Re: virus: Re: neural nets

Alex Williams (thantos@decatl.alf.dec.com)
Sun, 5 Jan 1997 16:00:01 -0500 (EST)


> Realistically you wouldn't use a Turing machine for anything other
> than proving theorems about computation. I would use a real processor
> like a Pentium and a modern compiler to model a function I know.

Turing machine in the generic sense as opposed to NNets and other such
non-Turing devices; of course the Pentium (or Alpha) on your desk is a
Turing machine itself. A Turing machine can simulate a NNet but the
burden of complexity lies the the other way when NNets try to simulate
Turing machines.

> True, perhaps in conjunction with a genetic algorithm or simulated
> annealing. Backpropagation and other supervised learning techniques
> are becoming obsolete.

Whatever method of encoding you use to have your NNet learn, it'll do
better than a Turing on things you don't `understand' to the level of
the formulaic, or don't have a good model for as yet.

> > That still doesn't mean that there's `computation' going on inside the
> > NNet, just as there isn't really `planning' going on inside a
> > spreading activation agent network. Certain things threshold at
> > certain times and certain actions/behaviours result.
>
> You are using "computation" in a sense I don't recognize. What do
> you mean by it?

Computation in the symbolic sense; there isn't symbolic understanding
or comparison going on inside a NNet structure, some patterns just get
more weight than others, trickle down through the layers and finally
render /this/ output pattern. A given NNet isn't a general-purpose
computing device while a Turing machine is.

> I wouldn't think the memes would need to be reorganized in the process
> of thinking, only act.

But isn't `thinking' an act, itself, even though its only targets are
other abstract entities? Are memes that act on other memes alone
still memes? That's the basic question.