The Applicability of Computational Process Models For Representing Consumer Behavior

ABSTRACT - This paper presents an overview of computational process models. The first section of the paper reviews the use of these models in cognitive psychology and artificial intelligence and discusses the many benefits of using this methodology to understand human information processing. The next section reviews different areas of human cognition where these models have been developed, and then issues in developing and testing computational process models are discussed. Finally, three different computational process models of consumer behavior are presented.

Citation:

Andrew A. Mitchell and Terrence R. Smith (1982) ,"The Applicability of Computational Process Models For Representing Consumer Behavior", in NA - Advances in Consumer Research Volume 09, eds. Andrew Mitchell, Ann Abor, MI : Association for Consumer Research, Pages: 125-131.

Advances in Consumer Research Volume 9, 1982      Pages 125-131

THE APPLICABILITY OF COMPUTATIONAL PROCESS MODELS FOR REPRESENTING CONSUMER BEHAVIOR

Andrew A. Mitchell, Carnegie-Mellon University

Terrence R. Smith, Universit of California, Santa Barbara

[This research was partially funded by a grant from the Department of Transportation. The authors thank Eric Johnson for his comments on an earlier version of this paper.]

ABSTRACT -

This paper presents an overview of computational process models. The first section of the paper reviews the use of these models in cognitive psychology and artificial intelligence and discusses the many benefits of using this methodology to understand human information processing. The next section reviews different areas of human cognition where these models have been developed, and then issues in developing and testing computational process models are discussed. Finally, three different computational process models of consumer behavior are presented.

INTRODUCTION

The building of computational process (CP) models to represent and understand cognitive processes has become a standard methodological approach in cognitive science. In these models, cognitive processes are represented by a computer program which is generally written in a list processing language such as LISP (e.g., Winston and Horn. 1981). The use of this approach is based on the premise that both humans and computers are manipulators of symbols and that both can be described as a physical symbol system (e.g., Newell and Simon, 1976 Newell, 1980a). These systems are defined by a set of symbols and a set of rules defined over these symbols (e.g. Haugeland, 1981; Newell, 1980a).

Although this logic seems rather straightforward. there are many issues concerning the construction and use of these models. In the next section, we discuss two different approaches for developing CP models. We then present the benefits of using CP models and describe research areas where they have been applied. Finally, we discuss issues in wing CP models as well as examples of their use in consumer behavior research.

Artificial Intelligence and Cognitive Psychology

Computational process models are constructed and implemented in both artificial intelligence and cognitive psychology, although the approaches and goals characterizing the two areas differ. The goal in artificial intelligence is to develop programs that exhibit "intelligence" behavior, although the exact definition of intelligence is left vague. A loftier goal that is frequently stated, is the understanding of intelligent behavior in general (e.g., Winston, 1981). In this area, a problem is defined, (e.g., to develop a computer system that will beat a chess master), and then one or more researchers work to develop a working computer program that will solve the problem. No pretence is made that the resulting system is necessarily a representation of how humans solve the problem, although, in many cases, an examination of how humans solve the problem represents a first approximation to the resulting procedure. Consequently, some empirical studies are done but rarely do they involve tightly controlled experiments to test hypotheses. Instead, the purpose of the empirical studies is to give the researcher an indication of how humans solve the problem and then to use these ideas to develop a computer program.

In cognitive psychology, the goal is to understand aspects of human information processing activities such as the retrieval of information from long term memory or problem solving. Computational process models are used here as a theoretical representation of human cognitive processes. Tightly controlled experiments are used to build and test the resulting models.

Although the two approaches have different goals and employ different means for constructing models, they are closely related and ideas flow freely between them. For instance, Schank and Abelson's notion of a script, which evolved from artificial intelligence (Schank and Abelson, 1977 ), has been used as a hypothetical construct by psychologists (e.g., Bower, Black and Turner, 1980; Abelson. 1981).

