Artificial Intelligence and Human Cognition - Essay Example

Running Head: ARTIFICIAL INTELLIGENCE AND HUMAN COGNITION Artificial Intelligence And Human Cognition [The Writer’s Name] [The Name of the Institution] Artificial Intelligence And Human Cognition Contemporary psychology, in broad outline, is dominated by three general viewpoints or theories: evolutionary theory, psychoanalytic theory, and computational theory. In evolutionary theory, the environment reinforces certain behaviors of the organism, resulting in survival. Behavior that is reinforced is adjustive and adaptive to the environment. In psychoanalytic theory, adaptation and adjustment result from the harmonious integration of dynamic aspects of the personality. Conflicts between motives and values result in anxiety. Defense mechanisms control anxiety, but at the cost of producing rigidity and neurosis in the personality. In computational theory, the mind is viewed as mechanism. The mechanisms of human thought can be described mathematically. Reasoning is but reckoning. Problem solving is but calculating. The computational theory of psychology finds its model in artificial intelligence, the science that holds that computers, by virtue of their mathematical structure, can reason. Artificial intelligence does not require the computer to understand what it is reasoning about. The reasoning mechanism is a calculus indifferent to its content. In contrast, the calculus of human thought is, as demonstrated in the psychoanalytic theory of psychology, distinctly responsive to the content and import of personal ideation. (Hall, 2004) Thus, at most, artificial intelligence can model only the mechanics of human reasoning and human problem solving. The mathematical descriptions of human thought and computer thought may approach an identical form. Such a universal mathematical description of reasoning and problem solving can be valuable for both human psychology and artificial intelligence. Advances in knowledge of the mechanisms of thought in one domain benefit the other domain as well. Just as mathematical description is a language expressing the essentiality of relationships between theoretical variables, so symbolic logic in the language of the propositional and predicate calculus expresses the essentiality of the structure or architecture of thought; and just as mathematical symbols can be manipulated instead of manipulating physical reality, so the logical calculus can be manipulated instead of manipulating cognitive reality. (Nilsson, 2001) The language of mathematics is to physical reality as the language of the predicate calculus is to cognitive reality. The mapping of cognitive reality in a general programming language that would accommodate the general structure or architecture of thought has been attempted by a number of cognitive scientists. An important example of John Anderson's unitary theory of cognition: I would like to head off two possible misinterpretations of my position. First, the unitary position is not incompatible with the fact there are distinct systems for vision, audition, walking, and so on. My claim is only that higher-level cognition involves a unitary system. Of course, the exact boundaries of higher-level cognition are a little uncertain, but its contents are not trivial; language, mathematics, reasoning, memory, and problem solving should certainly be included. Second, the unitary position should not be confused with the belief that the human mind is simple and can be explained by just one or two principles.( Anderson, 2003, p. 5) A different, more abstract, and inclusive general unified theory of intelligence can be formulated on the basis of the logic of implication. This fundamental theorem of intelligence would hold that the logic of implication (if p, then q) subsumes both the formal structure of human reasoning and problem solving and the formal structure of artificial intelligence. The logic of implication is foundational to mathematical and scientific reasoning and to the reasoning of everyday behavior as well, and is foundational to programming logic and knowledge representation formalisms in artificial intelligence systems. The production system consists of three modular elements: a global database, a set of production rules, and a set of control structures. The modularity of the elements allows for their modification without any complicating interaction effects. The content of the elements consists of encoded knowledge in a given problem domain. Production rules are composed of condition--action pairs. Satisfaction by the database of the conditions of production rules instigates their operation. The determination of the specific sequence of production rules in a cycle of operations is a major function of the control structures. In applying production systems to problem solving, pathways through the problem space (the set of possible problem states) are searched until the goal state is achieved. The sequence of operations of the production system directs a search trajectory. Trajectories are mapped onto a search tree structure consisting of nodes that represent problem states and directed arcs that represent production rules. The emulation of cognitive processes by artificial intelligence systems is a major theme being discussed in research circles. Systematic comparisons of similarity and contrast between human thought and computer thought are drawn, and the distinctive qualities of intellectual performance that characterize each mode of cognition are examined in depth. The intellectual contributions of artificial intelligence to mathematical reasoning may be exemplified by the problem of proving the four-color conjecture. The establishment of this proof had eluded the best mathematicians for over a century, and required the acceptance of the computer as a genuine contributor to mathematical reasoning. The section concludes with a comparison of mathematical theorem