I guess there are several attributes you'd like to consider when evaluating
any potential system.
(i) Grain - are agents each a production rule or are they large grain
'intelligent agents'
(ii) If large grain, how much 'intelligent' support in the language is there ?
IMHO, there is a trade off between easy to program (e.g. built-in
coordination techniques, KR, learning etc), with flexibility.
(iii) How real world is it - e.g. industrial strength, bug free, etc
(iv) platforms it's capable of supporting, inc windowing systems.
(v) Are they open or closed. e.g. do they support something like CORBA,
Internet protocols, do they use DARPA's Knowledge Sharing Effort
components.
It appears that every DAI group has it's own local architecures as well
(naturally as they are experimenting).
Some system software you might use as a basis for your system are:
(i) KAPI (eit's comms software for the KSE) (tested - v.nice)
(ii) Oz (a blackboarding system by DFKI in germany - ijcai93)
(iii) GRATE*/ARCHON (Queen Mary's College, London - Jennings et al)
(iv) MACE (classic system)
(v) CLIPS with DYNACLIPS - apparently an industrial strength expert
system language, DYNACLIPS giving MAS capabilities (though not sure about
it's logic
support).
(vi) parochially you could look at agent-K, but it has no UI support, you'd
have to port KAPI and our Quintus Prolog implementation to your machines,
build GUIs and
probably add bits and pieces to support what ever local tasks you want to
do. Good side is that it more or less supports Shoham's original AOP
paradigm, bad point is that I feelthat the AOP is far from being industrial
strength, and unproved in large scale projects. We will be evaulating it by
using it to implement a distributed data-mining system - (my phd thesis
proposal).
(vii) contact UMIST (Manchester, UK) DAI group for details of MADE - which
I've just heard of.
(viii) other names I've heard of: LINDA , Prolog II.
IMHO (2nd time around), it's about time the DAI/MAS community got its act
together and collaborated on one language - like Common Lisp - (but I think
a Logic Programming Language -(Prolog etc), How about CLAS (common lisp
agent system...:-))
or Praglog (Programming Agent Logic :-)). It would have to support
everyones different logics, comms and cooperation paradigms, but still make
it easy to build agents. I'm sure our agent group would join a
consortium...
Cheers,
Winton