Российская академия наук институт проблем управления им. В. А. Трапезникова
 Скачать 1.88 Mb.
|
|
4. Conclusions The paper discussed a theory of cognitive control highlighting the cognitive mechanisms underlying the selection of human behav- iour. In order to allow a technical system to decide on its user’s next behavioural goal by mirroring the human cognitive mechanisms, a way of how the theory of cognitive control can be implemented was presented. References 1. BAILEY, D. M., DEFELICE, T. Evaluating movement for switch use in an adult with severe physical and cognitive im- pairments. American Journal of Occupational Therapy, 1991, 45(1), 76-79. 2. BARTOLEIN, C., WAGNER, A., JIPP, M., BADREDDIN, E. Multilevel intention estimation for wheelchair control. Proc. of the European Control Conf., Greece? 2007. 3. LANKENAU, A., RÖFER, T. (2000). Smart wheelchairs - state of the art in an emerging market. Zeitschrift Für Künstliche Systeme. Schwerpunkt Autonome Systeme, 2000, 14(4), 37-39. 4. DEMEESTER, E., NUTTIN, M., VANHOOYDONCK, D., & VAN BRUSSEL, H. A model-based, probabilistic framework for plan recognition in shared wheelchair control: Experiments and evaluation. IEEE/RSJ Intl. Conference on Intelligent Robots and Systems, Las Vegas, Nevada, 2003. 242 5. JIPP, M., BARTOLEIN, C., & BADREDDIN, E. An activity- theoretic approach to intention estimation. Submitted for publi- cation, 2008. 6. RASMUSSEN, J. Skills, rules, and knowledge: signals, signs and symbols, and other distinctions in human performance mod- els. IEEE SMC, 1983, 1(3), 257-266. 7. ANDERSON, J. R. ACT: A simple theory of complex cognition. American Psychologist, 1996, 51, 355-365. 8. Badreddin, E. (1995). Control and System Design of Wheeled Mobile Robots. Habilitationsschrift, ETH Zurich, Switzerland, 1995. 243 CONSTRUCT VALIDATION: THEORY OF STRUCTURED INTELLIGENCE Badreddin, E., Jipp M. (Automation Laboratory, University of Mannheim, Germany) badreddin@ti.uni-mannheim.de, mjipp@rumms.uni-mannheim.de Keywords: intelligence, innovation, experience, learning 1. Introduction 1.1. Motivation In many disciplines including cognitive sciences, psychology, or computer sciences lots of attention has been devoted is intelligence. This is why a variety of theories of intelligence have been developed in the past (for a summary, see [1]). These theories, however, are isolated approaches developed for the requirements of research in their underlying discipline. For example, in psychology, intelligence theories have – amongst others – been used in order to predict suc- cess at work or differentiate between potentially qualified applicants from unsuitable ones (e.g., [2]). The resulting intelligence constructs are significant predictors, however, do only classify “intelligent behaviour” in categories but do neither provide an explanation of intelligent behavior nor do they allow for implementation in technical systems. An exception is the theory of structured intelligence, which is described in [1, 4] and in Section 3.1. Although research on intelli- gence has been going on for many decades, there is no generally agreed upon a definition for intelligence and there is no common structured model. Both are, however, necessary for guiding cross- disciplinary research in developing e.g., intelligent decision support systems, efficient human-machine systems or expert systems. 1.2. A short history of research on intelligence The first global intelligence models are characterized by the as- sumption that intelligence is a homogeneous holistic aptitude to master a given situation. Later models added specific factors to this 244 general factor of general intelligence. Other researchers disapproved the existence of a general factor of intelligence and proposed a vari- able number of intelligence factors. These factors were ordered hierarchically by some researchers, while others arranged them on the same level of abstraction and negated a hierarchy between them. A more thorough overview and clear definitions are given in [3]. 2. Problem Statement For understanding and implementing intelligent behaviour, it is required to propose a model structure and parameter yielding an experimental platform for further interdisciplinary investigations. In order to yield such a theory, the Theory of Structured Intelligence is described in Section 3.1 and its relationship with other theories discussed (see Section 3.2). The latter is an important step in con- struct validation and crucial for yielding an accepted theory. 3. Solution Approach 3.1. Theory of Structured Intelligence The definition of intelligence, which the Theory of Structured Intelligence makes use of, assumes that the latent construct shows and determines the problem solving capability [1]: A problem is posed through the environment to the intelligent agent. The solution obtained from the intelligent agent is applied to the environment and, herewith, modifies the environment eventually posing new or addi- tional problems to the intelligent agent. This is why intelligence is measured in terms of a metric involving the number of problems and their complexity. Complexity measures are taken from classical complexity theory based on “space” and “time” required to solve a given problem. The theory’s main structure (as depicted in Fig. 1) assumes that solving a problem consists of two parts, a solution proportion which stems from experience/memorization and a solution proportion which is entirely new stemming from innovation. Both parts are combined and the resulting overall solution to the given problem is learnt, such 245 that the memorization “data-base” is updated. Accordingly, the theory