From: Michael Anissimov (michaelanissimov@gmail.com)
Date: Mon Aug 21 2006 - 20:23:14 MDT
On 8/21/06, Philip Goetz <philgoetz@gmail.com> wrote:
> I recently took an IQ test - unfortunately, it was to help someone in
> training, and so they couldn't tell me what my score was - but there
> were hundreds of questions, and maybe 3 of them required thought, and
> the only part of the test on which it seemed possible to lose points
> was a speed-test in copying figures which seemed to have nothing to do
> with intelligence.
Despite numerous claims and anecdotes like the one above, IQ scores
are predictibly valid for performance in dozens of domains, and the
correlation between IQ and performance is much stronger than the
correlation between performance and any other metric. IQ tests are
consistent, as in, if you take the test twice, you will score around
the same as long as you're trying seriously on both trials.
Shane Legg in a post on this list just a few months ago
(http://sl4.org/archive/0603/14358.html), responded to another IQ
doubter as follows:
> I find it kinda funny that {IQ/ objective measures of intelligence} is so
> often hailed around Transhumanist / Technophile circles, when it's so often
> derided as bunkum in psych circles (to which many cog-sci stuff is
> connected}.
"In pop-science books that's the case, and now most of the population
seems to believe it. However if you look at the actual research that's
been done into IQ testing it's solid. Indeed in terms of various metrics,
such as consistency of results and the ability of the test scores to
predict future performance in mentally demanding tasks (e.g. future
academic performance), IQ tests are THE most rigorously validated
and proven psychological tests that exist. They reliably measure
something very important, there's really no doubt about it. "
Shane Legg also mentions that he has collected hundreds of research
papers on this, and you can email him offlist if you want to check
them out.
Undoubtedly the best paper on IQ (g actually) and its importance is
"Why G Matters" by Linda Gottfredson:
http://www.udel.edu/educ/gottfredson/reprints/1997whygmatters.pdf. I
was somewhat skeptical about the power of IQ tests before reading this
paper, but the evidence presented in it is very compelling. If you
are one of the millions of people who doubt IQ, you should read this
paper first, and if you are still doubtful afterwards, then I won't
argue.
I doubt all sorts of personality tests, the Myers-Briggs Type
Indicator in particular, which is probably one of the most condensed
nuggets of pure bullshit in psychology. )See
http://skepdic.com/myersb.html for example.) So naturally I doubted
IQ tests as well... it doesn't *feel* like the questions are important
or difficult, but after looking at the relevant research, it turned
out that they in fact are.
> 1) IQ tests are designed for placing people into one of 3 categories:
> below average, average, or above average. They are neither designed,
> nor able, to distinguish between 2 sigma and 3 sigma above average.
According to much research, this statement is wrong. The only reason
it is difficult to evaluate its effectiveness in the outer limits (4
sigma, not 3 sigma) is because of the scarcity of data points. They
*are* designed to distinguish between 2 sigma and 3 sigma, and do so
with great regularity.
> 2) If a person whose IQ "should" be 180 takes one of these tests, and
> gives an answer different from the one that is considered correct, I'd
> trust the answer of the 180-IQ test-taker more than that of the
> test-makers. The presence of any ambiguous questions containing
> subtleties that the test designers were not aware of, means that the
> most-intelligence test-takers will not have the highest scores.
According to Linda Gottfredson, "complexity is the active ingredient
in IQ tests", and the specifics of the questions are not exceedingly
important. Because the tests are calibrated using mountains of
statistical data, it is not necessary that the test-writers be as
smart as the highest scorers to measure them correctly. This is
hopeful for future AI development instances where we may need to write
intelligence tests for entities smarter than ourselves.
-- Michael Anissimov Lifeboat Foundation http://lifeboat.com http://acceleratingfuture.com/michael/blog
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:57 MDT