What is a problem? Theoretical conceptions problem solving
Transcription
What is a problem? Theoretical conceptions problem solving
EUROPEAN JOURNAL OF BEHAVIOR ANALYSIS 2008, 9, 157 - 172 NUMBER 2 (WINTER 2008) 157 What is a problem? Theoretical conceptions and methodological approaches to the study of problem solving Per Holth Akershus University College, Norway The current paper describes important events in the history of research on problem solving and discusses questions regarding the classification of problems. First, the main research traditions and major research findings in the field are described. Some major problems in the attempt to define a problem, including the issue of novelty, are discussed. A solution is suggested, based upon the definition of a problem situation as one that no prevailing three-term contingency has established as a discriminative stimulus for an action that produces a reinforcer. The final section is concerned with some differences between cognitivistic and behavioral treatments of problem solving. Key words: problem solving, novelty, mediating behavior, cognition, behavior analysis. The literature on problem solving According to PsychInfo, the first publication to mention problem solving either in the title or in the abstract was a paper by Willard S. Small, published in 1900. Small investigated what he called the “persistence of useless motor habits” and “fortuitous associations . . . in problem solving.” The title of his article was “An experimental study of the mental processes of the rat.” After that, only two publications on problem solving appeared in 1907, two in 1914/15, before the rate of publication began to accelerate in the 1920’s and continued to do so until the early 1990’s, when the rate leveled off, but increased again from 2004 (see Figure 1). By the end of 2007, the sum total of such publications was 22,088. Some of the early studies on problem solving appear strikingly modern. For instance, the research questions asked by B. W. Daily (1925) to a large extent match research questions asked in Hank Kahney’s (1993) book on current issues Address correspondence to: Per Holth, Akershus University College, Box 423, N-2001 Lillestroem, Norway. E-mail: Per. [email protected] Figure 1. The numbers of annual publications that, according to PsychInfo, mention ‘problem solving’ either in the title or in the abstract from 1900 through 2007. 157 Per Holth 158 in problem solving. On the other hand, the early days of the problem solving literature showed little sign of band-wagon research in which particular research questions guided a larger body of empirical or theoretical work. Some early research was concerned with methods of teaching arithmetic, some were concerned with showing that “insight” in Köhler’s sense of the word occurred or did not occur in children or in adults in different types of problems, and some were concerned with the special limitations of ‘idiots’. From 1930 to 1932, for instance, C. G. Aldrich and coworkers published a number of articles on problem solving in subgroups of people (e.g., Aldrich, 1931; Aldrich & Doll, 1931). Research traditions In 1966, Forehand, among others, distinguished between four different bodies of research into problem solving, which might be designated as follows: (1) the Gestalt/cognitive tradition, (2) the learning tradition, (3) the computer/information-processing tradition, and (4) the psychometric/component-analysis tradition. The Gestalt/cognitive tradition comprises a large body of experiments aimed at demonstrating ‘insight,’ ‘productive thinking,’ and ‘structural reorganization.’ Early work included both animal studies (Köhler’s monkey studies), and studies of human problem solving by Wertheimer (1945; posthumously), Duncker (1945), and Maier (1930, 1931). They were interested in the extent to which problem solutions emerged in a step-by-step fashion versus as a sudden, complete whole. Studies of animal ‘insight’ naturally went out of favor within this tradition after Birch’s (1945) and Harlow’s (1949) careful demonstrations of how ‘sudden’ solutions shown by their monkeys depended upon specific types of individual histories. As pointed out by Wertheimer (1945), the Gestalt tradition was primarily interested in what happened at the moment of ‘sudden insight’ in human subjects, but they did not completely ignore the relevance of ‘past experiences’. In fact much work was invested in demonstrating how training histories often hindered rather than fostered ‘insight.’ They called it a functional ‘fixedness,’ ‘fixity,’ ‘einstellung,’ or ‘set.’ In Duncker’s candle problem, for instance, subjects were given a candle, a box of nails, and some other objects, and they were asked to attach the candle to a wall above a table so that it did not drip onto the table. Duncker’s own solution was to use the nailbox as a candle holder which could be nailed to the wall, but very few subjects did that. According to Duncker, subjects were ‘fixated’ on the normal function of the box, and problem-solving success was hampered by ‘reproductive problem solving behavior,’ consisting only of previously directly reinforced responses. The results of Maier’s string problem could be interpreted similarly. Maier had two strings hanging from the ceiling and certain objects, such as pliers, lying around. When participants were asked to tie the two strings together, they discovered that when they took hold of one string, the other was out of reach. After a while, some participants would attach an object to one of the strings and make it move like a pendulum and, while holding one string, the other could be caught on its up swing and then the two could be tied together. However, two thirds of Maier’s subjects failed to simultaneously grab two ropes hanging from the ceiling – to tie them together – although objects were present which could have been tied to the end of one rope to produce a pendulum effect which would obviously be helpful. Luchins (e.g., 1942; Luchins & Luchins, 1950) went one step further and actually provided his subjects with an experimental history that produced such rigidity in problem solving. Useful as some of these experimental findings may be in predicting behavior, the overall attempt to show that “organisms organize” may add very little, and was what Boring (1957) called “the great tautology of psychology.” Moreover, the Gestalt question of whether problem solving is “productive” or just “reproductive” may constitute a pseudoempirical rather than an empirical question – simply because ‘novelty’ may, as we shall see, constitute a defining characteristic of a ‘problem.’ What is a problem? What has been called the learning tradition (sometimes, the S-R tradition) is hardly a unitary approach except, perhaps, for a clearly pronounced common interest in the relevance of variables in the history of each subject. Interestingly, Watson (1920) may, in fact, have been the first to systematically use verbal protocols in experiments on ‘thinking.’ For instance, he asked subjects to talk aloud when given assignments characterized by different degrees of novelty to the subject. He could observe characteristic response patterns under these different conditions, with an increasing number of errors, false starts, “hanging of the head” and even “blushing” when subjects were given the more novel assignments. Watson pointed out that he had “often felt that a good deal more can be learned about the psychology of thinking by making subjects think aloud about definite problems, than by trusting to the unscientific method of introspection” (1920, p. 91). Although verbal protocols have been widely used in the Gestalt tradition (e.g., Wertheimer, 1945) as well as in modern cognitive psychology (see Ericsson & Simon, 1980), they have only recently been employed seriously by others in the radical behaviorist tradition (e.g., Hayes, White, & Bissett, 1998). After the work of Birch (1945) and of Harlow (1949), Kendler and Kendler (e.g., 1962) analyzed problem solving in terms of combinations of separately established repertoires, and such procedures were explicitly used by Epstein and colleagues (Epstein, Kirshnit, Lanza, & Rubin, 1984; Epstein, Lanza, & Skinner, 1980, 1981) to simulate ‘insight’ and other “higher mental processes” in pigeons in what was called the Columban simulation project. Beyond that, behavior analysts were stuck with exploring basic behavioral processes for a long time, and their treatment of more complex and interesting human phenomena have largely been theoretical interpretations as in much of Skinners later work (e.g., Skinner, 1953, 1957, 1966, 1968, 1971, 1978). Direct experimental analyses of more complex human performances, such as stimulus equivalence, rule-governed behavior, naming, and ‘joint control’ have only recently begun to flourish within the behavioral tradi- 159 tion (e.g., January issue of JEAB, 1998). The Computer/information processing approach has been particularly influential in modern cognitivist research and theorizing on problem solving, and the work of Newell and Simon (1972) has been called the bedrock of the information-processing framework (e.g., Eysenck & Keane, 1995). In the ideal case, human problem solving is depicted metaphorically in a computer language that allows for a direct simulation of complex human phenomena in the computer. Thus, input ‘information’ is taken in by the organism, is ‘processed’ according to a ‘program,’ the information is said to be ‘encoded,’ that is, converted to a form that can be handled. It is ‘represented’ within and may occupy ‘space,’ much like bytes in the ‘working memory,’ and it may be ‘stored’ in ‘long-term memory’ – such as on a hard disk – ‘tagged’ for possible ‘retrieval’ in the future to produce an ‘output’ – a ‘printout.’ As pointed out by Davis (1966), the computer simulation procedure makes the demand that ‘prior information’ as well as possible solutions and the sequences of steps in problem solving are specified by the researcher. This is typically attempted in terms of problem space theory, which describes the possible sequences of steps or states from an initial state to a goal state in the form of a flow chart. Whereas a state space flowchart is considered as an omniscient observer’s view of the structure of a problem, the problem solver is construed as going through corresponding sequences of mental knowledge states. Derived from this tradition are some widely used heuristic devices, such as means-ends analysis, as well as partial answers to some structural questions pertaining to problem solving, such as what distinguishes experts from novices. The fourth line of research, the psychometric/ component-analysis line, aims to identify units, called factors, of abilities or skills which, once measured, allow for the prediction of success on other tasks. Through factor analysis (the multiple correlations of test scores across test batteries), psychometricians have identified variable numbers of intelligence factors, for instance, ranging from Spearman’s (1923) Per Holth 160 original one, (‘g’), to Guilford’s (1967) 120, but leveling out in more recent literature (e.g., Kahney, 1993) at around five or six. However, cognitivists have moved on to try to answer not just “What are the factors?” but also “How do they work?” The cognitive-component method of Sternberg (1977) with respect to problem solving, for instance, is described as consisting of two steps: (1) An armchair analysis of the kinds of processes that might be involved, in accord with information-processing theory, and (2) an experimental test based on the assumption that each component mental event takes time to execute. Adding components to be repeated, or subtracting components by guiding subjects through some of the components in what is called pre-cueing, helps decide whether each hypothesized component actually takes time, and how much. Types of problems In a paper titled “Current status of research and theory in human problem solving,” Davis (1966) wrote that “Research in human problem solving has a well-earned reputation for being the most chaotic of all identifiable categories of human learning.” According to Davis (1966), “. . . virtually any semi-complex learning task which does not clearly fall into a familiar area of learning can safely be called ‘problem solving’.” In fact, even wider definitions of problem solving have been suggested. For instance, according to Radford and Burton (1974), “. . . when psychologists investigate problems, they are concerned with any situation in which the end result cannot be reached immediately.” Thus phenomena described under the heading of problem solving range from Thorndike’s studies of cats in puzzle boxes through the human problem solving experiments carried out within the tradition of Gestalt- and cognitive psychology, to the discoveries of Galileo and Einstein. In an attempt to bring some order and continuity into the field, Davis (1966) classified different types of problems according to two broad sets of criteria. In problems of type O (overt), the outcomes of different response alternatives were not known to the problem solver, so that overt testing of alternatives was necessary. Hence, the problem solving behavior was observable, and it was usually described by what he called “behavioristic learning concepts.” A prototypical example would be the behavior of cats in Thorndike’s puzzle box studies. Once outcomes of response alternatives are known to the subject, however, these can be considered covertly, as Type C (covert). Such covert problem solving tended to be described in mentalistic terms, often within the Gestalt vocabulary. A typical example would be the engagement in ‘mental arithmetic’ to solve math problems. An overarching goal for research in problem solving may still be as Kahney (1993) wrote, “to be able to say, ‘Look, there are essentially five (or ten or fifteen or whatever) types of problem. Any problem you name can be categorized as one of these types, or some combination of them’.” (p.15). The standard practice is to classify problems mainly or solely in terms of their structure. Thus, problems may be characterized in terms of their “deep structure.” Different problems are said to have the same deep structure on the basis of whether their state spaces (derived from state space analyses) are identical (e.g., Kahney, 1993). Problems with identical “deep structures” are called isomorphic, but the empirical evidence thus far suggests that often surprisingly little transfer occurs even between isomorphic problems (e.g., Reed, Ernst, & Banerji, 1974). That is, of course, an interesting research issue in its own right, but it is not promising with respect to developing structural classes that correspond to functional classes of problems. In actual practice, the problems studied experimentally are often simply classified according to their structural similarity with some prototypical experiments, often named after the materials used and/or the person who first described