Since researchers in artificial intelligence and cognitive psychology employ fundamentally different approaches. there have been numerous arguments concerning the validity of each approach ( e.g ., Miller, 1978). Research in psychology is criticized because it generally examines only subsets of the human information processing system. Mini theories are developed to explain these subsets; however, little effort is directed at integrating the mini-theories into a coherent understanding of the system. Research in psychology is also criticized because the resulting theories are never made sufficiently explicit for a computer program to be written representation of the theory. For instance, the research on visual imagery (e.g., Kosslyn, 1976; Kosslyn and Pomerantz, 1977) was originally criticized by researchers in artificial intelligence because Kosslyn never explained exactly how individuals construct a visual image (e.g., Pylyshyn, 1973). Later, however, Kosslyn developed a CP model of these processes (Kosslyn and Schwartz, 1977).

On the other hand, researchers in psychology criticize research in artificial intelligence because the resulting models are rarely tested empirically. Instead they tend to rely on tests of face validity and on the Turing test. This latter test involves having an outside observer monitor the output from the computer program to determine whether or not it can be distinguished from the behavior of a human. Colby (1973), for instance, developed a computer program that emulates the response of a paranoid individual and trained psychoanalysts were unable to distinguish these responses from the responses of an actual paranoid patient.

In general, research which involves the development of computational models can be divided into three general areas. The first area is represented by the work of cognitive psychologists who use CP models to understand human behavior and employ tightly controlled experiments to develop and test these models. The ACT model of memory developed by Anderson (1976) is an example of this approach. In the second area, a claim is made that the resulting models are representations of human behavior, although little or no effort is made to test the models empirically. The models developed by Schank (1980,1981) are examples from this area. The third area is the development of computer models for performing a particular task where no pretence is made that the resulting model is a model of human behavior. Examples of this research include DENDRAL (Buchanan, Sutherland, and Feigenbaum, 1969) a program for analyzing mass spectrogram data and MYCIN (Davis, Buchanan and Shortliffe, 1975) a program for aiding physicians in diagnosing and treating certain bacterial infections.

Benefits of Approach

The benefits of using CP models for understanding human behavior are numerous. First, they provide an explicit representation of our understanding of a particular cognitive phenomenon. As such, they provide a means for communicating and testing this understanding. The only other currently available means for representing this understanding are natural language and mathematical models. Natural language is very ambiguous and does not provide precise predictions of particular outcomes. Although mathematical models provide precise predictions, the use of algebra or calculus currently appears to be an inefficient way of representing this knowledge, since the resulting models are generally so complex as to be intractable.

Second, the use of a CP model forces one to be precise in thinking about a particular phenomenon. Usually, there are large gaps in our understanding of a particular phenomenon and frequently we are unaware of these gaps. It is only when we attempt to create an explicit representation of this knowledge, such as in constructing a CP model, do we become aware of these gaps.

Third, building CP models forces one into a more deductive mode of thinking. Our natural mode of learning is essentially inductive, and this carries over into research. There is always a tendency to gather data and to allow the data to guide our thinking, as opposed to thinking through the problem first and then designing an experiment to test specific hypotheses. In general, our understanding of a particular phenomenon will increase much faster if we adopt the latter approach.

Finally, we believe that many aspects of consumer behavior, such as decision making, involves many complex cognitive processes that are highly interactive. Any given study will provide only a limited number Of data points for understanding these processes. Consequently, programmatic research is required and the building of CP models provides a way for representing and testing the accumulated knowledge of this research. Failure to use this approach may result in studies which either examine only the outcome Of these processes or which do not provide an understanding of the cognitive processes which cause the behavior measured in the study.

In summary, we believe that CP models will prove useful in understanding a number of different aspects of consumer behavior. In addition, we believe that it is necessary to adopt this approach if we are to move from the very descriptive approach that characterizes most information processing research to one that provides an understanding Of the causal relationships underlying the generation of the data we observe. (e.g., Brucks and Mitchell, 1981).

Areas of Application for CP Models

In this section, we discuss four general areas of human cognition where CP models have been applied. These are perception and pattern matching, language and text com prehension, memory structures and processes, and decision making and problem solving.

In the area of perception and pattern matching, the goal has been to develop computer systems that are able to interpret information from the environment. Primary emphasis has been placed on interpreting visual information and identifying objects in the environment based on two dimensional patterns of light and dark. Research in this area and on animal vision indicates that considerable computation is involved in interpreting objects in our environment. Most objects, such as telephones, will look very different when examined from different perspectives and under different lighting conditions. Consequently, considerable knowledge about the objects to be identified is required by the system. Examples Of procedures and models developed in this area are by Horn (1974,1975), Marr (1975) and Waltz (1975).

