Re: Expert Systems

Hi Carl,

> Actually -- just a quick note from the trenches
> here -- that's not how expert systems work. Expert
> systems are just symbolic systems full of rules,
> hard-coded into them. If you're thinking of
> training, you may be thinking of neural nets.

Sorry, yes I thought they had to be trained
using probabilities in the same way as neural nets.
Thanks for the correction.

I had realised that they were rule based in
some way however and that they were distinct
from neural nets.

Do they in some way involve the idea
of modelling the way that experts
on the subject make decisions about
things? An example I seem to remember
was of an expert system that embodied
the way expert doctors diagnosed heart
conditions from heart beats I think,
so it would involve using heuristics
based on analysing the way experts
do the diagnoses - does that involve
a training step of some form
(if not probabilistic)?

Or if not, how does the expert knowledge
get encoded into the software?

Robert

Heya,

[expert systems]

> Do they in some way involve the idea
> of modelling the way that experts
> on the subject make decisions about
> things?

I think that's accurate, yes.

> An example I seem to remember
> was of an expert system that embodied
> the way expert doctors diagnosed heart
> conditions from heart beats I think,
> so it would involve using heuristics
> based on analysing the way experts
> do the diagnoses - does that involve
> a training step of some form
> (if not probabilistic)?

Yes, medical diagnosis is a classic
example. In fact there's a company
that claims there system is better
than doctors.

> Or if not, how does the expert knowledge
> get encoded into the software?

You're right: it's hard-coded in by human
experts. Expert systems are the ideal
target for people like Searle. Outside
of a few hangers-on like CycCorp, though,
I don't think anyone takes them seriously
in the AI deptartment anymore.

-Carl