AI Work Research Articles

Modelling students' construction of energy models in physics

One of the main differences between novice and expert problem solving in physics is that novices mostly construct problem representations from objects and events in the experimental situation, whereas experts construct representations closer to theoretical terms and entities. A main difficulty in physics is in interrelating these two levels, i.e. in modelling. Relatively little research has been done on this problem, most work in AI, psychology and physics education having concentrated on how students use representations in problem solving, rather than on the complex process of how they construct them. We present a study that aims to explore how students construct models for energy storage, transformation and transfers in simple experimental situations involving electricity and mechanics. The study involved detailed analysis of problem solving dialogues produced by pairs of students, and AI modelling of these processes. We present successively more refined models that are capable of generating ideal solutions, solutions for individual students for a single task, then models for individuals across different tasks. The students' construction of energy models can be modelled in terms of the simplest process of modelling — establishing term to term relations between elements of the object/event ‘world’ and the theory/model world, with underlying linear causal reasoning. Nevertheless, our model is unable to take into account more sophisticated modelling processes in students. In conclusion we therefore describe future work on the development of a new model that could take such processes into account.

Formal specification and decision support

To gain widespread acceptance, decision support systems must be built to the highest possible standards. We believe techniques of formal specification and refinement have a valuable role to play in the development of certain components of decision support systems. We present a tutorial study of the use of formal specification focused on a system for maintaining deductive extensions of a knowledge base. The system is specified using an object-oriented variant of the specification language Z. The relationship of the formal specification with existing theoretical work in AI is discussed together with its refinement into a demonstrably correct implementation.

Scalable Parallel and Distributed Expert Database Systems with Predictive Load Balancing

Much prior work in AI on various attempts to speed up rule-based systems by parallel processing has been reported. Unfortunately, many of these results indicate that there is limited parallelism to be found when rules are applied to relatively small amounts of data. Thus, one can predict that much greater parallelism can be extracted when rules are applied to large amounts of data. However, traditional compile-time parallelization strategies as developed for main-memory based systems do not scale to large databases. We propose a scalable strategy for the efficient parallel implementation of rule-based systems operating upon large databases. We concentrate on load balancing techniques in a synchronous model of rule execution, where the variance in runtime of the distributed sites is minimized per cycle of rule processing, thus increasing utilization and speedup. We demonstrate that static load balancing techniques are insufficient, and thus low overhead dynamic load balancing is the key to successful scaling. We present a form of dynamic load balancing that is based upon predicting future system loads, rather than conventional demand-driven approaches that monitor current system state. We analyze a number of possible predictive dynamic load balancing protocols by isoefficiency analysis to guide the design of a parallel database rule processing system.

A hybrid algorithm for determining protein structure

At Thinking Machines, my colleagues and I have developed a hybrid system combining a neural network, a statistical module, and a memory-based reasoner, each of which makes its own prediction. A combiner then blends these results to produce the final predictions. This hybrid system improves its ability to determine how amino acid sequences fold into 3D protein structures. It predicts secondary structures with 66.4% accuracy. Both the neural network and the combiner are multilayer perceptrons trained with the standard backpropagation algorithm; this article focuses on the other two components, and on how we trained the hybrid system and used it for prediction. I also discuss how future work in AI and other sciences might meet the challenge of the protein folding problem.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Beyond the Fifth Generation: parallel AI research in Japan

Many AI researchers, particularly in the USA, were surprised when Japanese AI scientist Hiroaki Kitano was chosen to receive the 1993 'Computers and Thought' award for his work in AI and parallelism. Their astonishment derived from the pervasive perception in the USA that the Japanese Fifth Generation Project, seemingly the major Japanese effort in AI and parallelism, was a failure. "How," they asked, "was a Japanese scientist able to become a leader in this area?" This confusion about the state of parallel AI research in Japan is primarily due to lack of knowledge about Japanese efforts.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">&gt;</ETX>

