Volume 14, Number 3
PATTERN RECOGNITION LETTERS
premature promises made to funding agencies. Still, Hecht-Nielsen shares a widely held feeling that research on artificial neural networks was severely set back by Minsky and Papert's book.
In uncertain and ill-structured domains, i.e., in almost all areas of human activity, alternate possibilities in the form of different objectives, different perceptions and different partial information should be systematically considered. That is the motivation underlying the work presented in recent papers by Bhatnagar and Kanal (1993 a,b,c) and Bhatnagar's Ph.D. dissertation (1989). Our inspiration also derives from the Kantian and Hegelian models of enquiry developed in the writings of C.W. Churchman and his students ((Churchman (1971), Mitroff and Turoff (1973)). Instead of having a single domain model on which one performs Bayesian reasoning, we assume that the overall domain model is a forest which has embedded in it a multiplicity of situation models. The dissertation presents a hypergraph-search based methodology for hypothesizing alternative situation models for reasoning and planning. This produces inferences of interest based on different objectives of inquiry, rather than determining inferences in pre-determined causal models. The work develops what I think is a novel and useful method for representing qualitative knowledge about the known causal relations along with probabilistic knowledge about the domain. Formalisms for reasoning based entirely on qualitative cause effect relationships fail to capture the associated uncertainty. On the other hand, reasoning based on probabilistic methods alone takes into account statistical correlations and ignores the available -knowledge about the qualitative cause-effect rela- tionships. This is not unlike the situation in pattern recognition years ago when it became apparent that both probabilistic and structural or linguistic aspects of patterns need to be employed. The problem of automated abductive reasoning which we address in Bhatnagar and Kanal (1993a,b), is to find interesting explanations, i.e.; situation models for a set of partially observed events, in terms of the known qualitative cause-effect relationships while using statistical correlations whenever the qualitative cause-effect knowledge is unavailable. We also use probabilistic and fuzzy (Bhatnagar and Kanal (1993c)) knowledge about the domain to determine the uncertainty associated with various parts of an explanation. The capability to hypothesize interesting, alternative causal structures is needed in several reasoning and planning domains, e.g., legal battles in a courtroom, military battles, medical diagnosis, and fault diagnosis.
A variety of approaches to reasoning in uncertain domains is surveyed in (Bhatnagar and Kanal (1992)) and developed in several volumes of Uncertainty in AI in the book series Machine Intelligence and Pattern Recognition published by Elsevier/North-Holland. An excellent talk by Prof. Edwina Rissland on AI and Legal Reasoning, given at the 1992 AAAI conference, covered a lot of recent Al research on the topic in a very interesting manner. A starting point for references on this topic is Rissland (1988). A recent book by Peng and Reggia (1990) develops abductive reasoning techniques for diagnostic problem solving.
Alternate realities are very much a part of the riddles being tackled by quantum physicists as is discussed in an interesting article (Horgan (1992)) in the July 1992 issue of Scientific American.
Recent workshops indicate that many in neuroscience, pattern recognition, neural networks and related fields are strongly attracted to theoretical approaches based on quantum physics. At some workshops there is much discussion of the mind-body problem, consciousness, and materialist versus dualist philosophies. All very interesting and lots of fun. Sometimes the temptation to make grand claims is hard to resist, even for Nobel prize- winners:
"We are at the beginning of the neuroscientific revolution. At its end, we shall know how the mind works, what governs our nature, and how we know the world."
Gerald M. Edelman (1992)
One should never say never, and perhaps Edelman will one day be proved right. But at least in this lifetime of mine, on this claim as on many other such claims, I think I will bet on the following words from Goethe:
Faust: Wohin der Weg? Mephistopheles: Kein Weg! Ins Unbetretene.
(Where lies the way? No way! Its Untrodden.)
In the early 1960's, conferences titled Adaptive systems, Self-Organizing Systems, Bionics, and Pattern Recognition, attracted scientists from many different disciplines, including psychology, engineering and