[Newbies] Studying Artificial Intelligence

karl karl.ramberg at comhem.se
Fri Mar 14 16:44:46 UTC 2008

Ryan Mitchley wrote:
> Hi Thiago
> I think there used to be a Squeak Prolog somewhere, but I'm not sure where
> it went or if it's compatible with the current images (I'm pretty new to
> Squeak myself). That could certainly form the basis of an expert system in
> Squeak.
There is a Squeak prolog version here:

> There's a Prolog interpreter written in OMeta
> (http://www.cs.ucla.edu/~awarth/ometa/ometa-js/prolog.html) and JavaScript,
> desribed in the "STEPS Toward The Reinvention of Programming" paper
> (http://www.vpri.org/pdf/steps_TR-2007-008.pdf). It shouldn't be too hard to
> turn those into working Squeak code.
> Of course, vanilla Prolog is lacking some features that would make it useful
> for general purpose AI, e.g. reasoning under uncertainty, notions of time
> and sequence, constraints. I think the idea of classes  simplifies reasoning
> under uncertainty, since they make the dependence amongst rules and terms
> much more explicit (solving the "frame problem" to some extent... i.e. the
> normal problem with Bayesian reasoning is that joint probabilities grow
> exponentially with the size of the system being modeled).
> I hope someday to spend some time implementing a kind of Agent oriented AI,
> where an Agent is defined as an object with goals, knowledge, and means to
> achieve those goals using that knowledge. Learning involves the assertion
> and retraction of meta-knowledge... (I think it was the extremely
> interesting SOAR project that viewed learning as chunking - i.e. learning is
> simply a way of pruning the search tree through the use of inferred
> meta-productions). The reflective and meta-programming features of Smalltalk
> makes collaboration amongst agents much easier - since agents can describe
> their own abilities (I guess methods and behaviours).
> This is, of course, if you're on the symbol-processing side of the AI
> fence... (although, it's interesting to note that neurons and rules have a
> similar N-in, 1-out structure. I like to think of a rule base as converged
> neural network :)
> Anyway, I just need to get rid of my full-time engineering job so that I can
> actually explore these ideas someday!
> Ryan

More information about the Beginners mailing list