Beyond Rationalism: Symbols, Patterns and Behavior

Abstract Rationalism has been referred to as the tradition of explaining cognition in terms of logical structures. Much of the work in traditional AI can be seen within a rationalistic framework. Because of the problems with traditional AI, connectionist models have been proposed as an alternative. Connectionist models do solve a number of problems of AI in interesting ways, e.g. learning, generalization, and fault and noise tolerance. However, they do not automatically provide solutions to the basic conceptual problems which can be traced back to a neglect of the relation of AI systems with the real world. We will argue that if we are to make progress in the understanding of (intelligent) behavior the real issue is not whether connectionism is a better paradigm for cognitive science than traditional AI but whether a rationalistic perspective is appropriate and if not what the alternatives are. It is suggested that studying physically instantiated autonomous agents is an important step. However, we will show that building autonomous agents alone does not solve the problem either. What is needed is an appropriate embedding in a non-rationalistic framework. We will discuss a potential solution using an approach we have been developing in our group, called ‘distributed adaptive control’.

Argument and belief: Where we stand in the Keynesian tradition

There is the idea that rational belief for a single individual can be constructed via a process of unilateral argument. To preempt antipathy between the AI communities that can claim the idea that rational belief can be so constructed, we trace the idea to the beginning of this century, to Keynes' dispute with Russell over logic and probability. We review how Keynesian ideas were revived in AI's work on non-monotonic reasoning and parallel developments in philosophical logic.

Artificial intelligence and molecular biology

Molecular biology is emerging as an important domain for artificial intelligence research. The advantages of biology for design and testing of AI systems include large amounts of available online data, significant (but incomplete) background knowledge, a wide variety of problems commensurate with AI technologies, clear standards of success, cooperative domain experts, non-military basic research support and percieved potential for practical (and profitable) applications. These considerations have motivated a growing group of researchers to pursue both basic and applied AI work in the domain. More than seventy-five researchers working on these problems gathered at Stanford for a AAAI sponsored symposium on the topic. This article provides a description of much of the work presented at the meeting, and fills in the basic biology background necessary to place it in context.

Pragmatism and purism in artificial intelligence and legal reasoning

The paper identifies and assesses the implications of two approaches to the field of artificial intelligence and legal reasoning. The first — pragmatism — concentrates on the development of working systems to the exclusion of theoretical problems. The second — purism — focuses on the nature of the law and of intelligence with no regard for the delivery of commercially viable systems. Past work in AI and law is classified in terms of this division. By reference to The Latent Damage System, an operational system, the paper articulates and responds to conceivable purist (jurisprudential and AI) objections to such a program. The methods of the pragmatist are also called into question and refined. The author concludes that pragmatism within a purist framework is the only sound approach to developing reliable AI systems in law.

Exploring the No-Function-In-Structure principle

Abstract Although much of past work in AI has focused on compiled knowledge systems, recent research shows renewed interest and advanced efforts both in model-based reasoning and in the integration of this deep knowledge with compiled problem solving structures. Device-based reasoning can only be as good as the model used; if the needed knowledge, correct detail, or proper theoretical background is not accessible, performance deteriorates. Much of the work on model-based reasoning references the ‘no-function-in-structure’ principle, which was introduced by de Kleer and Brown. Although they were well motivated in establishing the guideline, this paper explores the applicability and workability of the concept as a universal principle for model representation. This paper first describes the principle, its intent and the concerns it addresses. It then questions the feasibility and the practicality of the principle as a universal guideline for model representation.

Perspectives on neural networks: [formula omitted]. KRUG International, Technology Services Division, 406 Breesport, San Antonio, TX 78216 USA

There are several perspectives or frameworks from which neural networks can be viewed. Each perspective gives a different view of where the field is going and what can be accomplished: (a) the historical evolution of AI; (b) generalization of hardware/software/knowledge engineering; (c) the modeling approach; (d) non-linear field theories: (e) relaxation and annealing; and (f) biological analogy (the existence proof). The SISD, SIMD, MIMD perspective considers the evolution of AI work: initial efforts were to use theorems to encode knowledge and theorem proving to generate results. The knowledge engineering phase accepted that the knowledge base must be extensive but still used theorem proving to apply it. The third phase, connectionism, accepts both the need for a large knowledge base and a computationally extensive approach to using it. Thus neural networks can be seen in this light as a normal progression in the utilization of larger memories and faster or cheaper silicon.

