Physical Symbol System Hypothesis Research Articles

Conscious AI at the Edge of Chaos

The main problem for AI consciousness is to operate within the right kind of AI. We distinguish between the traditional computing (GOFAI), and the computing based on stochastic pattern optimization. The latter will be called here computing at the edge of chaos. Optimization of learning patterns, which is the gist of its success, often happens between the areas of too much repetitive order and those of hard to predict and control stochastic processes. This is to change the focus from the opposition of symbolic versus sub-symbolic computing; symbols can appear at different granularities and the hedge between The Physical Symbol System Hypothesis and neural nets seems no longer the most productive cut to make. Computing at the edge of chaos is promising for AGI, especially for AGI consciousness. The second problem for AI consciousness is to work with the right definitions of consciousness.

The Dynamical Hypothesis in Cognitive Science: A Review Essay of Mind As Motion

Sometimes it is hard to know precisely what to think about the Dynamical Hypothesis (hereafter, DH), the new kid on the block in cognitive science and described succinctly by the slogan "cognitive agents are dynamical systems" (Van Gelder, 1998). Is the DH a radically new approach to understanding human cognition? (No.) Is it providing deep insights that traditional symbolic artificial intelligence overlooked? (Certainly.) Is it providing deep insights that recurrent connectionist models, circa 1990, had overlooked? (Probably not.) Is time, as instantiated in the DH, necessary to our understanding of cognition? (Certainly.) Is time, as instantiated in the DH, sufficient for understanding cognition? (Certainly not.) The DH is often touted as being a revolutionary alternative to the traditional Physical Symbol System Hypothesis (PSSH, renamed in Mind As Motion, the Computational Hypothesis) (Newell and Simon, 1976) that was the bedrock of artificial intelligence for 25 years. We disagree. First, as pointed out by the authors themselves, the use of dynamics as a framework to comprehend the brain and cognition is not new. In the early 1950’s, for example, W. Ross Ashby wrote a wonderful little treatise called Design for a Brain (1952) based on the recurrent, dynamical nature of the brain. The whole field of cybernetics (Wiener, 1948), developed during the late 1940’s, was also deeply concerned with feedback and stability in complex evolving systems. The explicit goal of these authors and others was to develop a general framework within which biological and cognitive systems could be studied and understood. These efforts, while vision in scope, did not bear fruit, in part because the systems they hoped to understand – in particular, human brain function and cognitive activities – were extraordinarily complex and there were no realistic means of empirically testing the ‘hypotheses’ generated by this approach to cognition.