makes use of four major blocks: Fig. 1. Theory of Structured Intelligence (adapted from [1]) - Memorization is a knowledge base of past solutions stored in an associative memory. It could, e.g., be realized on the basis of a neural network. - Innovation is a mechanism providing unpredictable solu- tions. The goal of innovation is to increase the entropy (entropy rate in connection with dynamic systems) of the solution and accordingly, provide a source for updating the memorization knowledge base, which otherwise would remain unchanged. A good model for the innovation process is a chaotic dynamical system (e.g., a double pendulum). - Fusion is an important mean for merging the solution propor- tions to the final solution, that is, the one applied to deal with the problem at hand (e.g., network of generalized binary operators). - Learning refers to the evaluation of a solution (i.e., was it successful or not successful) and updating the knowledge base used in the memorization accordingly. 3.2. Empirical and Theoretical Construct Validity of the Theory of Structured Intelligence In the following, the relationship of the proposed Theory of Structured Intelligence with other theories in the field is discussed, which is an important step in the classification of the newly devel- oped theory. Innovation Memorization Problem Innovative solution proportion Memorization-based solution proportion Fusion Solution Learning 246 3.2.1 Theory of Structured Intelligence and Theory of Cognitive Control The Theory of Cognitive Control [4] describes the cognitive mechanisms underlying action selection starting from routinised per- ception-motor responses in very familiar situations to problem solving mechanisms in unknown environments. In [4], it is argued how these mechanisms can be implemented in a technical system and used for intention estimation. In parallel to the Theory of Structured Intelli- gence, the Theory of Cognitive Control yields similar outputs: The first states that, while confronted with situations in which no standard solution is available, innovation-based processes determine behaviour or the solution. These innovation-based behaviours equal the creative actions proposed by the Theory of Cognitive Control. When con- fronted with a problem to which the solution is already saved in the knowledge base and no innovation-based propositions are required, routinized behaviour takes place, which is predictable, in contrast to the creative actions. However, while the Theory of Cognitive Control cannot explain why routinized behaviour is – presumably without reason – changed from the actor, the innovation-based processes of the Theory of Structured Intelligence provides appropriate means. 3.2.2 Theory of Structured Intelligence and Theory of Individ- ual Determinants of Skill Acquisition The Theory of Structured Intelligence has been put in theoretical and empirical relationship with Ackerman’s skill acquisition theory [5]. The theoretical and empirical relationships have been discussed in [6], which clearly demonstrate that the Theory of Structured Intel- ligence uses a similar concept of intelligence as do the Theory of Individual Determinants of Skill Acquisition and that its distinction between innovation- and memorization-based solutions can be found in the behaviour of human beings acquiring some skills. Especially, when the user is confronted with new situations, his/her behaviour is hardly predictable, while the predictability increases with his/her experience with the situation. 247 4. Conclusions Although research on intelligence has been going on for more than 100 years, a unified definition and information on its structure is not yet available. To provide such an approach, the Theory of Structured Intel- ligence has been introduced in [1]. To gain relevance especially in the different fields interested in intelligence, the theory must be tested and its adequacy proven in various settings. Especially in psychology, the approach of construct validation is an important mean for demonstrating the added value of newly developed theories. For this purpose, the Theory of Structured Intelligence has been put in relationship with the Theory of Cognitive Control and the Theory of Individual Determinants of Skill Acquisition. Similarities between these theories have been demonstrated. The added-value of the Theory of Structured Intelligence is its explanatory component stating how intelligent behaviour is ac- complished and the definition of ways of its implementation. Future work aims at implementing the proposed behaviours. References 1. BADREDDIN, E., & JIPP, M. Structured Intelligence. Interna- tional Conference on Computational Intelligence for Modelling, Control, and Automation. CIMCA 2006, Sydney, Australia. 2. KLEINE, D., JÄGER, A. O. Kriteriumsvalidität eines neuartigen Tests zum Berliner Intelligenzstrukturmodell. Diagnostica,1989, 35(1), 17-37. 3. CONRAD, W. Intelligenzdiagnostik. In: K.-J. Greffman, & L. Mischel (eds.). Enzyklopädie der Psychologie, 1983, p.104-201. 4. JIPP, M., BARTOLEIN, C., & BADREDDIN, E. Intention Esti- mation and the Concept of Cognitive Control. 7 Th International Conference on Cognitive Analysis, 2007, Moscow. Russia. 5. Ackerman, P. L. Determinants of individual differences during skill acquisition: Cognitive abilities and information processing. Jour- nal of Experimental Psychology: General, 1988, 117(3), 288-318. 6. Jipp, M., Badreddin, E. (2007). Theory of Structured Intelligence: Results on innovation-based and experience-based behaviour. In- ternational Conference on Informatics in Control, Automation and Robotics, Angers, France, 2007. |