Journal of the Operational Research Society (1995) 46, 274-275
OVER SIMPLISTIC COGNITIVE SCIENCE
In his paper on cognitive processes and OR [1], Klein has argued that the benefit of my enhancements to SSM conceptual models [2] are questionable. My point had been that SSM models were not adequate to model causation. This is because an adequate model of causation needs to be in terms of connectives corresponding to necessary conditions, sufficient conditions and necessary & sufficient conditions. SSM models have only a single connective which corresponds to a necessary condition. If these models are to be used as a basis for the design of information systems that can be informative about events in the physical world they need to be enhanced, at least tacitly, with two additional connectives. Such enhancement might be undertaken by a professional analyst working with the models after the stakeholders' iterative debate has finished. My point is that the stakeholders can play a greater part in the design of the system if they build a conceptual model with more of the required complexity.
It is not clear why Klein objects to this. As the relevant section discusses the desirability of simplicity and causation, it could be that he disagrees with my account of causation or because he thinks enhanced SSM models would be too complex.
With regard to causal modelling in Eden's work, Klein says "Cognitive mapping ... reduces the causal complexity of the world to just two types of relationship: positive and negative". If this is the case I then wholly disagree with cognitive mapping. I cannot see how (and it is certainly not explained how) necessary, sufficient and necessary & sufficient conditions can be represented by models of this form. My account of causation is quite traditional. It can be found in first year undergraduate logic textbooks and is bound to be mentioned in any book in the philosophy of science that discusses causation in any depth. If Eden has been putting forward a new account of causation I am puzzled as to why he never mentions the traditional account.
I don't think that Klein has looked at causation in any depth. The main thrust of his argument seems to be (a) that cognitive science shows that simple models are easier to understand and (b) that simple models are those with the fewest modelling rules. Both of these arguments are open to dispute. Many of the models used in cognitive science are very complex. Klein cites Sowa's work in cognitive science with evident approval so I think it is fair to take Sowa's conceptual graphs as an example. These are far more complex than the models I proposed. Indeed in a recently published conceptual graph [3] the verb "to run" is represented by ten arrows and ten bubbles, and each bubble is of a different logical type. Klein, therefore, does not present a balanced view of cognitive science and operational researchers who are unfamiliar with the literature should not take his paper as being representative of a consensus in this subject.
I also dispute that models with the fewest modelling rules are the easiest to understand. A good example is propositional logic. This standardly uses five constants - negation and four logical connectives. These are the biconditional, implication, alternation and conjunction. Each has its own set of rules. If we want to say "p if and only if q" we would normally write it using the biconditional as ' p <-> q'. However, if we wanted to reduce our modelling rules we could eliminate the biconditional entirely from the system and replace it with implication. 'p <-> q' could be expressed as '(p -> q) & (q -> p)' this could be rendered in English as "if p then q, and if q then p". We could reduce our modelling rules even further and eliminate all the connectives except conjunction. In this case 'p <-> q' becomes '¬ (p & ¬ q) &¬ (q & ¬ p)' this can be expressed in English as "it is not the case that p is the case and q not the case in any circumstances where it is not the case that q is the case and p is not the case". It is hard to see how this is easier to understand then "p if and only if q", yet it belongs to a system with fewer modelling rules.
The reason logicians use a system with four logical connectives rather than a system with one is the fact that the system with four connectives can be more easily mapped to statements made in European languages. This is, I think, the key factor. People will find models easy to understand in so far as they resemble, or follow from, a language which they already understand. The number of modelling rules is a secondary factor. Cognitive science contains work that supports this view as well as the work that Klein mentions.
1. J. H. Klein (1994) Cognative processes and operational research: a human information processing perpective. J. Opl Res. Soc. 45, 855 - 866.
2. F. Gregory (1993) Cause, effect, efficiency and soft systems models. J. Opl Res. Soc. 44, 333 - 344.
3. J. F. Nogier and M. Zock (1992) Lexical choice as pattern matching. Knowledge Based Systems, 5 (3), 200 - 212.
City Polytechnic of Hong Kong Frank H. Gregory