Re: Metarationality (was: JOIN: Alden Streeter)

From: Eliezer S. Yudkowsky (sentience@pobox.com)
Date: Mon Aug 26 2002 - 09:45:16 MDT


Ben Goertzel wrote:
>>If you like, consider it a challenge: Name a nonaccidental success, and
>>I'll show you how it's rational or why rationality lies beneath
>>its surface.
>
> Here is a nonaccidental success:
> The composition of the particular melody characterizing the final chorus of
> the "Ode to Joy" by Beethoven.
>
> Why was it rational to choose this series of notes rather than another one?

Your asking "Why was it rational... to choose" implies that all of the
rationality of this process is attributable to specific thoughts of
Beethoven, which may or may not be the case; "rationality" should be
attributed to an entire physical process and its complete causal history,
including all of Beethoven's evolved adaptations for the auditory sensory
modality.

In this case, while not myself a musician, I would tend to suggest that
the following processes that partially reflect or instantiate
"rationality" are sufficient to explain Beethoven's composition:

1) The evolution of Beethoven's auditory cortex, a process of
evolutionary pseudo-induction (not really Solomonoff induction, since
Solomonoff induction applies equally to an entire dataset, while evolution
weights recent data more heavily).

2) The ability of given melodic sequences to produce given emotional
states as a result of cognitive processes within the auditory system and
emotional system; these processes may have started as "accidental" results
of something else, probably speech prosody, but given the universality of
rhythm in human culture there would tend to be a selection pressure behind
the *universality* of melody, even if the specific melodic correspondences
are essentially arbitrary. This is the reason - which may be viewed as
accidental or nonaccidental, it doesn't really matter - that Beethoven,
imagining a melody that produces a specific emotion locally, thereby
imagined a melody that would be effective in other humans as well.

3) Beethoven's internal processes for tweaking the melody to find one
that achieved the effect he wanted are not all known. But if you name any
specific one, I will undertake to show how rationality underlies it. For
example, if Beethoven consciously tweaks the melody and chooses the one
that "sounds better" under (2), that's obviously a case of rationality.
If Beethoven uses intuitions that have been honed by previous positive and
negative feedback about which melodies tend to work, then the neural
potentiation procedures would implement rationality (or if you claim they
don't, I'd like a specific counterexample). In particular, the
"rationality" they implement is the exploitation of a regularity in
reality whereby what worked on one occasion is likely to work again; this
potentiation procedure is in turn the result of evolution exploiting a
regularity in what kinds of learning procedures are likely to work.

4) The fact that you selected Beethoven, in particular, is because
Beethoven composed a very good melody; Beethoven may have succeeded
"accidentally" in some ways, i.e., due to the random selection of a
variable whose particular value happened to have a large payoff, but the
fact that Beethoven's work is currently better-known is nonaccidental; it
is the result of the selection process for fame, which is nonaccidentally
correlated (though not perfectly correlated) with the strength of
emotional impact repeatably produced by Beethoven's music.

5) Beethoven may have been influenced by other musicians, selected by
(4), employing (1) (2) (3) and their own versions of (5). Again,
rationality is a property of a process's complete causal history.

Those are all the factors that spring to mind offhand. I may have left
some out.

> What if I succeeded, and you did not, largely because I were more
> persistent? [Please note: this is just a hypothetical example, I am not
> trying to say that you lack persistence in reality.] In your sense, this
> persistence (a personality trait) would be an aspect of my rationality,
> because it led me to overall effectiveness?

Your persistence would be an aspect of your rationality if:

1) You consciously chose persistence because you had a correct
expectation for correct reasons that this persistence would pay off, i.e.,
you noticed that persistence had tended to pay off in the past, and
hypothesized that it would pay off in the future. (This is a kind of
hypothesis that humans tend to make because, evolutionarily, tending to
make this class of hypotheses pays off often enough for it to be an
evolutionary advantage. Or you might consciously reason about the
similarity of {the past cases of persistence paying off} to the problem at
hand.)

2) Evolution tended to select for persistence - because of an exploitable
regularity in reality, governing both the past cases of persistence being
adaptive, and the present case of persistence being useful to create AI.

-- 
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 - 04:00:40 MDT