[comp.theory and comp.ai.genetic dropped. this is specifically
becoming philosophy of neuroscience...]
Dr Chaos <mbkennelSPAMBEGONE at NOSPAMyahoo.com> wrote in message news:<slrncd2ai8.ssv.mbkennelSPAMBEGONE at lyapunov.ucsd.edu>...
> Empirically, it seems that one may be able to train a small fraction
> of humans to simulate the execution of Lisp code, given a large amount
> of time and additional aids such as pencil and paper.
Yes, still it's an interesting question to me to find out how easy it
is to train a random person to evaluate expressions in a formal
language he has not seen before.
> In other words,
> "yes, but very poorly." Many more humans can swim in water for a
> short time but it is nevertheless clear that we are not an aquatic
> species and we have no mechanism for aquatic propulsion and control.
No, most definitely we are not an aquatic species. But I did not find
the analogy watertight: I mean mental operations, not muscular.
Learning to swim is mostly mental, which is the hard part to learn,
but to my mind it is a little more interesting kind of trait to be
able to acquire a language of *universal computation*, than a set of
specific programs for adaptive buoyancy or whatever.
> Most modern experimental and theoretical neuroscience focuses on what
> humans, and other organisms, can do automatically across all healthy
> members of their species and the neurobiology and the architecture which
> supports it.
Yes, I could agree with that. But I also think that is not a very
promising research agenda.
> Nothing in this enormous field looks mechanistically like anything
> remotely Lispish or appears provides fundamental mechanisms similar to what
> a Lisp implementor might want.
I believe you are actually wrong.
Neural network models of cortical columns seem to provide for
Turing-completeness. You may look for it on the net, there are online
papers.
That is sufficient for the "fundamental mechanisms similar to what a
Lisp implementor might want".
A human *can* evaluate LISP expressions. In my opinion, this is
amazing! How can humans do that at all? I think you'll have a hard
time explaining this from the POV of reactive-systems oriented
neuroscience research.
The hard problem is not what kind of hardware is needed. That is
almost certain, it ought to be Turing-complete, it ought to provide
for a time-space product that is large enough, it should be fast
enough in its elementary operations. Those are all things that we can
immediately know from theory, without consulting to experimental
neuroscience.
The hard problem in my opinion is, what kind of software abstractions
the brain uses to solve some of the problems it solves, such as
learning to evaluate LISP expressions. You won't find any of these
abstractions in the static (hardware) structure of the brain, just as
you won't find any LISP expressions in the static structure of a VLSI
or RAM chip. We need to deduce that from the spike code, but I am
afraid it might not be as easy as written. Hence, my suggestion for a
behavioral technique! (Not something I would usually endorse as you
may well know!!! Maybe, all this idle talk with Longley-bot might
actually prove useful, of course I don't support any of his silly
philosophy; philosophy is not needed to study these problems...)
Comments most welcome,
Regards,
--
Eray Ozkural