Defeasible Reasoning

What philosophers call defeasible reasoning is roughly the same as nonmonotonic reasoning in AI. Some brief remarks are made about the nature of reasoning and the relationship between work in epistemology, AI, and cognitive psychology. This is followed by a general description of human rational architecture. This description has the consequence that defeasible reasoning has a more complicated structure than has generally been recognized in AI. We define a proposition to be warranted if it would be believed by an ideal reasoner. A general theory of warrant, based on defeasible reasons, is developed. This theory is then used as a guide in the construction of a theory of defeasible reasoning, and a computer program implementing that theory. The theory constructed deals with only a subset of defeasible reasoning, but it is an important subset.

Turing's Test and the ideology of artificial intelligence

A close examination is offered of the Turing Test, whose lasting influence is ascribed to its power as a myth which is expressive of the ideology that underlies strong AI claims. Arguments are advanced to show that the Test is not a bonafide thought-experiment, nor an acceptable substitute for the question, Can a Machine Think? AI work is held to be highly valuable, but by reason of its achievement of thought-like results through the power of algorithms, not through its supposed emulation of thought. Finally, evidence is offered for thinking that the issue at hand is not only a practical, but an urgent one.

Artificial intelligence research and applications at the NASA Johnson Space Center

This is the second part of a two-part article describing AI work at the NASA Johnson Space Center (JSC). Research and applications work in AI is being conducted by several groups at JSC. These are primarily independent groups that interact with each other on an informal basis. In the Research and Engineering Directorate, these groups include (1) the Artificial Intelligence and Information Sciences Office, (2) the Simulation and Avionics Integration Division, (3) the Avionics Systems Division, and (4) the Tracking and Communications Division. In the Space Operations Directorate, these groups include (1) the Mission Planning and Analysis Division (MPAD) - Technology Development and Applications Branch, (2) the Spacecraft Software Division, and (3) the Systems Division - Systems Support Section. This second part of the article describes the AI work in the Space Operations Directorate. The first part of the article, published in the last week of AI Magazine, (7:1, Summer 1986) described the AI work in the Research and Engineering Directorate.

Object-oriented programming: Themes and variations

Many of the ideas behind object-oriented programming have roots going back to SIMULA. The first substantial interactive, display-based implementation was the SMALLTALK language. The object-oriented style has often been advocated for simulation programs, systems programming, graphics, and AI programming. The history of ideas has some additional threads including work on message passing as in ACTORS, and multiple inheritance as in FLAVORS. It is also related to a line of work in AI on the theory of frames and their implementation in knowledge representation languages such as KRL, KEE, FRL, and UNITS.

A formal model of diagnostic inference. I. Problem formulation and decomposition

This paper, which is Part I of a two-part series, introduces a new model of diagnostic problem solving based on a generalization of the set-covering problem. The model formalizes the concepts of 1. (1) whether or not a set of one or more disorders is sufficient to explain a set of occurring manifestations, 2. (2) what a solution is for a diagnostic problem, and 3. (3) how to generate all of the alternative explanations in a problem's solution. In addition, conditions for decomposing a diagnostic problem into independent subproblems are stated and proven. This model is of interest because it captures several intuitively plausible features of human diagnostic inference, it directly addresses the issue of multiple simultaneous causative disorders, it can serve as a theoretical basis for expert systems for diagnostic problem solving, and it provides a conceptual framework within which to view some recent AI work on diagnostic problem solving in general. In Part II, the concepts developed in this paper will be used to present algorithms for diagnostic problem solving.