Developing computer systems that understand natural language or language comprehension is another important area of research. Originally it was thought that these systems could be developed using only lexical and syntactic knowledge. It was soon found that general knowledge was also required, since much of our language is very content specific . Consequently, considerable attention has been directed at the different types of knowledge structures that are required in natural language processing. Some of these proposed knowledge structures include scripts (Schank and Abelson, 1977), memory organization packets (MOP's) and thematic organizational packets (TOP's) (Schank, 1981), schemata (Rumelhart and Ortony, 1977) and frames (Minsky, 1975). Examples of language comprehension systems include SHRDLU (Winograd, 1972) for written text and HEARSAY II (Lesser, Fennell, Erman and Reddy, 1975; and HARPY (Newell, 1980b) for spoken text.

Computational process models of memory structures and processes have been developed primarily by cognitive psychologists. Most of these models are based on the verbal learning tradition and have been concerned with how individuals process, store, and retrieve semantic information. In these models, this information is generally organized in networks and is represented by propositions. The work of Anderson and Bower (1973), Norman, et. al. (1975), and Anderson (1976) are examples of these types of models.

Finally, a number of CP models have been developed to represent decision making and problem solving. An early model in this area was the General Problem Solver of Newell, Shaw and Simon (1963). The construction of CP models in this area has provided an understanding of how individuals solve different types of problems and a general framework for understanding problem solving (Newell and Simon, 1972). Various problem solving procedures have been identified (i.e., means-ends analysis) and important concepts critical in understanding problem solving (i.e., problem space) have been identified. Recently, there has been much effort directed at understanding the differences between experts and novices in problem solving (e.g., Larkin, 1981) and developing CP models to represent learning in problem solving (e.g., Neves and Anderson, 1981).

Issues Concerning CP Models.

In this section we will consider five basic issues concerning computational process models. Although the issues are treated separately, they are clearly interrelated.

The first issue concerns the level at which cognitive processes are represented. Any cognitive activity occurs at a number of different levels simultaneously. These levels may range from the physiological level, the firing of neurons, to the level of conscious thought processes. CP models have been developed at a number of different levels. Examples of CP models at the level of conscious thought processes include the problem solving models of Newell and Simon, (1972), while examples of CP models at a lower level are the models of perception and pattern matching (e.g., Marr, 1975). Models of memory and retrieval -processes lie between these two levels. To date, little modelling is performed at the physiological level, although it has recently been argued that more attention should be paid to actual physiological processes in developing CP models (e.g., Hinton and Anderson, 1981).

An important question concerns the level at which CP models are most valid and what interactions between levels must be taken into account. Marr (1976) essentially argues that CP models are valid primarily at the levels of physiological processes. At higher levels, he argues, there are too many interactions between processes, and these interactions make it difficult to construct valid models. Pylyshyn (1980) takes a similar position in arguing that effort should be directed primarily at understanding human information processing at low levels. Anderson and Hinton (1981), on the other hand, argue that since most of the processes at lower levels are essentially parallel, it is not valid to represent such processes with serial CP models. Since higher level thought processes are essentially serial, however, CP models are valid at this level.

The second issue concerns the type of equivalence between computational process models and the human information processing system. Pylyshyn (1980), for instance, has argued for a strong equivalence between the two. Only when we do this, he argues. will we be able to make the greatest scientific advances. He suggests that there is a "functional architecture" of the mind which is equivalent to the notion of a virtual machine. This "functional architecture" lies between the physical and the representational level, is biologically based and is not affected by cognitive factors such as beliefs and goals. It is at this level. according to Pylyshyn, that we are most likely to discover causal relationships that provide an understanding of human behavior. Therefore, effort should be directed primarily at understanding the computational processes of this level and when we do this the resulting algorithms of the models will be equivalent to those cognitive processes.

This view of strong equivalence is not shared by most researchers in the area (e .g., see Colby, 1980; Grossman 1980; Haugeland 1980). Most would prefer to stay at a level of functional equivalence where the resulting model is a representation of cognitive processes.