it. Hence, we have Duncker’s candle problem, string problems of the Maier type, Luchin’s water jar problem, Tower of Hanoi problems, and Missionary and Cannibals problems. Otherwise, problems are said to be well defined when the problem space is easily specified, and ill-defined if it is not, and problems are called convergent or divergent What is a problem? depending upon whether there is only one, or more possible solutions. Related to questions concerning the ecological validity of laboratory experiments on problem solving, a distinction has also been made between “contrived” problems and real-life problems (Sinnott, 1989). Whereas real-life problems are considered to occur in a meaningful context and perceived as worth solving, contrived problems are typically de-contextualized and abstract (Lebow & Wagner, 1994). Accordingly, empirical studies suggest that problem solving may be more adaptive and inventive in real-life than in contrived laboratory problems. Problems in defining a ‘problem’ (a) Related terms. Authors have often pointed out a close relation between ‘problem solving’ and other concepts used as headings for research areas in psychology, such as ‘learning,’ ‘thinking,’ and ‘creativity.’ For instance, Davis (1966) wrote that “. . . virtually any semi-complex learning task which does not clearly fall into a familiar area of learning can safely be called ‘problem solving’.” Both Radford and Burton (1974) and Skinner (1968) underscored the close relation between problem solving and thinking. According to Radford and Burton (1974), J. P. Guilford added ‘creativity,’ stating that ‘creative thinking’ can be equated essentially with ‘problem solving.’ In a broader perspective, there is a sense in which all basic research areas in psychology may be related to ‘problem solving,’ as pointed out by Kahney and by Skinner, although in different terms (see Table 1). However, attempting to define a particular area of research that may require some sort of 161 special treatment, different researchers have postulated definitions of ‘problem’ and ‘problem solving’ that vary in broadness. (b) Some suggested definitions. Radford and Burton’s (1974) definition of ‘problem’ as “. . . any situation in which the end result cannot be reached immediately” (p. 39) has already been mentioned. According to Kahney (1993), “whenever you have a goal which is blocked for any reason – lack of resources, lack of information, and so on – you have a problem. Whatever you do in order to achieve your goal is problem solving.” (p. 15) Newell and Simon (1972) wrote that: “A person is confronted with a problem when he wants something and does not know immediately what series of actions he can perform to get it” (p. 72). In Skinner’s (1966) terms, “behavior which solves a problem is distinguished by the fact that it changes another part of the solver’s behavior and is reinforced when it does so (p. 225). Finally, Davis (1973) suggested that “A problem is a stimulus situation for which an organism does not have a ready response” (p. 12) (c) Problems with the definitions. Even if, at first sight, the definitions may look clear enough, “closer analysis easily muddies the waters” (Davis, 1973, p. 13). Let us consider some issues related to each of the definitions. First, if you are seated on a train on your way from Oslo to Stockholm, for instance, the end result cannot be reached immediately. According to the definition suggested by Radford and Burton (1974), then, you would have a problem whereas, in fact, you have not. Obviously, reaching the “end result” or “goal” must depend upon some previous action by the Table 1. Comparing Skinner’s (1966) and Kahney’s (1993) views on the relevance of basic processes in the analysis of problem solving. Every advance in understanding perception, memory, language and problem solving will feed into endeavours to help people become better problem solvers. (Kahney, 1993, p. 143) Since there is probably no behavioral process which is not relevant to the solving of some problem, an exhaustive analysis of techniques would coincide with an analysis of behavior as a whole. (Skinner, 1966, p. 225) 162 Per Holth problem solver. This is taken care of in Skinner’s (1966) definition, but including all the cases in which some behavior “changes another part of the solver’s behavior and is reinforced when it does so” also embraces cases which would not normally be considered as problems. It would include any regular behavioral chain, such as making coffee, getting to work, getting started on your computer, and so on, but standard usage suggests that these would not be considered as problems unless some sort of ‘obstacle’ arises. Raaheim (1988) criticized Skinner for “omitting ‘difficulty’ as a defining characteristic of a problem” to which Skinner (1988) simply replied that “There are easy problems and there are hard ones, and they are both problems” (p. 260). However, this may still require a consideration of what characterizes a minimal problem, because problem solving would otherwise actually coincide with the whole field of operant behavior. Indeed, Reese (1994) listed ‘obstacle’ as a crucial feature of a problem: “If no obstacle hinders progress toward a goal, attaining the goal is no problem” (p. 200). But what constitutes an obstacle? There are certainly problems without physical obstacles. If I ask you to come up with the answer to 372 (square) “in your head”, there are no particular physical obstacles to stop you. You are free to tell the answer, but it might still be a problem. Contrary to Newell and Simons’ (1972) definition, you may even “know immediately” a series of actions would lead to the answer, but still be unable to carry through with it. The cognitivist notion of “lack of information”, as in Kahney’s (1993) definition, may seem attractive from a colloquial point of view, but such terms are, as we shall see, undermined by experimental findings. Finally, Davis’ (1973) definition is flimsy for at least two reasons. First, it lacks a reference to what might be called the motivational aspect of a problem. Luckily, not all stimulus situations require ready responses. Second, and less trivially, what, does it mean to “not have a ready response”? Can it be determined ad hoc only, or is it predictable, at least in principle, from other variables? If we want to study how people solve problems, we do obviously want to know how to construct a problem. Donahoe and Palmer (1994) have suggested that the definition of a problem should include the requirement that a solution must be possible for the subject. This is in agreement with Saugstad (1977) who argued that “it does not make sense to speak of a problem unless the individual is in some way capable of solving what is regarded as the problem” (p. 115). However, experiments on problem solving have, in fact, occasionally used problems for which there is no solution, although proving that there is no solution is sometimes then itself the solution. (e.g., J. R. Hayes, 1978; as referred to in Kahney, 1993). A more important point brought up by Saugstad (1977), among others, is that “problem solving cannot be said to take place when an individual merely reproduces what he has previously learnt.” (p. 116; also Davis, 1973). Hence, as Saugstad (1977) pointed out, a “main difficulty in making theoretical statements on problem solving originates in the difficulty of saying what is meant by ‘novelty’” (p. 115). In a trivial sense, every stimulus and every response are novel. They are never