Re: requirements for systems that learn

casey <jgkjcasey@xxxxxxxxxxxx> wrote:
On May 24, 10:45=A0am, c...@xxxxxxxx (Curt Welch) wrote:

But where's the evidence to support that a very
similar algorithm is inadequate to explain human
behavior? And by very similar, I'm talking some
sort of neural network program running on a normal
computer, which is trained by reinforcement.

Only by running the program and seeing how it works
and what it does, to determine if we have said what we
meant to say, indeed, whether we have said anything
at all with all our high level abstract talk.

Well, that's consistent with what I say about the way you like to think.
If an idea can't be translated into hard code, you seem to think it doesn't
exist, or has no meaning. You always seem to focus on the next bit of hard
code you can write and for the most part, seldom seem to philosophize
abstractly about what the code is missing, or what your next program needs.

This style of thinking seems to be a general personality trait in people.
In the Myers Briggs personality test, it's the S vs N dimension I'm
thinking of. That is, sensing, vs intuition.

People that test heavy on the S end of the scale are very much "show me"
people. They focus on what's in front of them and they prefer to be shown
real examples to understand something. They are very literal and
disciplined people that don't tend to make heavy use of metaphors and
similes. They favor hard facts, over vague theories. They more quickly
spot what is different about two examples, than what is similar about them.
They favor the concrete over the abstract.

People that test far on the N end of the scale, are the opposite. They
favor the abstract, over the concrete. They tend to focus on what's
similar, instead of what's different. They prefer the abstract, over the
concrete.

I test far on the N end of the scale. You I would guess are more to the
middle or towards the S side. I test as an ENTP. I would guess you test
as something like an ESTP? It would be interesting to know how you test as
on an MB type test.

As such, I don't think there's much you are going to follow about my
thought process until I can turn the next idea into some more concrete code
for you to play with.

... but to try an argue the confectionist approach

Do you mean connectionist approach?

That's funny! When I use my spell checker to fix my errors, I really
should trying reading what it's done!

_cant_ explain human beahvior would require supporting
evidence that I've never seen.

I suspect you have never bothered to look for counter
evidence? It would be like asking a creationist to look
at the fossil evidence. Anyway today we can put these
centuries old doctrines of connectionism to the test
with actual programs.

Sadly, not until you have the next algorithm to test - and you can't find
that, by waiting for someone else to show it to you!

JC

--
Curt Welch http://CurtWelch.Com/
curt@xxxxxxxx http://NewsReader.Com/
.

Relevant Pages

  • Re: More math stuff, truth and social reality
    ... > out that I use brainstorming, where you generate lots of ideas during ... fast as other efficient factoring algorithms. ... I don't see evidence of lying, ... algorithm) is in fact the truth. ...
    (sci.math)
  • Re: How to portray yourself as a caring person..
    ... to review and change the guidelines. ... Having evidence of which algorithm you followed is ... Yeah, I guess, although the 'evidence' is kept by the trainer, not the ...
    (uk.rec.motorcycles)
  • Re: Solving k^m = q mod N
    ... otherwise can only be done by brute force. ... posters appear to be taking my algorithm far too seriously. ... reason in his head that it is. ... I made by response on the basis of the evidence. ...
    (sci.math)
  • Re: How to convert Infix notation to postfix notation
    ... If you think the grammar is correct, ... You present no evidence for the first claim. ... means you've never seen this simple algorithm. ... The evidence of the first is the truth of the second. ...
    (comp.lang.c)
  • Re: wav file
    ... Suppose you have developed the floating point algorithm and it works ... well, then whatever Q format you use, just scale the step size ... the smallest the step size is the slower your algo 'll be. ... algo use a power estimation (transform or freq lms) the result is ...
    (comp.dsp)