[extropy-chat] 12,000 IQ and nothing on?
asa at nada.kth.se
Tue Apr 10 18:18:39 UTC 2007
The "12,000 IQ superbeing" discussed in Lee's thread brings up another
interesting issue: what does 12,000 IQ really mean?
Clearly we are not interested in its ability to solve IQ tests, where it
would do 793 standard deviations above us. Even constructing such a test
would require a long chain of ultrabright entities (make a normal IQ test,
callibrate it. Then make a test that is very hard, and callibrate its
scores so that if you get X points on a normal IQ test and Y on this, you
get the same IQ. Then continue the chain upwards). In fact, having these
ultrabrigths around would change the population mean and undermine the
point of IQ scores.
The real meaning of the statement is probably along the lines of having
lots and lots of g, general intelligence. But it is not entirely clear
how to compare that. One could imagine putting various beings into
competition where they play randomly selected games neither has ever seen
before. Their ELO scores would to some extent general intelligence over
the domain of these games. The problem is that that domain may be rather
narrow. The no free lunch theorems show that a general intelligence that
is applicable to all domains does not exist. So we have to select a
reasonable subset, and that is iffy: what maked a "reasonable" game?
Chess? Buying and smuggling drugs in South-East Asia? Winning the war on
terrorism? Coloring maps on Calabi-Yau manifolds? In many of these domains
certain cognitive architectures would be at a total disadvantage (it is so
easy to overload human working memory).
Many domains probably have a kind of understanding limit where a
sufficiently smart mind can understand all relevant aspects of them and
achieve desired goals efficiently. There are also clearly domains where
there is no structure at all and no amount of smarts help. And there is
the usual wonderful mess of intermediate domains where it may be
impossible to tell how close you are to encompassing them (it is more or
less Godel-Chaitin land).
The effort it takes to understand a domain may still be nontrivial even
when you have enough general smarts, and this is why we often prefer to
learn from teachers, textbooks or letting an expert do the job. Sure, I
*could* learn quantum chromodynamics, but when the physicist upstairs
already knows it it makes more sense for me to ask her to solve my problem
than to spend my valuable time on it - we all benefit from division of
labour and comparative advantage. I would bet this is true among the
superbeings too. Why reinvent the wheel all the time when you can
specialize? And that suggests that all that enormous, wonderful general
intelligence tends to be crystalized among them too. One IQ 12,000 being
will know everything worth knowing about clenchirations in armiphlanges
(and a million other subjects), another will be the corresponding master
of usulism as applied to road planning.
If past history is a guide, the number of domains of human activity is
increasing tremendously. Many domains do not appear to be terribly
complex, but there are so many of them. We cannot google them all to do
them. With even greater brains (and posthuman bodies and extraterrestrial
environments) that number of domains would expand enormously. Even if
general intelligence can conquer them all given time and motivation, I
would suspect that there will never be enough time (and economics).
Compared to the range of domains most of us are acting in today,
posthumans might actually appear terribly specialised. But thanks to the
distributed nature of knowledge, if you ever asked the IQ 12,000 being
about usulism, it would probably efficiently get the answer from its
relative so quickly that we would think it knew it all along.
Oxford Uehiro Centre for Practical Ethics
Philosophy Faculty of Oxford University
More information about the extropy-chat