quite exactly like any previous instance. The issue, therefore, is one of characterizing the type or degree of novelty required for some situation to constitute a problem. The “novelty” in a problem. Saugstad (1977) wrote that: “Apparently if some definite task is mastered by an individual by an activity which may be described as remembering, we should not regard this as problem solving.” But certain types of remembering have been convincingly interpreted precisely as problem solving by Palmer (1991), Donahoe and Palmer (1994) and Delaney and Austin (1998). In the following, a few technical terms will be needed: (1) The concept of the three-term contingency, SD R SR, that is, discriminative stimulus, response, and a reinforcer, and (2) the concept of an establishing operation (EO). An EO serves two functions: First, to establish the reinforcing effect of certain stimulus changes and, next, to initiate behavior previously followed by those What is a problem? stimulus changes. Palmer (1991) distinguished between two classes of contingencies for which the term ‘memory’ is usually said to be required. The first is a matter of direct stimulus control, as when you are asked to tell the answer to “5 times 7”, or you answer Rome, promptly, to the question “What is the name of the capital of Italy?” As a result of standard education, your answer is under direct stimulus control of the question as a verbal discriminative stimulus, it does not depend on some intermediate behavior on your part, and it is explained satisfactorily by a prevailing 3-term contingency in the presence of relevant motivational variables – EO’s. There is, of course, a temporal gap between the contingencies that established the behavior and the current instance, and it is obviously tempting to fill that gap in order to complete the account. As Palmer pointed out, “for example, one may propose that synapses have been modified or created, resulting in an organism that responds in a particular way in a particular setting, but accounts of this sort are not necessary if our criterion of explanation is to be the prediction and control of behavior. A physiological explanation supplements the behavioral one; it does not replace it” (1991, p. 266). However, it is the second memory class characterized by Palmer which is of primary interests here. Consider a somewhat more complex math question, like “67 times 73.” Standard math education has not prepared us to answer this question directly, that is, we have not encountered the relevant 3-term contingency in which the question could be established as an SD for the correct response. Similarly, a question such as “What did you have for dinner last Tuesday” does not usually work directly as an SD for the correct answer, because the appropriate events according to the three-term contingency have not occurred. In sum, then, the novelty which should enter into the definition of a ‘problem’ occurs when some behavior is scheduled for reinforcement (that is, an EO is in effect), but no prevailing three-term contingency has established the current situation as an SD for an action that produces the reinforcer. Additional complexities arise here. How 163 do we know that some behavior can not be ascribed directly to the relevant three-term contingency? The three-term contingency is sometimes described as the A–B–C of behavior analysis, suggesting an antecedent, behavior, and a consequence, but the formula does not apply to just any arbitrary subdividing of successive events into Antecedent, Behavior, and Consequence. Sometimes the formula does not seem to readily fit the observed facts. Yet, behavior analysts have often neglected to state the criteria for when the formula applies and when it does not. Identifying instances of SDs and Rs. The SD is defined as a stimulus in the presence of which a particular response is typically reinforced and which, as a result, controls the response in the sense that in the presence of the stimulus, the response is more likely to occur than in the absence of that stimulus. According to the definition, then, a history of direct reinforcement of the response in the presence of that stimulus is in some sense required in order for the stimulus to qualify as an SD. However, as Skinner (1935) pointed out, “it is very difficult to find a stimulus and response which maintain precisely the same properties upon two successive occasions” (p. 347). Thus, Skinner (1935) showed that it was necessary to consider classes of stimuli and classes of responses. The class concept requires some defining property that allows for the determination of class membership. Hence, as pointed out by Skinner (1969): “The topography of an operant need not be completely fixed, but some defining property must be available to identify instances.” (p. 175). In accord with Skinner’s previous work, Catania (1973) argued that the concept of the operant grew out of a correlation between two response classes, one descriptive and one functional. The descriptive operant is the class of responses for which consequences are arranged and the functional operant is the class generated by that contingency. While any instance of the descriptive operant class can be identified when it occurs, we can only infer that particular instances are also members of the functional class. A functional class involves a controlling relation, and controlling relations are never 164 Per Holth directly observable (cf., Sidman, 1979). Several manipulations, and observations of instances and non instances may be required for the identification of controlling relations, and even when a rat “lever presses for food” in an experimental chamber, any particular instance of lever pressing might occur “for other reasons.” A descriptive operant class sometimes involves only response-descriptive features. In discriminated operants, however, the descriptive classes also involve stimuli or stimulus properties in the presence of which responses are followed by certain consequences. By extending Catania’s (1973) analysis of the operant class to the case of the discriminated operant, it is clear that the class of antecedents called SD similarly requires a correlation between a descriptive and a functional class of “SDs.” Evidence of SD control, therefore, requires a demonstration of a correlation between the stimulus class in the presence of which particular responses are reinforced and the stimulus class in the presence of which those particular responses subsequently are more likely to recur. The existence of a descriptive discriminated operant class (such as responding correctly to some specific types of multiplication tasks) may not be accompanied by a corresponding functional class (i.e., actually responding correctly to all such multiplication tasks) unless specific precurrent behavior is established. If so, the presented task is not an SD for uttering or writing the correct answer, but for the precurrent behavior which may lead, eventually, to the correct answer. In ‘Relational Frame Theory,’ Hayes and coworkers have described a number of different complex relational frames in accord with which humans sometimes respond, such as coordination (equivalence is one example), opposition, distinction, comparison, and hierarchical, temporal, and spatial relations . In their chapter on thinking and problem solving, Hayes, Gifford, Townsend, and Barnes-Holmes (2001) defined “thinking in a verbal sense” as “a reflective behavioral sequence, often private, or pragmatic verbal analysis that transforms