23 Structure-Mapping Theory 23.1 Introduction
2.6 Case-based Reasoning
The relative paucity of information concerning adaptation in psychological terms contrasts with the research undertake in Al, specifically that m Case-based reasoning. While it is not assumed that Case- based reasoning programs should be interpreted as psychologically accurate they may provide a reasonable basis on which certain hypotheses m ^ be founded, especially if they cohere with ideas current in psychological research. Consequoitly, the following sections will be concerned with an account of Case-based Reasoning.
2.6.1 Introduction
The essential character of Case-based reasoning can be stated quite simply; 'a Case-based reasoner solves new problems by adapting solutions that were used to solve old problems.' (Riesbeck & Schank 1989). The notion of Case-based reasoning has been developed fi*om earlier ideas concerning scripts (Sdiank& Ableson 1977), and contrasts sharply with rule-based accounts of thought (e.g. Newell & Simon 1972).
Rule Based Reasoning
Rule-based accounts of thought maintain that a set of specific mechanisms or rules underpins human intelligence. An AI goal is to render a machine which operates fi^om first principles and is able to solve any problem by reasoning firom these principles (Chamiak & McDermott, 1985).
The cmtrast between rule based and Case-based accounts of reasoning turns on the issue of whether reasoning or understanding a given situation or event is a matter of applying general rules and perfoiming pofaaps ccxnplex chains of inferoice cf identifying and adapting cases similar to the present instance. It is Reisbeck and Schank's (1989) contention that it is the latter form of cognition that people actually perform when constructing an explanation.
Examples of Case-based reasoning, in a naturalistic setting, are cited in the case of a radio interview conducted amongst American citizens concerning recent foreign pohcy issues, namely increased American involvement in El Salvador {ibid.) to construct plans concerning the relative merits of increased involvement, individuals cite previous examples of American involvement in foreign countries e.g.
What happened in Vietnam should have taught the American people a lesson.
I'm fearful that it might lead to a situation similar to what we faced in Vietnam 15 years ago, and therefore I would examine carefully anything we did before we got ourselves stuck in a situation we can't get ourselves out of.
Reisbeck and Schank go on to state Virtually when ever there is a prior case available to reason from, people will find it and use it for their future decision making.' Although a case is simply defined as e?q3erience. HowevŒ, it may be necessary to adapt a case before it can be applied to a current situation.
2.6.2 Case-based Reasoning and Adaptation
The importance of adaptation in AI research on Case-based reasoning has long been recognised (e.g. Reisbede & Sdiank 1989). The Case-based reasoning hterature (ibid.) has identified several different types of ad^tation. The most straight forward form of adaptation identified is that of'null adaptation': taking the solution to a previous problem and applying it directly to the present problem. This is what occurred with reference to the Vietnam / £1 Salvador example cited above. Instances of null adaptation are most fiequently found in cases where voy simple answers are required, e.g. 'should the US become involved in El Salvador: Yes / No ?'.
Reisbeck and Schank (ibid.), however, go on to state that a categorical answer alone (e.g. 'Yes' / 'No'), is of little use in the real world, what is required is the chain of reasoning that lead to the answer, i.e. how that answer was arrived at in the first place, and why that previous chain of reasoning is applicable to the current problem.
Reisbeck and Schank (ibid.) advance two types of adaptation technique: structural adaptation and derivational adaptation. These, however are types of adaptation wfrich take place within the same semantic or conceptual domain, e.g. adapting a meat recipe so it suits a vegetarian, or applying a technique used to solve one international dispute to solve another. Typically, however, analogical problem solving has investigated the process of applying analogues to solve problems in semantically distinct domains.
Kolodner (1993) identifies a further technique termed 'transformation' which provides some indication of how the process of inter-domain adaptation could be effected. Kolodner introduces the concept of transformation with the following remarks :
Transformation is the process of transforming an old solution to fit the new situation by making deletions or insertions, or transforming some element of the old solution by deleting some part of it, inserting some new element, adjusting the amount of something, or making a substitution. (Kolodner, 1993, itahcs mine).
Transformation is then the process of identifying the constraints inherent in a new problem and transforming the old case in order that it should conform to or satisfy such constraints. This process is, in the present wodc, characterised as ‘verification’; whereby possible analogically derived solution plans are compared against the constraints in the target problem. Solution plans which satisfy the cmstraints are verified as valid Where th ^ do not satisfy the constraints the solution plans are treated as instances of failed solution plans, which then stand as candidates for adaptation. Verification characterised in this way is tested for in Experiments 1 and 2 of the present work.
The items itahcised in the above quote fi*om Kolodner indicate possible ways adaptation could be realised: by applying operations of deleting elements of the solution plan firom an old situation and adding elements to the solution plan to transform it into a vahd solution plan for a present problem. These possible adaptation operations are tested for in human participants in Experiments 3,4 and 5. More detailed comparison between these operations, in terms of their relative ease of application, occupy Experiments 6 and 7.
There is no claim that the Case-based reasoning programs capture psychological facts about people’s actual problem solving practices, apart for the general claim that people reason with cases. Consequently, the psychological research concerning adaptation and analogical problem solving will now be examined.