Knowledge Representation in Sanskrit and Artificial Intelligence

In the past twenty years, much time, effort, and money has been expended on designing an unambiguous representation of natural language to make them accessible to computer processing, These efforts have centered around creating schemata designed to parallel logical relations with relations expressed by the syntax and semantics of natural languages, which are clearly cumbersome and ambiguous in their function as vehicles for the transmission of logical data. Understandably, there is a widespread belief that natural languages are unsuitable for the transmission of many ideas that artificial languages can render with great precision and mathematical rigor. But this dichotomy, which has served as a premise underlying much work in the areas of linguistics and artificial intelligence, is a false one. There is at least one language, Sanskrit, which for the duration of almost 1000 years was a living spoken language with a considerable literature of its own. Besides works of literary value, there was a long philosophical and grammatical tradition that has continued to exist with undiminished vigor until the present century. Among the accomplishments of the grammarians can be reckoned a method for paraphrasing Sanskrit in a manner that is identical not only in essence but in form with current work in Artificial Intelligence. This article demonstrates that a natural language can serve as an artificial language also, and that much work in AI has been reinventing a wheel millenia old. First, a typical Knowledge Representation Scheme (using Semantic Nets) will be laid out, followed by an outline of the method used by the ancient Indian grammarians to analyze sentences unambiguously. Finally, the clear parallelism between the two will be demonstrated, and the theoretical implications of this equivalence will be given.

Reviews of "LISPcraft by Robert Wilensky", W. W. Norton &amp; Company Inc, 1984.

Traditionally, LISP has been the language for AI researchers, It has a number of unique features for the kind of programming that AI demands. However, till recently its use in the industry was restricted because the language is not standardized and many variations of the language exists. The MACLISP and INTERLISP versions were extensively used in earlier AI work. With the increasing popularity of the UNIX operating system, FranzLISP is currently being used in a number of projects.

Field Trials in Fixed Time Sheep A.I. in Ireland

Studies reported from this laboratory in the mid seventies showed that it was possible to achieve a conception rate of 73% to a single fixed-time insemination in the autumn breeding season in ewes previously treated with intravaginal progestagen and pregnant mare serum gonadotrophin (PMSG). The present work was aimed principally at an examination of the AI technique in commercial flocks when applied over a period of three years (1979–81) and when applied in sheep of different breeds (Galways and Suffolk-cross).In the course of three successive breeding seasons, a total of 2,704 cyclic ewes in 87 farm flocks were bred to Texel or Suffolk rams by A.I. using a sperm dose of 400 million in 0.2 ml volume of a skim milk diluent; only semen samples showing a wave motion rating of 3.5 or higher were used and accepted samples were pooled. A further 525 ewes in 20 farm flocks were bred by natural service (ratio of 1 ram: 10 ewes) in one area covered by the AI work. Control of oestrus and ovulation was achieved by a 12-day treatment using commercially available progestagen sponges (Chronogest, Intervet. Ltd.; Veramix, Upjohn Ltd.); at sponge withdrawal a single intramuscular injection of 500 i.u. PMSG (Intervet. Ltd.) was administered.

Artificial Intelligence, Psychology, and the Philosophy of Discovery

In their contributions to this volume, Bruce Buchanan (1983) and Lindley Darden (1983) have provided compelling reasons why philosophers of science concerned with the nature of scientific discovery should be aware of current work in artificial intelligence. Buchanan presents abundant examples of the mechanization of generation of empirical hypotheses in current AI programs, while Darden, at a more speculative level, discusses the use of analogy in theoretical discovery. The implication of both these studies is that the philosophy of discovery can profitably borrow notions from work in artificial intelligence, since useful analogies can be found between problems in that field and the problems facing philosophers studying scientific discovery.I have no quarrel with this claim, except that it is too weak. I shall argue that work in AI is not only useful but essential for investigations in philosophy of science concerning discovery. The connecting link between the two fields is empirical cognitive psychology.