A third issue concerns the scientific status of a computational process model. Here the argument centers on whether a computational process model is a theory. Moor (1978) has argued that a computational process model is not equivalent to a theory. He argues that theories must be stated in terms that are independent of the computer program. Schank ( 1980), on the other hand, argues that theories may not be able to exist independent of a computer program because the phenomena modelled are generally too complex.

The fourth issue concerns the testing of CP models. The problem arises because in most situations there are only a limited number of data points available for fitting the model. Consequently, the resulting model will have many more parameters than data points, and because of this, it will be theoretically possible for a number of different models to fit a given set of data well. For instance, Newell (1973), Anderson and Bower (1973), and Anderson (1976) have developed alternative models that provide a good fit to the data from Sternberg's scanning experiment (Sternberg. 1966).

Given this situation, Simon (1979) makes a distinction between necessary and sufficient conditions in model testing. If a computational process model is operational and is able to reproduce the available data, then it is a sufficient model. Recently, Newell (1980) has added that the structure of the model must also conform to what we know about the structure of the human processing system. For instance, it is well known that individuals can only attend to nine or fewer chunks of information at a time (e.g., Miller, 1956; Simon, 1958). Consequently, any CP model of human behavior requiring more than nine chunks of information to be attended to in active or short term memory would not be a sufficient model. In order for a model to satisfy the necessary conditions, it must be shown that no alternative model could fit the data. At present, it is probably impossible for any model to satisfy the necessary condition

There seems to be two possible ways out of this dilemma. The first is to discover as much as possible about the human information processing system and use this knowledge as constraints to reduce the number of possible CP models that will explain a given set of data. The second is to build general models that will explain human behavior in a number of different situations and then use tightly controlled experiments to test different aspects of the model. This is the approach used by Anderson (1976) in the development of the ACT model.

The final issue concerns the ability of CP models to explain all aspects of human behavior. For instance, these models have been criticized because they do not include emotional and -motivational factors. Simon (1969), however, has discussed ways in which CP models might include such factors. He suggests, for instance, that motivational factors might be integrated into these models through the use of goal structures and that some emotional reactions occur because the occurrence of a particular event does not conform to expectations. Recently, Bower and Cohen ( 1982), Bower (1981), and Smith, Mitchell and Meyer, 1982) have developed the designs for CP models which include emotion and affect, while other researchers have discussed the cognitive bases of emotion (e.g., Roseman, 1979; Weiner, 1952) . As of yet, however, there are no implemented models of which we are aware that include motivation. Although it is too early to know if motivation and emotion can be successfully incorporated in CP models we do not believe that CP models should be discarded because implemented models have not been developed which include these factors.

Many of these issues are clearly interrelated and center on the equivalency between the model and the human information processing system. As long as we do not demand strong equivalency from the model, many of these issues disappear. Any model of a physical process will be valid in some respects, but not in others (Pylyshyn, 1978,1981). The type of equivalency must be stated explicitly, however, in developing procedures for testing the model. For instance, if a particular cognitive process occurs in parallel it can be represented on a serial machine, although a comparison of the amount of time for subjects and the model to complete the execution of this process will probably not provide a valid test of the model.

Constructing and Testing CP Models

In developing a computational process model, two critical questions involve the selection of a particular architecture and the methods used for testing the model. The architecture of the model refers to the basic structure used in the model. Examples of different computational architectures are production systems (e.g., Davis and King, 1977; Newell, 1973), distributed processing systems (e.g., Minsky, 1980; Chandrasekaran, 1981) and memory activation (Anderson, 1976; McClelland and Rumelhart, 1981). The type of architecture selected will, of course, depend on the particular phenomena to be modelled

Production systems use a series of condition-act statements. When a particular condition is recognized (e.g., green light), a particular action is executed (e.g., cross the street). Consequently, this architecture is most appropriate for situations when the resultant cognitive processes are serial in nature. Examples of tasks where production system models have been applied are the solution of algebra problems (Neves and Anderson, 1972) and the Tower of Hanoi problem (Anzai and Simon, 1979).

In distributed processing systems a number of different processes occur in parallel. In developing these systems, the critical problem is to develop procedures by which the different subsystems may communicate with each other (e.g., Anderson and Hinton, 1981). This architecture is most appropriate for complex tasks that require rapid execution. Examples of problems where this architecture has been applied are planning (Hayes-Roth and Hayes-Roth, 1978) and speech understanding (Lesser, Fennel, Erman and Reddy, 1975).