proving by mathematicians and by computers. Modern experimental psychology has regularly studied human rationality and has used the normative standards of logic and mathematics to evaluate everyday reasoning. Formal constructs of rationality vary with respect to their existence as self-contained systems, and with respect to impingement on and interaction with the empirical concerns of science and everyday reasoning. Systems of formal mathematics and formal logic may be essentially autonomous deductive structures with little or no interpretive applications. Systems of mathematics may have intimate and mutually dependent relationships with theoretical and applied science. Mathematical models may be viewed as conceptual tools directed toward the improvement of understanding in physics, economics, and psychology. As discussed above, mathematical modeling of psychological reasoning is especially difficult, and the balancing of formal criteria and heuristic criteria in judging the quality of human rationality constitutes, in itself, a significant and formidable problem in rational analysis. As self-involved reasoners investigating our own rationality, we may be forever limited by self-recursive cognition and the absence of a comparative standard. However, comparison with the rationality of artificial intelligence may, to some extent and from certain perspectives, obviate the limitation. The intellectual range has been broad: from a comparison of mathematical proof by mathematicians and by computers, to a comparison of legal reasoning by human jurisprudence and computer jurisprudence; from the nature of rationality in mathematical models of economic behavior to the nature of rationality in probabilistic models of psychological judgment; (Carbonell, 2001) from the rationality of simple mathematical description to the rationality of theoretical mathematical analysis; from an analysis of computational approaches to problem solving in the natural sciences to an analysis of problem solving in medical diagnosis; and from a computer modeling of scientific discovery processes in biochemistry to a computational analysis and comparison of sentential and diagrammatic representation in information processing systems. The intellectual ability of the computer has increased in power and in scope, but its coextensivity with problems to which the human mind has been applied is still in the visible future. Moreover, the human mind is continuously extending its intellectual range and depth, and its accomplishments do not remain fixed. It is clear that intellectual advancement will characterize both the realm of human intelligence and the realm of artificial intelligence and that a cooperative and synergistic relationship between the realms will culminate in greater advancement than could be achieved by the separate endeavors of either. Logical approaches to artificial intelligence have demonstrated the power of axioms and deductive systems in the domains of scientific and mathematical reasoning, but the domain of commonsense reasoning contains formidable difficulties in that people do not employ theoretical knowledge of physics or psychology in their everyday reasoning about the physical or social world. Yet, people appear to have a satisfactory working knowledge of events and processes in the physical world. Mathematics and mathematical logic have been largely responsible for advances in the physical sciences. Whether mathematical logic can serve an analogous role in the development of the sciences of cognition remains to be seen. It may be that nonstandard forms of mathematical logic will have to be created to capture human thinking, which is more often finesse than ratiocination. (Sternberg, 2003) The field of artificial intelligence, a specialized discipline within general computer science, is directed toward the continuous augmentation of computer intelligence. The augmentation of intelligence in computers may be achieved by two general methods or by a combination of the methods. In the first general method, the computer models the cognitive processes of human intellect. Augmentation of computer intelligence through this method requires the continuous expansion of reliable and valid knowledge concerning human cognitive processes. In the second general method, the intelligence of the computer models formal logical structures and processes. Augmentation of computer intelligence through this method requires the continuous expansion of reliable and valid knowledge concerning the theory and application of systems of logic and coordinated sets of programming languages. (Cussins, 2000) These two general methods of augmenting computer intelligence depend for their physical realization on the continuous expansion of knowledge in the field of computer engineering. Improvements in computer engineering design and materials (for example, from serial to parallel processing, from electronic to optical circuitry) optimize the results of the application of the two general methods of artificial intelligence.( Lenat, 2003) In addition to the engineering objective of the augmentation of the intelligence of computers, the field of artificial intelligence also has a scientific objective concerned with the development of a general theory of intelligence. This abstract science of intelligence would systematically establish the general principles, commonalities, and singularities of human, animal, and computer intelligence. As is true of any science, it is important to inquire into the conceptual foundations of artificial intelligence. Because knowledge of the universe is the product of the human mind, discovery of the nature of the mechanics of the human mind would enable purely conceptual procedures to formulate the entire science of physics. Hypotheses have long before been developed in cognitive psychology and artificial intelligence that resulted in specific knowledge of the operations of the human