the functions of the environment so as to lead to novel, productive acts” (p. 95), and “verbal problem solving” as “framing events relationally under the antecedent and consequential control of an apparent absence of effective actions” (p. 96). Hence, it seems clear that these proponents of relational frame theory recognize examples of problem solving that consist of “behavioral sequences” that can mediate “novel, productive acts”. However, according to Hayes, BarnesHolmes, and Roche (2001), complex relational responding is typically established through multiple exemplar training, and they see no need to posit mediating behavioral responses to account for ‘derived relational responding’ (Hayes, Fox, Gifford, Wilson, Barnes-Holmes, & Healy, 2001). Yet, the prerequisites for, and the generality of relational responding without such mediation or precurrent behavior remain to be explored. For instance, it is not yet obvious that the initial trials of a stimulus equivalence test do not constitute a problem in the sense that precurrent responses of some sort are typically necessary in order to respond in accord with equivalence. Some experimental data suggest that the initial test trials do constitute a problem in this sense. For one thing, reaction times typically increase initially during testing, and in a study that more or less eliminated the possibility of precurrent responding by constraining reaction times initially during testing to 2s, the subjects did not respond in accord with equivalence (Holth & Arntzen, 2000). Although a ‘relational frame’ (such as coordination in this case) may suffice as a descriptive class, it may not be accompanied by a corresponding functional class. We do not know ‘the lines of fracture’ along which new instances actually emerge in the repertoires of organisms in the absence of functionally relevant intermediate behavioral events. Presumably, a problem only exists when they do not. Basic, unmediated SD R relations can be blurred by different variables, such as: (a) the lapse of time between the “SD” and the R, and (b) the complexity of a relational pattern between the “SD” and the corresponding R. In both cases, at some level, the SD ceases to function as such with respect to that particular response. When, in spite of this, some apparent functional relation between the S and the R still remains, the relation must be bridged or mediated by additional events. We have What is a problem? already discussed the case of the time delay in the example of memory as problem solving. Basic research with both humans and animals shows that in the absence of potentially mediating events, the basic functional relation breaks down if the interval exceeds a few seconds (e.g., Blough, 1959; Peterson & Peterson, 1959). As the complexity increases, the less likely it seems that the appropriate novel responding results automatically from a particular identifiable contingency. For instance no number of multiple exemplars with the multiplication of three- and four-digit numbers would suffice to establish the appropriate problem solving skills with respect to novel exemplars in the absence of precurrent problem solving skills. An example: Precurrent responses in continuous repertoires As Skinner (1968) has pointed out, precurrent responses need not be explicitly reinforced. But neither does the terminal response or “solution”. When precurrent behavior is not terminated by some explicit reinforcement of a successful response, we must point to other variables to account not only for why the precurrent behavior starts, but also for why it stops, i.e., when we “know we are right.” In some performances, such as imitation, instances cannot be identified in terms of stimulus and response properties alone, but in terms of correspondences between response dimensions and stimulus dimensions. Such cases, in which reinforcement is contingent upon a correspondence between response and stimulus dimensions, were described by Skinner (1953) as continuous fields. Wildeman and Holland (e.g., 1972) reported a couple of experiments on such contingencies and called the resulting behavior continuous repertoires. They found that such repertoires were easily established in young school children but not in pigeons. However, little more of the basic research was carried out. In a simple case of a potentially continuous repertoire, let us say we reinforce the selection of RIGHT 1 in the presence of LEFT 1 (see Figure 2, upper panel), LeftRIGHT 2 in the presence of LEFT 2, and RIGHT 5 in the presence of 165 LEFT 5. Now, what will happen if we present, say, LEFT 3 or LEFT 6? A subject could, for instance, respond in accord with a continuous repertoire by horizontal eye movements or by drawing a line from left to right and by stopping when arriving at a stimulus in the right-hand column. However, this strategy would have limited success across different task - as in the rotated dimension task (Figure 2, lower panel), or in arbitrarily-related dimensions task, or in, say, multiplication. If, instead, the subject emits a differential response, such as pronouncing the number (possibly as a result of counting) and rehearses it until the same topography is jointly controlled by a second stimulus, this would be applicable to a range of different kinds of problems (e.g., Lowenkron, 1991, 1996, 1998). Even in some cases that typically involve specific types of explicitly trained precurrent responses, joint control is involved when we are checking up on ourselves, as in self editing. For instance, 1 1 2 2 3 3 4 4 5 5 6 6 1 2 3 4 5 6 6 5 4 3 2 1 Figure 2. Continuous repertoire training materials. Upper panel: A vertical stimulus dimension to the left, and a corresponding response dimension to the right. Lower panel: A vertical stimulus dimension to the left, and a corresponding rotated 90O response dimension. 166 Per Holth having a couple alternative solutions to “73 multiplied with 77,” we might do “3 times 7 equals 21” and, as a result, decide on the one that “ends with 1.” A few words need to be said about the general reluctance among behavior analysts as regards what is conveniently called mediating or precurrent performances. This is clearly not an absolute question of accepting or rejecting the inclusion of mediating behavior in one’s analyses, since the area of problem solving is defined by the functional relevance of such performances. Thus, rather than questioning whether or not mediating performances should be accepted as part of the picture, we may be concerned with the levels of complexity at which some sort of problem solving is necessarily, or most likely, part of a successful repertoire. An important task, then, is (1) to find out when basic SDR relations actually break down, (2) to identify the types of events (such as transpositional repertoires) capable of mediating the still occasionally apparent Stimulus Response relation, and (3) to specify histories that give rise to those mediating performances. I will now turn to methodological and theoretical issues related to such mediating events, and to certain contrasts between cognitive and behavioral accounts. Contrasting cognitive and behavioral accounts (a) The question of mediation. That different sorts of covert mediation of behavior occurs in very much of human behavior is such an obvious fact that no one can seriously question it. If you just watch someone being asked when they went to bed last night, or to come up with the square of 65 with no aids, many very anomalous results, including