Connectionist Modelling and Education

Introduction The main aim of this paper is to describe and explain some recent work on artificial neural networks (ANNs) that should be of interest to researchers in educational studies. These networks are also known more generally in the literature as connectionist systems. Although models of cognition would be of obvious relevance to those seeking to understand teaching and learning, a subsidiary aim of the paper will be to show that ANNs are relevant to the study of a wider range of educational phenomena. The plan of the discussion is as follows. First, the development of ANN models is located within the context of the rise of what might usefully be called the `new cognitive science'. Then a taxonomy of the main models is provided, together with a fairly detailed account of the workings of one particular type. Three quite diverse applications are then discussed and their implications for education canvassed. Finally the significance of ANN models for representing both static and dynamical aspects of knowledge is touched upon. Since applications involve computer simulations, there will also be mention of a number of helpful software packages. Traditional cognitive science Fifty years ago, in a highly influential paper, the English mathematician Alan M. Turing (1950) sought to answer the question `Can machines think?' (p. 40). After transforming that question into one that was more manageable, he answered in the affirmative, speculating that `in about fifty years time' developments would be such that an interrogator would have only a 70 per cent chance of correctly distinguishing an (unseen) computer from a human after five minutes of questioning (p. 49). Despite the offer of a generous prize--the Loebner Prize, worth US$100,000, and a gold medal--and restrictions on the domain of questioning, no computer has been adjudicated as thinking since the annual contest commenced in 1991 (although adjudicators have classified a few humans as computers, Dennett, 1998, pp. 27-29). Matters of prediction aside, Turing's lingering influence on traditional conceptualisations of cognition, or perhaps his capturing of a widely shared intuition, lay in the way he transformed the original question. His first strategy was epistemological and amounted to a behavioural test for thinking: the Turing Test. How could we ever tell if a machine was thinking? Answer: by asking it unrestricted questions. If it successfully imitates human answers--plays well the imitation game--then we may say it is thinking (Turing, 1950, pp. 40-43). His second strategy was ontological and involved a reconstrual of what we might regard as a machine. Since he was concerned, not with whether some particular machine could think, but with `whether there are imaginable computers which would do well [at imitation]' (p. 43), an approach was proposed that abstracted almost entirely from physical detail beyond requiring adequate (program) storage space and suitable speed of operation. The upshot is that these two strategies conspire to yield a functionalist theory of mind, a very general set of constraints for determining the adequacy of accounts of cognition. The way it works is as follows. Roughly speaking, sets of inputs and outputs constitute the empirical evidence of cognition. Often winnowed selectively to comprise words, or other symbolic tokens, written or spoken, that elicit behaviours also often in the form of produced symbol tokens, cognitive acts are construed as the set of transformations that connect these inputs to their associated outputs. The best known general formulation of the class of intelligent cognitive acts is given by Newell and Simon (1976) in what they call `the Physical-Symbol System Hypothesis' which states that `a physical-symbol system has the necessary and sufficient means for general intelligent action' (p. 111). Construed as an empirical hypothesis, they are claiming essentially that the basis for intelligent action is to be found in the processes that operate on sets of symbols, or `symbol structures' to produce other symbol structures. …

Why the dynamical hypothesis cannot qualify as a law of qualitative structure

Van Gelder presents the dynamical hypothesis as a novel law of qualitative structure to compete with Newell and Simon's (1976) physical symbol systems hypothesis. Unlike Newell and Simon's hypothesis, the dynamical hypothesis fails to provide necessary and sufficient conditions for cognition. Furthermore, imprecision in the statement of the dynamical hypothesis renders it unfalsifiable.

Fundamentals of representation I: A traditional philosophy

Abstract Assumptions about the nature of representation can be expected to have considerable influence on the way in which the project of computational representation is approached. This paper examines a traditional philosophy of representation which has driven much of the project up to now. The use of propositional and predicate logic, and of various notational variations on that logic such as semantic nets, conceptual dependency, frames and scripts, has presupposed this philosophy—as indeed has the physical symbol system hypothesis of AI. This traditional philosophy is expounded and its weaknesses exposed, leading to an alternative philosophy of representation which might be expected to prove particularly helpful in the field of law.

Situated action, symbol systems and universal computation

Vera & Simon (1993a) have argued that the theories and methods known as “situated action” or “situativity theory” are compatible with the assumptions and methodology of the physical symbol systems hypothesis and do not require a new approach to the study of cognition. When the central criterion of computational universality is added to the loose definition of a symbol system which Vera and Simon provide, it becomes apparent that there are important incompatibilities between the two approaches such that situativity theory cannot be subsumed within the symbol systems approach. Symbol systems and situativity theoretic approaches are, and should be seen to be, competing approaches to the study of cognition.

Cognition — perspectives from autonomous agents

The predominant paradigm in cognitive science has been the cognitivistic one, exemplified by the “Physical Symbol Systems Hypothesis”. The cognitivistic approach generated hopes that one would soon understand human thinking — hopes that up till now have still not been fulfilled. It is well-known that the cognitivistic approach, in spite of some early successes, has turned out to be fraught with problems. Examples are the frame problem, the symbol grounding problem, and the problems of interacting with a real physical world. In order to come to grips with the problems of cognitivism the study of embodied autonomous systems has been proposed, for example by Rodney Brooks. Brooks' robots can get away with no or very little representation. However, this approach has often been criticized because of the limited abilities of the agents. If they are to perform more intelligent tasks they will need to be equipped with representations or cognition — is an often heard argument. We will illustrate that we are well-advised not to introduce representational or cognitive concepts too quickly. As long as we do not understand the basic relationships between simple architectures and behavior, i.e. as long as we do not understand the dynamics of the system-environment interaction, it is premature to spend our time with speculations about potentially useful architectures for so-called high-level processes. This paper has a tutorial and a review aspect. In addition to presenting our own research, we will review some of the pertinent literature in order to make it usable as an introduction to “New AI”.

