Sunday, September 19, 2004

the term "symbolic AI"

There seems to be a basic misunderstanding among many amateurs and students of AI (and shamefully, among some professionals as well) about what "symbolic AI" refers to.

Today there are many AI techniques that employ symbols -- logical AI obviously, AI based on the use of frames and semantic networks (which is distinct from logical AI because they often do not have a clear logical semantics), genetic programming (where one searches through spaces of symbolically represented genotypes), and Bayesian networks (where the nodes are often labeled with meaningful symbols.) Neural networks often do not have anything you could reasonably call a symbol in them, but there are communities within AI that have developed many ideas about how to express structures such as frames and semantic networks in connectionist substrates.

Yet despite this variety I often run into students who lump logical AI, frames, and semantic networks into one category; genetic programming into another; and Bayesian networks and other probabilistic models into a third. There seems to be no special reason for this lumping other than the so-called symbolic vs connectionist debates of the late 80s and early 90s which established this false dichotomy.

If AI is going to become a mature field we need to start teaching students to make finer distinctions -- which means that they will need to be trained more deeply in more subject areas, including all of the above.


Blogger Olmy said...

"However, on closer inspection, the differences between symbolic and subsymbolic systems are more superficial than actual"
"...all can be applied to the same class of inudction tasks, and can be compared to one another both experimentally and analytically. Future research ... sohuld attempt to see beyond the notational and rhetorical differences that divide these paradigms, attempting to understand the relative abilities of each approach rather than claiming at the outset that they are inherently different. Such work may even lead to novel ways of combining the inductive biases of different methods."

P. Langley Towards a Unified Science of Machine Learning Machine Learning Vol 3, No. 4 March 1989.

This problem can now be cast in terms of "causal diversity" and a systematic approch to managing the strengths and weaknesses of different approaches as their "inudctive bias" means they are useful in certain situations but not others.

September 22, 2004 at 10:40 AM  

Post a Comment

<< Home