Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
SYNOPSIS OF GENERAL AUDIENCE DISCUSSION - bst of the points raised during Session IV and the general discussion centered arc und two somewhat related issues: 1. the gap between behavioral (heuristic) and traditional (rule based) approaches to decision making, and 2. how to HE with shortcomings in one or the other that detract f ~ u system performance. m e Gap Issue m e observation was made that there seem to be two ways of thinking about decision problems, each with its own philosophy and research agenda, that are proceeding more or less independently. To some extent, it was pointed cut, the two papers in the session highlight the differences between the two approaches. me question was whether, and if so how, they should be integrated or links morn cicsely. Two conflicting views were offered. One was that since the differences are deeply rooted in their respective traditions and Captures, the barriers will not be broken down easily, and the anticipated payoff for NINA would probably not justify the time and cost necessary to bring about an integration. A number of other issues should take precedence over this one. - ~ ~ . ~ ~ ~ . ~ ~ ~ . ~ . The chairs visor was that the two approaches snatch ne nether Denigrated, peaceably can be if NOVA puts the issue on its r~r~ agenda, and in fact is being attempted in a small way trough r~r~ currently in progress in Fischhoff's Hi. Among the suggestions for an Integrative approach were the whole domain of fuzzy logic and the founded rationality concept (e.g. defining general goals and then "fiddling with the model at the margin as in 'satisficing"'). It was pointed out, however' that In the context of expert systems such approaches reduce to writing a lot of conditional rules over a large number of state variables. This-= one cannot summarize -~.=ily what the system will do over the full range of decision problems. 275
276 Applications, Or Dealing With Shortcomings Several options ware suggested for minimizing the effect of suboptimalities in human judgment. Training, while not universally effective in overcoming biases, has produced some not~hie suc-~-=s~s (e.g. weather forecasters). The key may well lie in the proper design of training programs (something that merits a continuing research effort). Increasing the trainee's sophistication in statistical concepts, however, is clearly of little help. Aiding in its various forms and with its inventory of existing models has its place but also has limitations. MUltiattribute utility theory, decision analysis, etc. are useful for solving well defined problems, but "bring no knowledge to the party." Often their logic is not trar~rent to the user and critical factors may be emitted. Thus Fife output may not be satisfactory in either an absolute sense or as E ~ hived by the user. When it conflicts with he m n intuition there is a problem nartim,Qarlv if the human doesn't understand the logic. ~ , ~ ~ User acceptance of even improved decisions becomes crcbiematic. . . One approach to dealing with these deficiencies in the aiding models was advocated by Davis: find cut what is missing an] build it An. Intuition and creative thinking are not magic' but rancher, "~iscaveredl rationality." R - earth Should try to expose that rationality (or reasoning) and apply it in creating more rcit~ust models, as well as mom transient ones. To Me extent that the research subs, it should be incorporated into training as well as aiding applications, and be result cock be better decisions and greater acceptance of those decisions by users (who wed now be more likely to appreciate the logic).
