A Cognitive science approach for modelling representational organization and learning in humans

Abstract

Artificial intelligence (AI) is concerned with discovering intelligence in human mind and its application to the problems in real world. As a consequence it is associated with understanding the nature of human mind, that falls under the area of cognitive science. Researchers from different disciplines including psychology, linguistics, neuroscience, and philosophy have also contributed to the development of cognitive science. Cognitive scientific theories are leading increasingly to practical applications. Most of the applications are in the domain of education and learning. Other applications of cognitive science are more surprising. Contemporary theories of human memory have been applied to the question of the reliability of legal witness. This research has already led to important changes in the role of eyewitness testimony in the legal process. Another example is the application of new psychological theories of skill acquisition and visuomotor imagination to the design of training programs for atheletes. AI has led to immense application areas like expert systems, decision support systems, knowledge acquisition tools, intelligent tutoring systems, etc. This thesis is concerned with the study of two important aspects of human behaviour: the organization of mental representational structures and the role of inquisitiveness of human mind in learning. This thesis is in cognitive science that is relevant to AI stream, and thus our aim is to always keep the computer before us, eventhough we tackle the problems concerned with humans. To make the machine behave intelligently, it is essential to examine what are the factors that affect intelligence in humans. Intelligence, as discussed by Simon is a behavioral property, and when we consider behaviour the representation of the world around us and the ability to learn strongly affects this behaviour, and hence the intelligence. Our aim in this thesis, thus, is to examine the representational organization in humans and to see how it restricts the current computing systems to behave intelligently. W e also try to model the interactive aspect of the human mind that leads it to learn actively. We discuss these two problems in the first and the second parts of this thesis respectively. We briefly describe these two problems in the following. Representational organization : The first part The types and the organization of the representational structures available in an information processing system (human/computer) affect the performance of that system. To realize the goal of making computers behave as intelligently as humans do, it was necessary for us to understand the structures available and their organization in humans. In literature it has been accepted that there are two types of representational structures for decalarative knowledge. These are the propositional (p-) type that are fact like representations which specify formal relationships among concepts and their associated properties and the non-propositional (n-p) types, e.g., mental images, that depict the objects by resembling or mimicking them in some way that is hard to characterize. There are researchers that disagree to the existence of these two types of structures. According to them only p-type of representational structures are used. W e accepted the existence of both the p-type and n-p type of representational structures and looked into the organization problem. The above mentioned two types of representational structures can be organized into three possible configurations as shown in Fig 1.1. These three models of organizations are : 1. Imagistic propositional parallel process model, proposed with respect to mental imagery , also called as the outrace model. Both the structures are invoked parallely whenever a query is addressed. 2. Imagistic propositional serial model, in which the non- propositional representational structure is invoked first and then the propositional structure. 3. Propositional imagistic serial model in which the top level is the p-type representational structure and hence is invoked first and then the non-propositional level is invoked. We call this model as two-level model. • With the help of experiments we have validated the proposed two-level model. Learning : A manifestation of an interactive mind As a thesis on cognitive science our aim in this work is to identify various components of human learning and if possible to bring out a unifying model of learning thatincludes existing learning strategies. We aim to implement this model and show its application in the knowledge acquisition domain. Learning has a direct relation to understanding. Although the term understanding is extensively used in various discussions, there is no formal definition of the term till date. In general, it is assumed that once we have understood a concept then we don’t have to learn any more about that concept. W e thus also aim whether this generalized model of human learning can lead to a simplistic formalization of the term understanding. An important feature of human mind is its interactive and inquisitive nature. How this inquisitiveness affects the human learning process has not been discussed in the machine learning paradigms? What are the components of this inquisitiveness in humans? is an important question that can lead to a learner’s model. Psychlogists have strongly emphasised the active participation of learner in the learning process. In this work we explored the active aspect of human mind, i.e., the exploration of ones own knowledge, the act of thinking, questioning, daydreaming, etc. This exploration led us to what we call as Introspective learning (I-learning), a learning strategy based on learner’s inquisitiveness.