Memory activation architectures are used in situations involving stored knowledge. Here external or internal cues activate portions of memory. This architecture has been used to examine the processing and retrieving of information (Anderson, 1976) and processes involved in reading (McClelland and Rumelhart, 1981)

Once the architecture for a problem has been selected and a model constructed, procedures are required to test the model. Four types of data are generally used for testing the model: protocols, response times, recall information and error rates. When building models of high level processes, such as those occurring in problem solving, protocols are generally used. These protocols can be analyzed to determine what type of strategy is used by the subJect and whether the strategy that is used is the one predicted by the model.

When examining lower level processes, such as memory retrieval, other measures need to be used since subjects generally do not have access to these processes. Hence response times, error rates and recall measures are the most useful. Generally, subjects are given a number of different conditions that will yield different response times, error rates and recall measures and then the model is run under these same conditions to determine if it yields the same pattern of differences

Computational Process Models and Consumer Behavior

In a previous paper (Brucks and Mitchell, 1981), two criticisms were directed at much of the current information processing research in consumer behavior. First, it was argued that much of this research was directed at describing processes instead of explaining them. For instance, research using the information board paradigm has found that information search strategies are very heterogeneous across individuals. Effort should now be directed at developing theories that will explain such differences.

Second, not enough research has been directed at understanding consumer behavior in "real world" situations. For instance, much of our research has examined information search and integration processes from information boards where verbal attribute information is presented for a number of different alternatives simultaneously. However, the presentation of information in this form rarely occurs in the environment.

Understanding consumer information processing activities in more "real world" environments will require a shift in focus to internal processes and memory structures. This shift in focus will probably require the use of CP models for understanding and representing internal processes and memory structures.

The papers that follow in this session represent examples of CP models for examining consumer behavior. Each paper examines a different aspect of consumer behavior and each model has a different structure or architecture. The first paper by Eric Johnson (1982) examines information search and integration processes within the information board paradigm. Since these processes are frequently constructive, he uses a production system architecture which consists of a series of condition-act statements. Here a specific action is executed (e.g., a specific search pattern) whenever a specific condition is recognized, As an individual searches for information, his or her knowledge about the alternatives change which, in turn, triggers a different production system which causes a different search pattern. A nice feature of his model is that it is not just a model of information search, but it also predicts what information will be recalled from memory after the search process.

The second paper is by Barbara Hayes-Roth (1982). In this paper she describes the cognitive processes involved in a purchase of a set of dishes. The processes that she described, which are probably very typical for many types of consumer purchases, take place over a period of months and are essentially opportunistic or constructive. The consumer acquires information about alternatives over time and formulates a new strategy for either more information search or for making a decision. This type of process has been modelled with the planning model that she has developed which uses a distributed processing architecture (Hayes-Roth and Hayes-Roth, 1980). Within this system there are different "experts" which, when called upon, make decisions based on the currently available information provided by other "experts."

The third paper, by Smith, Mitchell, and Meyer (1982), presents the structure of a model of the evaluation process. Within this model, the individual experiences a series of events and generalizations from these events may be formed into higher order memory structures. Each memory trace of an event has an emotional node (e.g., positive, negative or neutral affect) attached to it. If higher order structures are formed, evaluative nodes may also be attached to these structures. Evaluations, then, are formed by either activating the memory traces of the events or the higher order structures. Consequently, an activated memory structure architecture is used.

These papers all represent some initial work in using CP models to understand and represent different aspects of consumer behavior. The range of behavior that are explained by these models is very broad - ranging from information search strategies within the information board paradigm to constructive processes in planning a purchase. Obviously, CP models are also applicable in other areas of consumer behavior. For instance, we are currently working on a model of the cognitive processes that occur during exposure to an advertisement.

SUMMARY

In this paper, we have presented an overview of computational process models. We have provided examples of a number of different CP models and discussed their use in cognitive psychology and artificial intelligence. The benefits of using these models seems to be great. For instance, they are an explicit representation of cognitive processes and structures that yield empirical tests of the representation. They also are a way of identifying the gaps in our understanding of these processes and structures. Finally, they provide a means of accumulating and testing our knowledge of a particular cognitive phenomenon over a series of empirical experiments.

