From: Michael Wilson (mwdestinystar@yahoo.co.uk)
Date: Thu Sep 08 2005 - 16:05:42 MDT
I'm not going to jump into this exchange, but I had to shoot down
some obvious silliness;
rpwl@lightlink.com wrote:
> The Ptolemaists would have declared, in exactly the same way, that
> they were going to respond to any challenge by using an infinite
> number of nested epicycles.
You are confusing computing power and description length. Eliezer is
talking about a simple theory using unbounded (not infinite) computing
power. You are talking about a theory with unbounded complexity, which
would be impossible to /state/ rather than just impossible to evaluate.
While Eliezer's tendency to ignore tractability can be annoying, he has
a point; the universe itself (including your prized complex systems)
seems to favour short descriptions and computationally inefficient
implementations.
> No, the attack I am making is that there are an array of subtle faults
> that the Bayesian approach cannot get out of
Which you have yet to describe, or rather have tried and failed to
describe one.
> ******* Quote from a hypothetical CAS theorist *******
>
> If you succeed in doing this - having Bayesian global behavior *and*
> Bayesian local mechanisms - then the folks in the CAS community will
> want to know about it, because you will have produced a system that
> is utterly unique: it will have the same behavior at both global and
> local level. In all our experience, we have never seen such a thing:
> complex systems just don't do that!
So, you've never heard of fractals then? The idea of patterns of
structure and behaviour repeating again and again at different levels
of organisation is utterly novel to you? No wait, you admit that there
are 'recurring general behaviours' in complex systems, and you admitt
that complex systems can be nested, so you are just being
self-contradictory. A Bayesian AGI probably won't meet your definition
of Complexity anyway, being engineered to maintain hard constraints
and specific behavioural attractors, but that's almost beside the point.
> It might well be true that your core Bayesian ideas are that
> predictable, but that means nothing until you say exactly how a real
> computing system (not some fantasy with infinite computing power!)
> would implement a full AGI,
Here at last I have some sympathy for your position. It's true, neither
Eliezer nor I have specified how to actually build a tractable Bayesian
AGI. But then, you haven't specified how to build anything at all,
tractable or otherwise. As I see it the onus is on you to specify
something first.
> If you don't accept that an extended version of your Bayesian AGI,
> in the sense just described, would actually need any Complex Adaptive
> low-level mechanisms in the extended portion, then show us such a system
> that actually works.
Obviously we're working on that. In the mean time, you can't demonstrate
a Bayesian system that shows 'Complex adaptive' behaviour. Ignoring the
possibility that Bayes just doesn't scale at all, how are you managing
to claim a predictive accuracy better than indiference as to the outcome?
Indeed I can quite easily state that no causally chaotic system is
stable under self-modification, and that all such systems will rapidly
disintigrate or fall into a causally clean attractor on gaining the
ability to self-modify. If you don't accept that an AGI based on 'Complex
Adaptive' low-level mechanisms will inevitably fall into a 'non-Complex'
attractor, then show us such a system that actually works. In the mean
time I can declare victory with the same apparent level of success that
you can.
* Michael Wilson
___________________________________________________________
How much free photo storage do you get? Store your holiday
snaps for FREE with Yahoo! Photos http://uk.photos.yahoo.com
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:52 MDT