Reconstructing Physical Symbol Systems

In attempting to force ALVINN1 into their already bulging symbolist tent, Vera and Simon (1993) have burst the seams of the Physical Symbol System Hypothesis (Newell and Simon, 1976; Newell, 1980a). We would like to present another view of representation in connectionist networks and respond to statements about ALVINN by both Vera and Simon and Greeno and Moore (1993). In so doing, we will reconstruct the PSSH.

The case for research in decision support systems

The ‘Physical Symbol System Hypothesis’ proposed by Newell and Simon states that any system capable of manipulating a set of symbols is also capable of intelligent action. Newell and Simon assert that this hypothesis forms the basis for research in artificial intelligence (AI). The belief is that computers are just as capable of intelligent action as are humans. This paper argues that the basis for research in decision support systems (DSS) contrasts sharply with that of AI in that DSS research assumes that a man-machine symbiosis is capable of a higher level of intelligent action than either man or machine operating independently.

Approaches to the study of intelligence

How can human and artificial intelligence be understood? This paper reviews Rosenbloom, Laird, Newell, and McCarl's overview of Soar, their powerful symbol-processing simulation of human intelligence. Along the way, the paper addresses some of the general issues to be faced by those who would model human intelligence and suggests that the methods most effective for creating an artificial intelligence might differ from those for modeling human intelligence. Soar is an impressive piece of work, unmatched in scope and power, but it is based in fundamental ways upon Newell's “physical symbol system hypothesis”—any weaknesses in the power or generality of this hypothesis as a fundamental, general characteristic of human intelligence will affect the interpretation of Soar. But our understanding of the mechanisms underlying human intelligence is now undergoing rapid change as new, neurally-inspired computational methods become available that are dramatically different from the symbol-processing approaches that form the basis for Soar. Before we can reach a final conclusion about Soar we need more evidence about the nature of human intelligence. Meanwhile, Soar provides an impressive standard for others to follow. Those who disagree with Soar's assumptions need to develop models based upon alternative hypotheses that match Soar's achievements. Whatever the outcome, Soar represents a major advance in our understanding of intelligent systems.

Is intelligent computer-assisted language learning possible?

The limitations of existing CALL software [Self, J. (1985) Microcomputers in Education. Harvester Press] and the extravagant claims of achievements in artificial intelligence [Schank, R. (1984) The Cognitive Computer: On Language, Learning and Artificial Intelligence. Addison-Wesley] together with rapidly increasing power/price ratio of microcomputers has led to much speculation about the development of intelligent computer-assisted learning. Intelligent computer-assisted learning systems consist of four modules: a natural language interface, a pedagogical module, a model of the student, and a knowledge base of the subject domain [Dede, C. (1986) Int. J. Man-Machine Studies 24, 329–353]. The three latter modules are still at very early experimental stages of development and pose formidable problems. The basis of current research in artificial intelligence is the physical symbol system hypothesis [Newell, A. and Simon, H. (1976) Communications of the ACM 19, 113–126. It is argued that no knowledge base of the subject domain language can be constructed on that basis [Winograd, T. (1987) Paper presented at the Humans, Animals, and Machines: Boundaries and Protections Conference, Stanford University]. Thus, inherent limitations of the physical symbol system hypothesis make (current) artificial intelligence techniques inappropriate for the development of intelligent CALL. However, teachers can use computers in language teaching in ways which are adapted to pedagogical goals. In this sense at least, intelligent computer-assisted language learning is possible.

The physical symbol system hypothesis of Newell and Simon

The essay below deals with Newell and Simon's hypotheses about the nature of intelligent action. The material is suitable for a classroom demonstration of artificial intelligence at the high school level and above.Discussion questions following the essay are designed to encourage making explicit connections between computer science, philosophy, and the life sciences. They are part of an effort to formulate an information-oriented, algorithmic view of nature.