very unpredictable response latencies, will be found in the area of stimulus control if you exclude from consideration some mediating events that are even quite obvious to anyone who has engaged in a little self observation. Moreover, the cognitive literature is replete with empirical results that make sense only on assumptions of certain mediating events, such as in experiments on ‘mental rotation,’ by Shepard and coworkers (Cooper & Shepard, 1973; Shepard & Metzler, 1971), in experiments on visual tracking – even in the pigeon (Neiworth & Rilling, 1987), and in Kosslyn’s (1975, 1976) experiments on image tracing tasks. So, the question is really not at all whether such functional mediation takes place but, first, what are the criteria for inferring it and, second, in what terms should those mediating events be described? The criteria are different within cognitivist and behaviorist frameworks and, of course, so are the terms in which the inferred mediating events are described. Before contrasting behavioral and cognitive interpretations of cognition and problem solving, it is tempting to describe some essential features of the behavioral view that are often misrepresented in psychology. For instance, the behavioral perspective is actually a selectionist view rather than a stimulus – response view, it is concerned with events that take place inside the skin of organisms, it does not depict behavior according to an input-output formula, and it does not generally ignore ‘cognitive’ phenomena, but it ignores internal duplicates or representations of those phenomena – particularly as automatically explanatory entities. Anyone can find out about that easily by checking primary behavioral literature sources. Moreover, the continuing misrepresentations of behavioral views in psychology are amply documented elsewhere (e.g., Catania, 1991; Schlinger, 2002; Todd & Morris, 1992). Several authors have been concerned with what is sometimes called a Procrustean bed of behavior analysis. For instance, Davis (1973) wrote: By dissecting human thinking and problem solving into the theoretical language of simple conditioned responses, the theorist must ignore too much beautifully conscious and deliberate mental behavior. . .To even attempt a complete picture of such activities, one must use a phenomenological language which accepts the complexity of the conscious, thinking, and feeling human being (Davis, 1973, pp. 58-59). What is a problem? The same problem, however, is faced by other sciences. I recall an incident where a physiology teacher said that: “Erection is a vascular-dilatation phenomenon which can be understood from the structure of the penis.” Some of his students disagreed red-bloodedly. We often want more than the dry and narrow scientific facts, but it is not obvious that the typical cognitive vocabulary will come off much better here. (b) Cognitivist guidelines for inferring cognitive events. As pointed out by Reese (1994), even the most ardent cognitivist, such as a Piagetian, does not freely invent cognitive phenomena, but actually constrains the inventions on the basis of quite objective evidence. Such objective evidence can consist of at least one of the following: (1) Behavior occurring in accord with a certain pattern or a specific rule, (2) verbal self reports on cognitive activity, (3) behavioral discontinuity (as in ‘insight curves’), (4) long reaction times, (5) basic behavioral processes, such as ‘reinforcement,’ ‘extinction,’ ‘generalization,’ or ‘discrimination,’ or (6) the survival of behavior over time. As Reese (1994) also pointed out, no such evidence is very compelling by itself, and the inference that an unobservable cognitive event occurred is fundamentally a syllogism of the following form: Major premise: If a person performs A (a specified cognitive activity), then the person will exhibit B (e.g., a specified behavioral pattern). Minor premise: The person exhibits B. Conclusion: The person performed A. Although this involves the logical fallacy of affirming the consequent, it is the standard pattern of scientific inference, and it does not distinguish cognitivist from behaviorist inferences. (c) Behaviorist guidelines for inferring mediating behavior. The main distinguishing feature of interpretations in behavior analysis as compared with traditional cognitive psychology is that the terms and principles relied on in a behavioral interpretation are restricted to those obtained directly from the experimental analysis of behavior. An interpretation that includes private 167 events makes the provisions that the private event is probable, given current controlling variables, and that control by these variables is plausible considering the phylogenic and ontogenic histories of the organism (Palmer, 1991). Computer simulations of ‘cognitive’ phenomena are increasingly popular within the behavioral tradition but, then, the same criteria apply - no new principles may be introduced: the principles informing the program are restricted to those that are derived from research on the relevant biobehavioral processes. As Donahoe and Palmer (1994) recounted: “[Scientific] interpretation is a consumer, not a producer of principles” (p. 127). Why, then, would some researchers impose such restriction upon their own activities at odds with the psychological community at large and, definitely, at odds with the prestigious field of modern cognitive psychology? Problem solving as behavior: The utility of the behavioral view As pointed out by Catania (1992), when math problems are solved by paper and pencil, the intermediate products obviously serve as discriminative stimuli that may occasion the solution. Furthermore, “presumably, the intermediate products would still enter into the solution, even if there was no written record of them. If we did not say them aloud, an observer might say we had engaged in ‘mental arithmetic’ . . . But the role of the intermediate products is the same even if they are more public and more permanent in the first case than in the second. We still have much to learn about such processes, but we need not treat them as something other than behavior” (p. 348). However, the question for many psychologists remains: Why should we treat problem solving strictly in terms of behavior? ‘Reflection’ and ‘ratiocination,’ for instance, are not intuitively examples of behavior, and Harré (1988), for example, stated that “such problem solvings . . . are precisely not behavior” (p. 246). In reply, here are a few things that I believe will have to be seriously considered: 1. The case of negative information. The 168 Per Holth common finding that people learn more from positive instances than from negative ones, even when they are equally ‘informative,’ poses a particular problem for information processing theories. Wason (1960; see also Wason & Johnson-Laird, 1972), for instance, presented four cards with a letter on one side and a number on the other, with the upsides typically showing one vowel, one consonant, one odd, and one even number (such as A, D, 7, 4), and told his subjects to turn only the cards they would have to in order to find out if the following was true: “If there is a consonant on one side of the card, then there is an even number on the other.” Subjects rarely turned over the card(s) that could, in fact, have disproved the rule. This phenomenon was investigated in more detail by Dinsmoor (1983) in an experiment on observing behavior in pigeons. During a mixed EXT/ VR schedule of reinforcement, a pigeon can be taught to peck an additional key that produces stimuli correlated with each component schedule. In colloquial terms, the pigeon pecks one key to see if there’s any reason to peck the second key. But does it really? Dinsmoor modified the procedure so that the first peck only produced one of the stimuli – either the one correlated with extinction or the one correlated with variable ratio reinforcement. The results showed that the pigeon continued pecking on the first key only when it produced the stimulus that was correlated with reinforcement, even if the stimulus correlated with extinction was equally ‘informative.’ Hence, it is the reinforcing rather than the informative value of stimuli that affects observing behavior. This suggests that the basic idea of information processing may need to be reconsidered. 2. Interpretation of imaging, image tracing, and mental rotation. According to Kosslyn (1981), cognitivists generally hold the view that images are quasi-spatial entities generated from some store of perceptual experiences in long-term memory. As pointed out by Eysenck and Keane (1995), two worries accompany such symbolic-representation theories: First, they tend to become quite complicated, even with respect to simple tasks and, second, “we are left with no idea of how these symbols are repre- sented and manipulated at the neural level.” In contrast, the operant view is fairly simple. First, as expressed by Skinner: Seeing can occur “in the absence of the thing seen” (e.g., Skinner, 1974), but “so far as we know, nothing is ever seen covertly which has not already been seen overtly at least in fragmentary form.” This view is fully in accord with Kosslyn’s (1975, 1976) experiments on image tracing tasks, experiments on so-called ‘mental rotation’ by Shepard and coworkers (Cooper & Shepard, 1973; Shepard & Metzler, 1971), and so on. Second, the behavioral perspective makes no claim to even suggest what happens at the neural level, but is concerned with specifying the behavioral facts which the neurologist must have in order to complete the task. Moreover, as Skinner (1988, p. 337) pointed out, whatever symbolic representations may or may not occur, something called ‘seeing them’ would still be required. 3. Interpretation of introspective data and verbal protocols. The standard practice in cognitivist interpretations is to make some resigned reservations as to the accuracy of self reports. Kahney (1993), for instance, wrote that “the best we can hope for is information in a subject’s statements that permits us to infer that particular mental processes occur in a given task situation” (p. 54). Although the warnings of Nisbett and Wilson (1977) are often echoed, no serious efforts are made to tackle basic issues concerning the limits to self observation and its nature of origin in the first place. In addition to the obvious limits set by the structure of the nervous system, the behavioral view is that awareness of private events depends on some sort of correspondence between private and public events – accessible to the verbal community (see Skinner, 1957, p. 131ff). Hence, for instance, visual imaging must have something in common with regular seeing, just as auditory imaging must have common properties with regular hearing, and so on. This is actually an exciting view, because it suggests that the covert actions performed during problem solving, for instance, can be taught at the overt level. 4. Reinterpretation of ‘cognitive’ findings. Finally, if empirical findings in the cognitive literature on problem solving are reinterpreted What is a problem? in terms of behavior, there is a substantial body of basic research which may be immediately relevant. For instance, there are impressive bodies of research on complex stimulus control, multiple exemplar training, and on rule-governed and contingency-shaped behavior, which seem directly relevant, respectively, to the questions of the characteristics of ‘expert behavior,’ of variables that influence transfer of problem solving skills across tasks, variables involved in the heuristics or in the contingency shaping of ‘avoiding loops.’ According to Kahney (1993): The idea, simply put, is to try to figure out how to make ‘C’ students into ‘A’ students. That’s a worthy goal. But at the moment the goal has not been achieved. (p. 148) In conclusion, although a behavioral view of problem solving is clearly less intuitively attractive than cognitivist formulations, the behavioral view is more parsimonious, and it has an enormous advantage in its potential with respect to practical applications. I suppose that if Kahney’s goal is ever to be reached, cognitive psychology cannot afford to neglect the basic knowledge of behavioral principles that behavior analysis has to offer. References Aldrich, C. G. (1931). Experimental studies of idiot behavior. Training School Bulletin, 28, 151-159. Aldrich, C. G., & Doll, E. A. (1931). Problem solving among idiots. Journal of Comparative Psychology, 12, 137-170 Birch, H. G. (1945). The relation of previous experience to insightful problem-solving. Journal-of-Comparative-Psychology, 38, 367-383. Blough, D. S. (1959). Delayed matching in the pigeon. Journal of the Experimental Analysis of Behavior, 2, 151-160. Boring, E. G. (1957). A history of experimental psychology. New York: Appleton-CenturyCrofts. 169 Catania, A. C. (1973). The concept of the operant in the analysis of behavior. Behaviorism, 1, 103-116. Catania, A. C. (1991). The gifts of culture and of eloquence: An open letter to Michael J. Mahoney in reply to his article, “Scientific psychology and radical behaviorism.” The Behavior Analyst, 14, 61-72. Catania, A. C. (1992). Learning (3rd ed.). New York: Prentice Hall. Cooper, L. A., & Shepard, R. N. (1973). The time required to prepare for a rotated stimulus. Memory and Cognition, 1, 246-250. Daily, B. W. (1925). Ability to select data in solving problems. Teachers College Contributions to Education, 190, 1-103 Davis, G. A. (1966). Current status of research and theory in human problem solving. Psychological Bulletin, 66, 36-54. Davis, G. A. (1973). Psychology of problem solving. New York: Basic Books. Delaney, P. F., & Austin, J. (1998). Memory as behavior: The importance of acquisition and remembering strategies. The Analysis of Verbal Behavior, 15, 75-91. Dinsmoor, J. A. (1983). Observing and conditioned reinforcement. Behavioral and Brain Sciences, 6, 693-728. Donahoe, J. W., & Palmer, D. C. (1994). Learning and Complex Behavior. Boston: Allyn and Bacon. Duncker, K. (1945). On problem solving. Psychological Monographs, 58 (whole no. 270) Epstein, R., Kirshnit, C. E., Lanza, R. P., & Rubin, L. C. (1984). ”Insight” in the pigeon: Antecedents and determinants of an intelligent performance. Nature, 308, 61-62. Epstein, R., Lanza, R. P., & Skinner, B. F. (1980). Symbolic communication between two pigeons (Columba livia domestica). Science, 207, 543-545. Epstein, R., Lanza, R. P., & Skinner, B. F. (1981). ”Self-awareness” in the pigeon. Science, 212, 695-696. Ericsson, K. A., & Simon H. A. (1980). Verbal reports as data. Psychological Review, 87, 215-251. Eysenck, M. W. & Keane, M. T. (1995). Cognitive Psychology. A Student’s Handbook. East 170 Per Holth Sussex: Psychology Press. Forehand, G. A. (1966). Constructs and strategies for problem-solving research. In: B. Kleinmuntz, (Ed.) Problem Solving: Research, Methods and Theory. New York: John Wiley. Guilford, J. P. (1967). The Nature of Human Intelligence. New York: McGraw-Hill. Harlow, H. F. (1949). The