[bafuture] G-factor versus general intelligence

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Fri Aug 12 2005 - 14:13:19 MDT

J. Andrew Rogers wrote:
> IQ is more of an enabler that bounds the possibility space; no amount of
> hard work is a substitute for being fundamentally too stupid to solve a
> given problem.

So-called "g-factor" - the large correlation that all intelligence tests
exhibit with each other and with real-life measures such as financial success
- measures *relative* intelligence between humans. g-factor discriminates
Einstein versus a village idiot, not between the small cluster consisting of
"all humans everywhere" versus the clusters for chimps, rats, and reptiles.
Chimps can't even take an IQ test - which only goes to illustrate the point.
We've got to distinguish between g-factor load, the correlation between tests
of intelligence within the human species, and human general intelligence,
which makes us capable of challenging problems that other species can't even

IQ tests don't measure what people can do in principle, what they are
fundamentally too stupid or too smart to do. IQ tests measure who does it the
fastest, most reliably, and with the least effort. I suspect that much of the
observed effect of g-factor arises from people specializing in what they can
excel at socially. What's the point of becoming a financial trader if you
can't out-trade the next guy? What's the point of becoming a molecular
biologist if you'll never be the first to invent a new theory? But that
doesn't mean that you'd never *ever* have gotten the answer, if you could have
studied all the evidence quietly without competition. If you expect to
straggle in last in the race, you never study the field - in which case you'll
*really* never get it.

So I have to question the term: "Fundamentally" too stupid? Really,
fundamentally too stupid? A chimp is fundamentally too stupid but is it
likely that a non-brain-damaged adult human would be *fundamentally* too stupid?

Complex adaptations are necessarily universal within a species. Suppose an
adaptation with ten necessary parts: if each of the genes are independently at
50% frequency in the gene pool - each gene possessed by only half the
organisms - then on average only 1 in 1024 organisms will possess the full,
functioning adaptation. Thus the ten parts must be driven to
near-universality before substantial selection pressure exists for an eleventh
part that depends on the other ten. Complex, interdependent machinery is
*necessarily* universal within a biological species; it cannot evolve
otherwise. One robin may have smoother feathers than another, but they will
both have wings. Natural selection, while feeding on variation, uses it up.
Individual differences between humans are a tiny froth on the surface of a
deep, still water.

I do understand where James Rogers is coming from on this one. In my youth to
earn pocket money, I tutored in algebra an adult student who couldn't keep
track of what exponentiation meant. I tried my best to go back to basics, but
I don't think I succeeded in helping her. I admit I got the impression - and
it was a horrifying thought - that some people, through no fault of their own,
could *never* understand linear algebra or calculus, never ever no matter how
much hard work they put in. Like her math module was just missing. Being
young and poorly trained in rationality, I might have refused to even
formulate such a thought, to contemplate an Earth so horribly unfair and
askew, if I hadn't already been planning to fix it all post-Singularity.

Maybe it takes childhood training to force the relevant area of cerebral
cortex to become a number sense. Maybe if you don't develop a number sense in
childhood you never develop a number sense. Maybe if you can't perform
certain mental operations very *reliably*, the conjunctive probability of
completing an entire chain of mathematical reasoning is too low, and you can't
learn from experience. But I would guess that even total mathematical
incompetents are not actually missing any brain modules and could understand
calculus given sufficient training, time, and effort. Evolutionary biology
*requires* that all the basic machinery that I have is present in them also.
The only differences can be speed, quantity, tilted balances, maybe some minor
surface variations in complex mechanisms, and most of all training.

> Giving a person a high IQ is kind of like giving a person a million dollars.
> A few individuals will do something interesting with it, but most will piss
> it away on trinkets and pointless exercises. Call it human nature I guess,
> but few people exploit the potential of the resources at their disposal.
> Doing something constructive is rarely, if ever, the path of least
> resistance.
> In practice, a very high IQ will not do anyone much good without discipline,
> focus, and ambition. If IQ is not effectively applied to real tasks or if
> one has no real goals to apply IQ to, it will not do a person much good to
> have a high IQ. While discipline, focus, and ambition can be developed to a
> certain extent, many of those properties are undoubtedly controlled to a
> significant extent by genetics and biochemistry. Hell, diet affects these
> characteristics to a significant extent.

Few people study rationality formally. Few people learn probability theory
and decision theory, heuristics and biases, social psychology, evolutionary
psychology. Even among those who conceive of rationality as a learnable
skill, it is not commonly realized that the art of rationality has progressed
to the point of including component maths and sciences, or that rationality is
an exact art - a dance instead of a walk where each step needs to land in
exactly the right spot, neither to the right nor to the left. More people
train their body to unforgiving precision to their mind, because it's easier
and more comfortable to overlook a fuzzy answer than to not notice when you
hit a wooden board wrong. There are many more fifth-dan martial artists than
fifth-dan Bayesians; hell, there aren't even formal schools for advanced
rationality training.

But despite appearances, rationality is no more bound to high IQ than martial
arts is bound to athleticism. People who aren't athletic or who don't enjoy
physical exertion aren't likely to learn martial arts because they won't excel
socially in that forum. But evolutionary biology requires that they have the
same muscular and skeletal anatomy as a ninth-dan judo master, and logically,
the same training ought to work.

I don't conceive of my rationality studies as training my high IQ. I conceive
of my rationality studies as training my human general intelligence. How
could you train g-factor, anyway? It'd be like training a measure of
correlation between performances on muscular tasks. Martial arts doesn't
train correlations or standard deviations from the mean, it trains muscles.

Eliezer S. Yudkowsky                          http://intelligence.org/
Research Fellow, Singularity Institute for Artificial Intelligence

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:51 MDT