C[omp]ute

From: "andrew cooke" <andrew@...>

Date: Thu, 15 Feb 2007 13:19:49 -0300 (CLST)

My (possibly somewhat limited :o) understanding of Kant is that he is
trying to bootstrap an epistemology by trying to find a self-consistent
(and necessary) set of requirements.  And in general terms he concludes
that you can't get anywhere without:

- An outside world
- Perceptions of the outside world that allow structured interpretation -
Principles that impose structure on perceptions
- Imagination that uses principles to impose structure on perceptions - A
persistent consciousness to provide temporal coherence

To me, one reason that this is interesting is that it suggests how we
might structure intelligent machines.  And, indeed, the parallels with
"old fashioned" AI are clear (and no doubt indebted to Kant).

This leads to thinking about people as "Kantian machines" - machines
structured following the principles outlined above.  When we do this one
thing we can look at is the jump from outside world to perceptions.  This
is the work of "detectors" - the retina, for example.

These detectors can be noisy.

Now Kant seems to argue that perceptions have no inherent structure
themselves.  The structure is imposed by our imagination, which applies
certain principles.  But presumably the outside world is itself structured
in a way that makes our principles effective, and so perceptions do have
some kind of latent structure.  So we might be able to distinguish between
noise and "genuine" perceptions because only the latter match our
principles.

But we know from experience - particularly if we have a scientific
training and experience with statistics - that at least some Kantian
machines impose structure on noisy inputs even when no noise is present. 
We know this because it is what people do.  They are unreliable witnesses,
seeing shapes in shadows.

This seems to be a problem for Kant's argument against problematic
Idealists (people like Descartes who argue that we cannot trust our
senses).  Kant's argument is that we must have once experienced the
outside world, or we would not be able to structure our imagined thoughts.
Yet we seem to be able to impose structure on noise.

So there seem to be two choices.  Either the principles that structure our
perceptions are learnt (and so require "genuine", structured perceptions),
or problematic idealism remains possible.  But principles cannot be both
learnt and apriori.

Andrew