Even though there are clear benefits from using these models, there are also a number of philosophical issues surrounding them. These include the scientific status of the model, the equivalency between the model and the human information processing system and methods of model testing.

Finally, we provided a brief overview of three different computational process models of consumer behavior and discussed different areas in consumer research where these models might be useful.

REFERENCES

Abelson, R. P., (1981), "Psychological Status of the Script Concept," American Psychologist, 36, 715-729.

Anderson, J. A. and G. E. Hinton (1981), "Models of Information Processing in the Brain." In G. E. Hinton and J. A. Anderson (eds.), Parallel Models of Associative Memory, Hillsdale, NJ: Lawrence Erlbaum Associates. 9-48.

Anderson, J. R., (1981), Cognitive Skills and Their Acquisition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Anderson, J. R., (1976), Language, Memory, and Thought, Hillsdale, NJ: Lawrence Erlbaum Associates.

Anderson, J. R., and G. H. Bower (1973), Human Associative Memory, Washington, DC: Winston.

Anzai, Y. and H. A. Simon (1979), "The Theory of Learning by Doing," Psychological Review, 86, 124-140.

Boden, M. (1977), Artificial Intelligence and Natural Man. New York: Basic Books, Inc.

Bower, G. H. (1981), "Mood and Memory," American Psychologist, 36, 129-168.

Bower, G. H., J. B. Black and T. J. Turner (1979), "Scripts in Text Comprehension and Memory," Cognitive Psychology, 11, 177-220.

Bower, G. H. and P. R. Cohen (1981), "Emotional Influences in Memory and Thinking." In M. S. Clark and S. T. Fiske (eds.), Affect and Cognition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Brucks, M. and A. Mitchell, (1981), "Knowledge Structures, Production Systems and Decision Strategies." In K. Monroe Advances in Consumer Research, Vol. 8, Ann Arbor, MI: Association for Consumer Research.

Buchanan, B., G. Sutherland and E. A. Feigenbaum, (1969), "Heuristic DENDRAL: A Program for Generating Explanatory Hypotheses in Organic Chemistry" Machine Intelligence, 4.

Carbonell, J. (1978), "POLITICS: Automated Ideological Reasoning," Cognitive Science, 2, 27-52.

Chandrasekaran, B. (1981)," Natural and Social System Metaphors for Distributed Problem Solving." IEEE Systems: Man and Cybernetics, 11, 1-4.

Colby, K. M., (1980), "Human and Computer Rules and Representations are Not Equivalent," Behavioral and Brain Sciences, 3, 134-135.

Colby, K. M., (1973), "Simulations of Belief Systems." In R. C. Schank and K. M. Colby Computer Models of Thought and Language, San Francisco: W. H. Freeman and Company.

Davis. R., B. Buchanan and E. Shortliffe, (1975), "Production Rules as a Representation for a Knowledge-based Consulting Program", AIM-266, The Artificial Intelligence Laboratory, Stanford University, Stanford California.

Davis, R. and J. King (19775. "An Overview of Production Systems," Machine Intelligence, 8. 300-332.

Fahlman, S. E. (1979), NETL: A System for Representing and Using Real-World Knowledge, Cambridge, MA: MIT Press.

Fahlman, S. E. (1981), "Representing Implicit Knowledge," In G. E. Hinton and J. A. Anderson (eds.), Parallel Models of Associative Memory, - Hillsdale, NJ: Lawrence Erlbaum Associates, 145-159.

Feigenbaum, E. A. (1963), "The Simulation of Verbal Learning Behavior." In E. A. Feigenbaum and J. Feldman (eds.), Computers and Thought, New York: McGraw-Hill, 297-309.

Geschwind, N. (1980), "Neurological Knowledge and Complex Behaviors," Cognitive Science, 4, 185-193.

Grossman, S., (1980), "Human and Computer Rules and Representations are Not Equivalent", Behavioral and Brain Sciences, 3, 136-138.

Haugeland, J., (1981), "Semantic Engines: An Introduction to Mind Design." In J. Haugeland (ed .) Mind Design, Cambridge, MA: The MIT Press, 67-94.

