From: Tennessee Leeuwenburg (tennessee@tennessee.id.au)
Date: Mon Sep 05 2005 - 19:22:50 MDT
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Eliezer has said he thinks that his approach can capture the
subtleties of human intelligence. I'm inclined to believe that his
approach can capture the subtleties of human deduction, and some
amount of inference, but I have a question :
How about the conceptual framework? aka Cognitive schema, noema. How
is it that human breaks the world into logical objects? Sometimes it's
done in an aristotelian way -- necessary and sufficient conditions,
the principle of conflict. By understanding phenomena as made up of
inter-relating objects, and then working out those objects are on the
basis of whether a candidate object has internal conflict, you can
break the world up into objects.
For example, the coke can and my desk, considered as a single
candidate object, exhibits conflict in that I can move the coke can
away from the desk. Thus it is better to think of them as two objects.
It seems like the human mind performs some kind of ad hoc reduction of
phenomena into objects. These objects, plus our set of beliefs about
relationships (such as beliefs about the relatedness of two particular
objects, or kinds of relation like above, below, or relatedness per
se) make up our tools for reasoning.
How is it that an Eliezer approach (Bayes plus whatever else he cares
to include) can do this phenomenological reduction?
How do our ideas about the phenomena we observe carry into logic
itself? It seems like logical structure is a priori true, but it seems
at least possible that some things like this are just the result of
observation. Anyway, I don't know exactly where all of this is headed,
but I think there's something interesting in there.
If you start with a bad noema, then no matter how good your abilities
to refine rules of relationship are, then you cannot succeed. If you
have a bad information representation, then you may fail to capture
truth. How would you go about performing generalisation under a
Bayesian system?
Cheers,
- -T
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (MingW32)
iD8DBQFDHO9qFp/Peux6TnIRAqG/AKCUWNvwJIXIC8e6nC2KnVV6g7F1FACbBgt2
IdSvbH7bE4g6jZRMU+6brmw=
=qONR
-----END PGP SIGNATURE-----
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:52 MDT