formation of learning sets. Psychological Review, 56, 51-65. Harre, R. (1988). Psychology as moral rhetoric. In A. C. Catania & S. Harnad (Eds.), The Selection of Behavior. The Operant Behaviorism of B. F. Skinner: Comments and Consequences (pp. 259-260). New York: Cambridge University Press. Hayes, J. R. (1978). Cognitive psychology: Thinking and creating. Homewood IL: The Dorsey Press. Hayes, S. C., Barnes-Holmes, D., & Roche, B. (2001). Relational frame theory: A précis. In S. C. Hayes, D., Barnes-Holmes, & B. Roche (Eds.), Relational frame theory: A post-Skinnerian account of human language and cognition (pp. 141-155). New York, Plenum Press. Hayes, S. C., Fox, E., Gifford, E. V., Wilson, K., Barnes-Holmes, D., & Healy, O. (2001). Derived relational responding as learned behavior. In S. C. Hayes, D. Barnes-Holmes, & B. Roche (Eds.), Relational frame theory: A post-Skinnerian account of human language and cognition (pp. 21–49). New York: Plenum Press. Hayes, S. C., Gifford, E. V., Townsend, R. C., & Barnes-Holmes, D. (2001). Thinking, problem-solving and pragmatic verbal analysis. In S. C. Hayes, D. Barnes-Holmes, & B. Roche (Eds.), Relational frame theory: A post-Skinnerian account of human language and cognition (pp. 87–101). New York: Plenum Press. Hayes, S. C., White, D., & Bissett, R. T. (1998). Protocol analysis and the “silent dog” method of analyzing the impact of self-generated rules. The Analysis of Verbal Behavior, 15, 57-63. Holth, P. & Arntzen, E. (2000). Reaction times and the emergence of class consistent responding. The Psychological Record, 50, 305-337. Kahney, H. (1993). Problem Solving: Current Issues (2nd ed.). Buckingham: Open University Press. Kendler, H. H., & Kendler, T. S. (1962). Vertical and horizontal processes in problem solving. Psychological Review, 69, 1-16. Kosslyn, S. M. (1975). Information representation in visual images. Cognitive Psychology, 7, 341-370. Kosslyn, S. M. (1976). Can imagery be distinguished from other forms of internal representation? Evidence from studies of information retrieval times. Memory and Cognition, 4, 291-297. Kosslyn, S. M. (1981). The medium and the message in mental imagery: A theory. Psychological Review, 88, 46-66. Lebow, D., & Wager, W.W. (1994). Authentic activity as a model for appropriate learning activity: Implications for emerging instructional technologies. Canadian Journal of Educational Communication, 23, 231-144. Lowenkron, B. (1991). Joint control and the generalization of selection-based verbal behavior. The Analysis of Verbal Behavior, 9, 121-126. Lowenkron, B. (1996). Joint control and wordobject bidirectionality. Journal of the Experimental Analysis of Behavior, 65, 252-255. Lowenkron, B. (1998). Some logical functions of joint control. Journal of the Experimental Analysis of Behavior, 69, 327-354. Luchins, A. S. (1942). Mechanization in problem solving – the effect of Einstellung. Psychological Monographs, 54, 95. Luchins, A. S., & Luchins, E. H. (1950). New experimental attempts at preventing mechanization in problem solving. Journal of General Psychology, 42, 279-297. Maier, N. R. F. (1930). Reasoning in Humans: I. On direction. Journal of Comparative Psychology, 10, 115-143. Maier, N. R. F. (1931). Reasoning in Humans: II. The solution of a problem and its appearance in consciousness. Journal of Comparative Psychology, 12, 181-194. Neiworth, J. J., & Rilling, M. E. (1987). A What is a problem? method for studying imagery in animals. Journal of Experimental Psychology: Animal Behavior Processes, 13, 203-214. Newell, A., & Simon, H. A. (1972). Human problem solving. Englewood Cliffs, N.J.: Prentice Hall. Nisbett, R. E. & Wilson, T. D. (1977). Telling more than we can know: verbal report on mental processes. Psychological Review, 84, 231-259. Palmer, D. C. (1991). A behavioral interpretation of memory. In L. J. Hayes and P.N. Chase (Eds.) Dialogues on verbal behavior (pp. 261-279). Reno, NV: Context press. Peterson , L. R., & Peterson, M. J. (1959). Short-term retention of individual items. Journal of Experimental Psychology, 58, 193-198. Raaheim, K. (1988). Is there such a thing as a problem situation? In A. C. Catania & S. Harnad (Eds.), The Selection of Behavior. The Operant Behaviorism of B. F. Skinner: Comments and Consequences (pp. 259-260). New York: Cambridge University Press. Radford, J. K., & Burton, A. (1974). Thinking: Its nature and development. Chichester: John Wiley & Sons. Reed, S. K., Ernst, G. W., & Banerji, R. (1974). The role of analogy in transfer between similar problem states. Cognitive Psychology, 6, 436-450. Reese, H. W. (1994). Cognitive and Behavioral Approaches to Problem Solving. In S. C. Hayes, L. J. Hayes, M. Sato, & K. Ono (Eds.), Behavior analysis of language and cognition (pp. 197-258). Reno, NV: Context Press. Saugstad, P. (1977). A theory of communication and use of language. Oslo: Universitetsforlaget. Schlinger, H. D. (2002). Not so fast Mr. Pinker: A behaviorist looks at the Blank Slate. Behavior and Social Issues, 12, 75-79. Shepard, R. N., & Metzler, J. (1971). Mental rotation of three-dimensional objects. Science, 171, 701-703. Sidman, M. (1979). Remarks. Behaviorism, 7, 123 126. Sinnott, J. D. (Ed.) (1989). Everyday problem 171 solving: Theory and applications. New York: Praeger. Skinner, B. F. (1935). The generic nature of the concepts of stimulus and response. Journal of General Psychology, 12, 40-65. Skinner, B. F. (1953). Science and Human Behavior. New York: MacMillan. Skinner, B. F. (1957). Verbal Behavior. Englewood Cliffs, N. J.: Prentice Hall. Skinner, B. F. (1966). An operant analysis of problem solving. In: B. Kleinmuntz, (Ed.) Problem Solving: Research, Methods and Theory. New York: John Wiley. Skinner, B. F. (1968). The Technology of Teaching. New York: Appleton Century Crofts. Skinner, B. F. (1969). Contingencies of Reinforcement: A Theoretical analysis. Englewood Cliffs, NJ: Prentice Hall. Skinner, B. F. (1971). Beyond freedom and dignity. Middlesex: Penguin Books. Skinner, B. F. (1974). About Behaviorism. New York: Knopf. Skinner, B. F. (1978). Reflections on behaviorism and society. Englewood Cliffs, N.J.: Prentice Hall. Skinner, B. F. (1988). [Reply to Raaheim]. In A. C. Catania & S. Harnad (Eds.), The Selection of Behavior. The Operant Behaviorism of B. F. Skinner: Comments and Consequences (pp. 260-261). New York: Cambridge University Press. Small, W. S. (1900). An experimental study of the mental processes of the rat. American Journal of Psychology, 11, 133-165. Spearman, C. (1923). The Nature of ‘Intelligence’ and the Principle of Cognition. New York: Macmillian. Sternberg, R. J. (1977). Intelligence, information processing, and analogical reasoning: The componential analysis of human abilities. Hillsdale, N.J.: Lawrence Erlbaum Associates. Todd, J. T., & Morris, E. K. (1992). Case histories in the great power of steady misinterpretation. American Psychologist, 47, 1441-1453. Wason, P. C. (1966). Reasoning. In B. M. Foss (Ed.), New Horizons in Psychology (Vol. 1). Harmondsworth: Penguin. Wason, P. C., & Johnson-Laird, P. N. (1972). 172 Per Holth Psychology of reasoning: Structure and content. London: Batsford. Watson, J. B. (1920). Is thinking merely the action of language mechanisms? British Journal of Psychology, 11, 87-104. Wertheimer, M. (1945). Productive thinking. New York: Harper. Wildeman, D. G., & Holland, J. G. (1972). Control of a continuous response dimension by a continuous stimulus dimension. Journal of the Experimental Analysis of Behavior, 18, 419-434.