mind. These advances in cognitive science have impressed the eminent theoretical physicists. The imitation game is a test of parity between computer intelligence and human intelligence and it is an inadequate criterion of psychological conjecture whose limits must be tested by the usual procedures of science. (Wagman, 2003) The computability of the brain and its intelligence must stand or fall in the outcome of empirical inquiry into neural and cognitive processes. (Haugeland, 2001) In cognitive psychology and artificial intelligence, production rules are theoretical variables that control the representation and processing of information or knowledge. The role of production rules in the human information processing system and their distinction from the stimulus response variables or behaviors are set forth in accounts. The general unified theory of intelligence describes the concept of production system as an implementation of logical implication. In turn, the production system concept is fundamental in artificial intelligence in that the many distinct methods of knowledge representation in artificial intelligence can all be reduced to the theory of production systems. A production system consists of a set of rules, each consisting of a left side (a pattern) that determines the applicability of the rule and a right side that describes the operation to be performed if the rule is applied. One or more knowledge/databases that contain whatever information is appropriate for the particular task. Some parts of the database may be permanent, while other parts of it may pertain only to the solution of the current problem. (Greiner, 2002) A central issue in comparing human and artificial reasoning concerns the importance of abstract rules of inference. Clearly, insofar as artificial intelligence employs a predicate calculus, its reasoning is abstract. It is also the case that logicians and mathematicians in their formal professional work use systems of deductive inference. A number of theoretical positions in psychology -- including variants of case based reasoning, instance-based analogy, and connectionist models -- maintain that abstract rules are not involved in human reasoning, or at best play a minor role. Other views hold that the use of abstract rules is a core aspect of human reasoning. There are eight criteria for determining whether or not people use abstract rules in reasoning, and examine evidence relevant to each criterion for several rule systems. There is substantial evidence that several different inferential rules, including modus ponens, contractual rules, causal rules, and the law of large numbers, are used in solving everyday problems. (Burstein, 2002) There are multiple implications for various theoretical positions and consider hybrid mechanisms that combine aspects of instance and rule models. A general unified theory of intelligence would need to be inclusive of a dual typology of human cognition and a dual typology of computational models. Deliberative human thought, as represented by reasoning, problem solving, planning, judgment, and decision-making, is distinguished from automatic human memory, as represented by retrieval and recognition processes. The first set of cognitive activities is best modeled by symbolic computational models, the second by connectionist computational models. A complete account of cognition will require an integration of the symbolic and connectionist architectures. In the general unified theory of intelligence, both the cognition and the models have their ultimate foundation in the logic of implication. It is important that research in the psychology of logical reasoning follows the principles of logic that govern the complex relationships between the validity-invalidity of arguments and the truth-falsity of premises and conclusions. Invalid argument patterns can consist of true premises and true conclusions, true premises and false conclusions, false premises and true conclusions, or false premises and false conclusions. Valid argument patterns can consist of true premises and true conclusions, false premises and true conclusions, or false premises and false conclusions. The pattern of valid arguments, true premises, and false conclusions is not extant in the principles of logic. References Anderson J. R., C. F. Boyle, and B. J. Reiser. 2003. Intelligent Tutoring Systems. Science 228: 4-6. Burstein M. 2002. Concept Formation By Incremental Analogical Reasoning And Debugging. In Machine learning: An artificial intelligence approach, ed. R. S. Michalski, J. G. Carbonell, and T. M. Mitchell, pp. 351-370. Los Altos, CA: Morgan Kauffmann. Carbonell J. G. 2001. Learning by analogy: Formulating and generalizing plans from past experience. In Machine learning: An artificial intelligence approach, ed. R. S. Michalski, J. G. Carbonell, and T. M. Mitchell, pp. 137-162. Palo Alto, CA: Tioga. Cussins A. 2000. The connectionist construction of concepts. In The philosophy of artificial intelligence, ed. M. Boden. Oxford: Oxford University Press. Greiner R. 2002. Learning by understanding analogies. Artificial Intelligence 35: 81-125. Hall R. P. 2004. Computational approaches to analogical reasoning: A comparative analysis. Artificial Intelligence 39: 39-120. Haugeland J. 2001. Semantic engines: Introduction to mind design. In Mind design: Philosophy, psychology, and artificial intelligence, ed. J. Haugeland. Cambridge, MA: MIT Press. Lenat D. B., and E. A. Feigenbaum. 2003. "On the thresholds of knowledge". Artificial Intelligence 47: 185-250. Nilsson N. J. 2001. "Logic and artificial intelligence". Artificial Intelligence 47: 31-56. Sternberg R. J. 2003. Intelligence, information processing and analogical reasoning: The componential analysis of human abilities. Hillsdale, NJ: Erlbaum. Wagman M. 2003. "Cognitive psychology and artificial intelligence: Theory and research in cognitive science". Westport, CT: Praeger. Read More