Haugeland, J., (1980), "Psychology and Computational Architecture." The Behavioral and Brain Sciences, 3, 138-139.

Hayes-Roth, B., (1982), "Opportunities in Consumer Behavior." In A. Mitchell (ed.), Advances in Consumer Research, Vol. 9, Ann Arbor: Association for Consumer Research.

Hayes-Roth, B. and F. Hayes-Roth (1979), "A Cognitive Model of Planning," Cognitive Science, 4, 275-318.

Horn, B. K. P., (1975), "Obtaining Shape From Shading Information." In P. H. Winston (ed.), Advanced Psychology of Computer Vision, New York: McGraw-Hill.

Horn, B. K. P., ( 1974), "Determining Lightness from an Image," Computer Graphics and Image Processing, 3.

Johnson, E., (1982), "DECIDER: A Production System Model of Choice", Manuscript in preparation.

Klahr, D. and J. Wallace (1976), Cognitive Development: An Information Processing View, Hillsdale, NJ: Lawrence Erlbaum Associates.

Kosslyn, S.H. (1976), "Can Imagery be Distinguished from Other Forms Internal Representations? Evidence fran Studies of Information Retrieval Time," Memory and Cognition, 4, 291-297.

Kosslyn, S. M. and J. R. Pomerantz (1977), "Imagery, Propositions and the Form of Internal Representations," Cognitive Psychology, 9, 52-76.

Kosslyn, S. H. and S. P. Schwartz, ( 1977), "A Simulation of Mental Imagery", Cognitive Sciences, 1, 265-295.

Larkin, J. H., (1981), "Enriching Formal Knowledge: A Model for Learning to Solve Textbook Physics Problems." In J. R. Anderson (ed.) Cognitive Skills and Their Acquisition, Hillsdale, NJ: Lawrence Erlbaum Associates

Lesser, V. R. and L. D. Erman (1977), "A Retrospective View of the HEARSAY-II Architecture," Proceedings of the 5th International Joint Conference on Artificial Intelligence, Cambridge, MA, 790-800.

Lesser, V. R., R. D. Fennell, L. D. Erman, and D. R. Reddy (1975), "Organization of the HEARSAY-II Speech Understanding System," IEEE Transactions: Acoustics, Speech and Signal Processing, 23, 11-33.

Marr, D. (197?). "Artificial Intelligence--A Personal View," Artificial Intelligence, 9, 37-48.

Marr, D., (1974), "The Computation of Lightness by the Primate Retina," Vision Research, 14.

McClelland, J. L. and D. E. Rumelhart, (1981), "An Interactive Model of Context Effects in Letter Perception: Part I. An Account of Basic Findings," Psychological Review, 88, 375-407.

McCorduck, P. (1979), Machines Who Think, San Francisco: W. H. Freeman and Company.

Miller, G. A., (1956), "The Magic Number Seven," Psychological Review, 63.

Miller, L. (1978), "Has A. I. Contributed to an Understanding of the Human Mind," Cognitive Science, 2, 111-127.

Minsky, M. A. (1975), "A Framework for Representing Knowledge." In P. H. Winston, (ed.), The Psychology of ComPuter Vision, New York: McGraw-Hill.

Minsky, M. A. (1980), "K-lines: A Theory of Memory." Cognitive Science, 4, 117-133.

Moor, J. H. (1978), "Three Myths of Computer Science," British Journal of the PhilosoPhy of Science, 29, 213-222.

Neves, D. M. and J. R. Anderson (1981), "Knowledge Compilation: Mechanisms for the Automatization of Cognitive Skills." In J. R. Anderson, (ed.), Cognitive Skills and Their Acquisition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Newell, A., (1980a), "Physical Symbol Systems, Cognitive Science, 4, 135-183.

Newell, A. (1980b), "HARPY, Production Systems and Human Cognition." In R. Cole (ed.), Perception and Production in Fluent Speech, Hillsdale, NJ: Lawrence Erlbaum Associates.

Newell, A. (1973), "Artificial Intelligence and the Concept Of Mind." In R. C. Schank and R. M. Colby (eds.), Computer Models of Thought and Language, San Francisco: W. H. Freeman and Company.

Newell, A. (1973), "Production Systems: Models of Control Structures," In W. G. Chase (ed.), Visual Information Processing, New York: Academic Press.

