrong. Using any single heuristic by itself is rarely a good approach.

Let me offer a real life example where a version of this heuristic seems valid: Fermat claimed to have a proof of what is now called Fermat's Last Theorem (that the equation x^n + y^n =z^n has no solutions in positive integers with n>2). This was finally proven in the mid 90s by Andrew Wiles using very sophisticated techniques. Now, in the 150 or so year period where this problem was a famous unsolved problem, many people, both professional mathematicians and amateurs tried to find a proof. There are still amateurs trying to find a proof that is simpler than Wiles, and ideally find a proof that could have been constructed by Fermat given the techniques he had access to. There's probably no theorem that has had more erroneous proofs presented for it, and likely no other theorem that has had more cranks insist they have a proof even when the flaws are pointed out (cranks are like that). If some new individual shows up saying they have a simple, elementary proof of Fermat's Last Theorem, it is reasonable to assign this claim a very low confidence because someone would have noticed it by now. Since so many people (many of whom are very smart) have been expressly looking for such a proof for a very long time, we can be pretty sure that if such a simple proof existed it would have been found by now.

The "somebody would have noticed'" heuristic thus functions like many other heuristics. In some cases the heuristic will fail. And the heuristic will likely fail more frequently in situations like Wednesday where the individual is either ignorant or surrounded by people who make basic mistakes in rationality. But properly used, the heuristic can still be useful and reliable.

Chess World Champions are sometimes notoriously superstitious, you can still rely on the consistency of their chess moves.

I'd love to do a real money prediction market. Unfortunately western governments seek to protect their citizens from the financial consequences of being wrong (except in state sponsored lotteries… those are okay), and the regulatory costs (financial plus the psychic pain of navigating bureaucracy) of setting one up are higher than the payback I expect from the exercise.

In the "moral values" domain, you're more likely to have discontinuous rules (e.g., "X is always bad", or "X<N is acceptable while X>N is not"), and be performing logical inference over them. This results in situations that you can't solve directly, and it can result in circular or indeterminate chains of reasoning, and multiple possible solutions.

This line of thinking is setting off my rationalization detectors. It sounds like you're saying, "OK, I'll admit that my claim seems wrong in some simple cases. But it's still correct in all of the cases that are so complicated that nobody understands them."

I don't know how to distinguish moral values from other kinds of values, but it seems to me that this isn't exactly the distinction that would be most useful for you to figure out. My suggestion would be to figure out why you think high IC is bad, and see if there's some nice way to characterize the value systems that match that intuition.