Newsgroups: comp.ai.philosophy
From: ohgs@chatham.demon.co.uk (Oliver Sparrow)
Path: cantaloupe.srv.cs.cmu.edu!das-news.harvard.edu!news2.near.net!MathWorks.Com!europa.eng.gtefsd.com!howland.reston.ans.net!pipex!demon!chatham.demon.co.uk!ohgs
Subject: Re: What is predictability?
Distribution: world
References: <CvAJ3w.1xo@aisb.ed.ac.uk> <33sqde$hth@search01.news.aol.com> <33t54d$t9b@urmel.informatik.rwth-aachen.de>
Organization: Royal Institute of International Affairs
Reply-To: ohgs@chatham.demon.co.uk
X-Newsreader: Demon Internet Simple News v1.27
Lines: 84
Date: Tue, 6 Sep 1994 08:06:43 +0000
Message-ID: <778838803snz@chatham.demon.co.uk>
Sender: usenet@demon.co.uk


What it is that constitutes a predictable system is, I should have thought, one 
of those questions where it really does depend what you mean by.....! It would 
be helpful to have an idea of the field in which you are working and what it is 
that your system is trying to predict.

This said, there is predictability (the rising of the sun in the morning) and 
there is knowing about predictability: "the morning is when Quetzal is 
struggling to be reborn after a period of little death and we must hasten to 
offer him blood such that the sun may rise". Knowing about predictability 
occurs when a model is adduced which appears to be homologous with what is 
observed: the following parameters offer the following C8/C10 ratios from the 
cracker; heart's blood gets the sun up in the morning. Some models are 
apparently homologous but untestable, for whilst we might try Quetzal on 
cornflakes, the outcome would be disasterous if it didn't work. Some models 
are untestable (the universe is run by little green men who manipulate things 
so that you can never truely be *sure* that they are there) but are useful 
to some because they offer comfort.

Models come in two flavours. There are phenomenological models, such as rules 
of thumb, that if you push this then that pops out, that 97% of the 
intercountry variance in energy demand can be explained by their respective 
GNP and average temperature (which it can, astonishingly). This form of model 
blends into the other flavour, which is the clockwork model. This
explains the emergence of the cuckoo in terms of the chain of events which 
make the clock function. Such a model seems complete, until you ask yourself 
why people want this to happen; and why in this way; and what you need to 
know about the differences between people in order to understand why they 
have selected this (infuriating) method of time keeping.

No model is either complete or known to be complete, therefore, and there are a 
variety of proofs that this is so. Technology, however, consists of working 
backward from a (primarily clockwork) model in order to create a physical or 
information-embodied system that is homologous with it. We reverse the nature 
flow; and, as we all know to our cost, spend a great deal fo time fine tuning 
the outcome until it is homologous with our expectations of it. It is, within 
its domain and subject to the predictability of its environment, itself 
predictable.

AI is the discipline which tries to establish a two way flow across this path. 
Natural intelligence takes in data from the outside world and builds an inner 
set of models to explain and to predict; technology (etc) takes elaborated 
versions of these models and imposes them on the outside world. AI tries to 
learn in order to impose - or anyway, sometimes it does. Systems are to learn 
and then to respond adaptively in the pusuit of some set of goals or other. 

The analogy with systems of feedback is compelling. One notes that dynamic 
balances are achieved all over the manmade and natural world. IN some 
instances, such balances are no more than a balance between two 
manifestations of the same thing, such as the pressure and temperature of a 
gas. In others -such as the equilibrium struck over the supply of and demand 
for a good, hugely complex structures (which need many layers of 
interpretation, in the manner of the cuckoo clock, if they are to be 
individually understood) somehow come into dynamic balance and behave in a 
manner which is broadly predictable. Our society is predicated upon such 
predictability - as are our bodies and the ecology in and on which we subsist. 

These systems are purely abstract: a market, a price  - these are not 
reducible to their constituent parts but are independent dimensions which 
have, in a curious way, transcended the model needed to capture what it is 
of which matter alone is capable. One can predict, at least in general terms, 
how the dance of attractors will evolve; but *only* if one has seen and 
understood how it works in its own terms. One looks, one evolves a set of 
more or less sophisticated models and one uses these to look better, with 
understanding and foresight. One may even be able to predict, to a degree. 

It is striking that one would not be able to predict the model itself, 
however, ex ante and with a knowledge base that knew only abou the mechanics 
of cuckoo clocks, atoms and the physiology of the poeple operative in the 
cuckoo clock market place, however. The high level model "emerged" from the 
interaction of the constituent actors, achieved a dynamic balance and created 
something which had not existed before. Noting this new actor, the behaviour 
of people and organisations, consumers and cuckoo clock makers (and no doubt, 
copper miners and wood cutters) minutely shift to take account of this new 
force.

To design something which can achieve understanding of this would be more 
than a triumph. To be able to predict it, ex ante, would be impossible. To 
return to my opening remark, therefore, it depends what you mean by "predict"!

 _________________________________________________

  Oliver Sparrow
  ohgs@chatham.demon.co.uk