Newell, A., J. C. Shaw, and H. A. Simon (1963), "Elements of the Theory of Human Problem Solving," Psychological Review, 65, 151-166.

Newell, A., and H. A. Simon, (1976), "Computer Sciences as Empirical Inquiry: Symbols and Search," Communications of the Association for computing Machinery, 19, 113-126.

Newell, A., and H. A. Simon, (1972), Human Problem Solving, Englewood Cliffs, NJ: Prentice-Hall.

Newell, A., and H. A. Simon (1963), "GPS, A Program that Simulates Human Thought." In E. A. Feigenbaun and J. Feldman (eds.), Computers and Thought, New York: McGraw-Hill.

Nilsson, N. J., (1981), "The Interplay Between Experimental and Theoretical Methods in Artificial Intelligence", Cognition and Brain Theory, 1, 69-74.

Norman, D. A. (1981), "Categorization of Action Slips," Psychological Review, 88, 1-15.

Norman, D. A., D. E. Rumelhart, and the LNR Research Group (1975), Exploration in Cognition, San Francisco: W. H. Freeman and Company.

Pylyshyn, Z., (1981), "Complexity and the Study of Artificial and Human Intelligence." In J. Haugeland (ed.) Mind Design, Cambridge, MA: a e MIT Press. 67-94.

Pylyshyn, Z., (1980), "Computation and Issues in the Foundation of Cognitive Science," Behavioral and Brain Sciences. 3, 111-169.

Pylyshyn, Z., (1978), "Computational Models and Empirical Constraints," Behavioral and Brain Sciences, 1, 93-99.

Pylyshyn, Z., (1973), "What the Mind's Eye Tells the Mind ' s Brain: A Critique of Mental Imagery," Psychological Bulletin, 80, 1-24.

Roseman, I., (1979), "Cognitive Aspects of Emotion and Emotional Behavior." Paper presented at the American Psychological Association Conference.

Schank, R. C., (1981), "Failure Driven Memory," Cognition and Brain Theory, 1, 41-60.

Schank, R. C., (1980), "Language and Memory," Cognitive Science, 4, 209-241.

Schank, R. C., and R. Abelson (1977), Scripts, Plans, Goals and Understanding, Hillsdale, NJ: Lawrence Erlbaum Associates.

Schank, R. C., and C. K. Riesbeck (1981), Inside Computer Understanding, Hillsdale, NJ: Lawrence Erlbaum Associates.

Searle, J. R. (1980), "Minds, Brains, and Programs," The Behavioral and Brain Sciences, 3, 417-424.

Simon, H. A., (1979), Information Processing Models of Cognition," Annual Review of Psychology, Palo Alto, CA: Annual Reviews, Inc., 363-396.

Simon, H. A., (1974), "How Big is a Chunk?" Science, 183. 268-288.

Simon, H. A., (1967), "Motivational and Emotional Controls of Cognition," Psychological Review, 74, 29-39.

Smith, T. R., R. Meyer and A. A. Mitchell, (1982), "A Computational Process model of Evaluation Based on the Cognitive Structuring of Episodic Knowledge." In A. Mitchell (ed.) Advances in Consumer Research, Vol. 9, Ann Arbor: Association for Consumer Research.

Sternberg (1966), "High Speed Scanning in Human" Science, 153, 652-654.

Turing, A. M., (1963), "Computer Machinery and Intelligence." In E. A. Feigenbaum and J. Feldman, Computer and Thought, New York: McGraw-Hill.

Waltz, D. (1975), "Understanding Line Drawings of Scenes and Shadows." In P. H. Winston (ed.), The Psychology of Computer Vision, New York: McGraw-Hill.

Weiner, B., (1982), "The E]notional Consequences of Causal Aspirations." In M. S. Clark and S. T. Fiske (eds.), Affect and Cognition, Hillsdale, NJ: Lawrence Erlbaum Associates.

Winograd, T. (1972), Understanding Natural Language, New York- Academic Press

Winston, P. H., (1977) Artificial Intelligence, Reading, HA: Addison- Wesley Publishing Co.

Winston, P. H. and B. P. Horn (1981), LISP, Reading, MA: Addison Wesley.

----------------------------------------