The dangers of genuine ignorance (was: Volitional Morality and Action Judgement)

From: Eliezer Yudkowsky (sentience@pobox.com)
Date: Wed May 26 2004 - 17:45:48 MDT


Ben Goertzel wrote:

>> Feel free to explain how a realistic and frightened medieval alchemist
>> can convince a hopeful, cheerful, vaguely mystical medieval alchemist
>> that there is no way to concoct an immortality serum by mixing random
>> chemicals together. Bearing in mind that the first alchemist has to
>> drink whatever the second alchemist comes up with. Bearing in mind
>> that there is in fact no way to do it, and that being ignorant of this
>> does not change Nature's law in the slightest.
>
> The problem is that the realistic and frightened medieval alchemist
> actually has no way of knowing that concocting an immortality serum is
> impossible. We know that NOW because of the science we've accumulated,
> but no one could know that in the medieval times. Alchemy was not as
> stupid then as it is now. Which is why there were more alchemists then
> than now.... Alchemy is idiotic only in hindsight, which is why someone
> as brilliant as Isaac Newton was an avid alchemist.

So the second alchemist, having triumphantly proved that he is genuinely
ignorant of the difficulty, drinks his immortality potion and dies. What's
wrong with this picture?

This is why it is important never to place yourself in a situation where
you have something to gain by proving your ignorance.

If the realistic and frightened alchemist happens to be a Bayesian
rationalist with a history of science to study, it's straightforward enough
in hindsight (once you realize the danger exists) to notice that nobody has
any legitimate reason to expect an immortality serum to pop out of randomly
mixing chemicals. Or think that the difficulty of mixing an immortality
serum might be comparable to the difficulty of building an airplane. Or
compare heartwarming mystical thinking about immortality serums to
heartwarming mystical thinking about vitalism. Or to observe that
present-day alchemy is a field in chaos, with many facts known, but the
generalizations mystical, heartwarming, and unpredictive. The mistake is
obvious in retrospect and once I caught myself at it - my previous self's
triumphant ignorance of the Singularity or the nature of consciousness -
the mistake also becomes obvious looking forward.

No, Isaac Newton didn't see the stupidity of alchemy. Isaac Newton had a
much shorter scientific history to study. Newton did not read about
evolutionary psychology, vitalism, Bayes (Laplace, actually), or Thomas
Kuhn. I am sick and tiring of hearing about allegedly smart people who
made mistake XYZ. Isaac Newton was irrational? Fine. Do you suppose that
no one is ever allowed to do better, lest it diminish the luster of sacred
Newton's name? Humanity has moved on, and today's geniuses may aspire to
higher standards. It's not like Isaac Newton applied his genius to the
mathematics of rationality. I have more shoulders to stand on, and I can
do better than that.

Rationality is not an innate talent, though it depends strongly on innate
talent. Training the innate talent requires knowledge humanity has only
recently discovered, not known to Newton's day. And learning the art of
rationality requires changing yourself as a person, an inconvenience to
which few scientists are willing to subject themselves. It is so much more
fun to leave the map blank, for then you can draw in the heartwarming lands
you would like to see.

If alchemy was less visibly stupid in medieval times, it is because there
was no history of science to tell worried alchemists the folly of
heartwarming ignorance, or the improbability of specific complex miracles,
or that Nature does not need to convince you of the danger before She is
allowed to kill you. Today that is not a valid excuse.

>> There is too much science. Funny, how the people asserting the
>> ignorance of science on some subject are so rarely specialists in that
>> *particular* field...
>
> I am a specialist in the field of dynamical systems (at least, I was a
> few years back, I published in the field, knew all the literature,
> etc.). So I think I know basically what is known about the attractors,
> terminal attractors, invariant measures, etc. of complex systems. And
> it ain't nearly enough to tell us anything about the attractors etc.
> that complex self-modifying AI's will get into.

Interestingly enough, my "too much science" quote was cribbed from an
unpublished work which reads:

"How could anyone possibly state with confidence that science did not know
a thing, if he were not a specialist in that field? And even then, outside
specialists might know detailed technical answers to questions that you
fancied untouched mysteries of your own field. I had lost track of the
number of papers I had read, offering up as mysteries questions another
field had solved. And when someone gave their favorite mystery as proof of
humanity's ignorance, more commonly than not there would be entire journals
devoted to the answer, a depth of technical knowledge that stretched back
for decades, international conferences and research institutes. There
might be multiple fields of science devoted to separate parts of the
question. Science was a huge edifice, already vastly more than any lone
human could absorb, and it kept on accumulating. No one could say what
science did not know. There was too much science. One needed to train
oneself to always ask if a mystery might already have a known answer,
rather than incredulously saying, 'But how could anyone know that?' For
seven-eighths of the time, someone would. That was what it meant to live
in a world of billions of individuals, rather than a hunter-gatherer tribe
of two hundred."

In this case the math of dynamical systems (of which I do know a little) is
simply inappropriate. The appropriate math is Bayesian probability theory,
and expansions of expected utility theory.

-- 
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 - 04:00:47 MDT