learning machines 
Author Message
 learning machines


>> I'm just curious if a machine can ->learn<- without hardcoding
>> everything (in an external file or in it's own source code).

> Sure it can :-)

Mind.Forth (q.v. infra) and its JavaScript Mind-1.1 AI tutorial
version learn basic factual knowledge from human user input --
with a plan for the future implementation of a learning "spiral."

> Knowledge can be viewed as a set of rules, associating events,
> i.e. conditions with results. All these can be formulated,
> constructed and adapted as the Ai system evolves over time,

Right now the Mentifex AI in all its various (a)life forms
evolves and speciates only as the individual coders hack away.

The dy(namic) lan(guage) Dylan or its counterpart may be
necessary if the current AI Minds are to modify and self-
organize their own underlying source code in true evolution.

> given the proper examples and experiences (in ANN this is
> called training, but it looks much more like optimization).
> The more abstract the encoded knowledge, the more appealing
> it appears to our "human" eyes.

Each mentifex-class AI is based on (implements) a theory of mind
that incorporates an "abstract" memory channel full of concepts.

> Of course, except some cruel GP tries, we have not yet found

"Found"?  Eureka!  I have _found_ a way to evolve the
learning of new and more complex grammar structures by
the infant AI Mind -- through a spiral mechanism of
the insertion and deletion of part-of-speech nodes on
syntactic structures as found in Chomskyan linguistics.

> a robust way to evolve the inference engine itself,
> which means that the learning system cannot "mature"
> over time - i.e. the AI system can be adapted to new
> knowledge but not to new ways of aquiring it. [...]

I have also eureka'd a way to display ancient Greek
words (e.g., "to pan" -- "the all") in the "AI4Udex"
bidirectional, hyperlinked index (see link below)
to the AI4U AI textbook now available on-line at
and other venues (or simply browsable online through the AI4Udex).

The AI4Udex may turn out to be more valuable in its own
right than the "AI4U: Mind-1.1 Programmer's Manual,"
because the two-way book-index provides extensive
philosophy-of-mind and AI background material
as hyperlinks for each AI Mind concept in the AI textbook.
Enjoy :-)

A.T. Murray
http://www.*-*-*.com/ ~mentifex/theory5.html -- AI4U Theory of Mind;
http://www.*-*-*.com/ ~mentifex/jsaimind.html -- Tutorial "Mind-1.1"
http://www.*-*-*.com/ ~mentifex/mind4th.html -- Mind.Forth Robot AI;
http://www.*-*-*.com/ ~mentifex/ai4udex.html -- Index for book: AI4U

Mon, 31 Oct 2005 00:38:39 GMT  
 [ 1 post ] 

 Relevant Pages 

1. challenge (playing games: the learning machine)

2. TEACH ME EASIEST/FASTEST way to learn machine language (pc) (please)

3. Python Framework for Learning Machine Challenge

4. CFP: C on Machine Learning

5. machine learning/genetic programming/... frameworks in smalltalk

6. Machine Learning Conference Scholarships

7. Machine Learning Conference Tutorials

8. Machine Learning Conferenc Workshops

9. Machine Learning Conference Call for Tutorial Proposals

10. Machine Learning Conference, Final CFP

11. Machine Learning Conference

12. Machine Learning with Functional Languages


Powered by phpBB® Forum Software