Re: comments on AI

Alexander Meystel (alex@cerebrum.impaqt.drexel.edu)
Sat, 26 Sep 1998 21:38:55 +0200 (MET DST)

Dear BISCers:

Lotfi's letters about AI remind us about the need to demonstrate clearly
the pitfalls of its (AI's) existing curriculum and attitudes. One of the
most important issues neglected by AI is the issue of granularity
(resolution, scale).

Soft Computing illuminates the relationships between AI and Control
bodies of knowledge in a specific way. It turns out, that granularity of
knowledge directs ALL CASES of problem solving efforts into the
multigranular (multiresolutional, multiscale) direction.

This leads to a number of interesting and important consequences.

Assuming, a robotic problem is given (which should have
planning/control/actuation/sensing/man-machine interaction intertwined).
It turns out that "planning" appeals to a particular resolution
(granularity), control evokes a higher resolution, its realization boils
down to even a higher resolution of knowledge.

Interesting resemblances can be observed in a multiplicity of practice
related problems. The following properties can be stated for a
multiresolutional (multigranular) analysis of a problem:
1. The degree of refinement (resolution) is growing top-down.
2. The time scale (the "tick" of the clock) is reducing top-down.
3. The degree of dynamics to be taking in account is growing top-down).

Any problem requires addressing at least three levels of resolution:
1. The one to which the minimum model of the system is related
2. The lower resolution level: the one where the goal of the problem is
formulated together with the cost-function of interest and the
constraints of importance.
3. The higher resolution level: the one where the results of the problem
solving are supposed to be implemented, realized, and refined.

This leads to a set of conclusions which can be considered instructive ones.
1. Even if the problem formulation does not require any high degree of
dynamics to be taken in consideration, at the next adjacent level below,
the dynamics should be taken in consideration in a higher degree.
2. Thus, any AI problem embeds the dynamics as the high resolution
explanation either for the parameters or for the rules of action.
3. Thus, dynamics is embedded in any AI problem.
4. Thus, the knowledge in "Control Theory" should be a prerequisite for
the advanced knowledge of AI and should be recommended to any specialist
in AI.

AI community is notorious about its orientation toward decoupling the
planning and control problems (although some outstanding AI researchers
are trying to resist this tendency, like T. Dean, S. Russell, the general
attitude is clear).

In my view, it is linked with underestimating the multigranular character
of Knowledge and the associated problem-solving processes.

Alex Meystel

############################################################################
This message was posted through the fuzzy mailing list.
(1) To subscribe to this mailing list, send a message body of
"SUB FUZZY-MAIL myFirstName mySurname" to listproc@dbai.tuwien.ac.at
(2) To unsubscribe from this mailing list, send a message body of
"UNSUB FUZZY-MAIL" or "UNSUB FUZZY-MAIL yoursubscription@email.address.com"
to listproc@dbai.tuwien.ac.at
(3) To reach the human who maintains the list, send mail to
fuzzy-owner@dbai.tuwien.ac.at
(4) WWW access and other information on Fuzzy Sets and Logic see
http://www.dbai.tuwien.ac.at/ftp/mlowner/fuzzy-mail.info
(5) WWW archive: http://www.dbai.tuwien.ac.at/marchives/fuzzy-mail/index.html