bundy@edxa@sri-unix.UUCP (10/01/83)
From: BUNDY HPS (on ERCC DEC-10) <bundy@edxa> THE DEPARTMENT OF ARTIFICIAL INTELLIGENCE AT EDINBURGH UNIVERSITY Artificial Intelligence was recognised as a separate discipline by Edinburgh University in 1966. The Department in its present form was created in 1974. During its existence it has steadily built up a programme of undergraduate and post-graduate teaching and engaged in a vigorous research programme. As the only Department of Artificial Intelligence in any university, and as an organisation which has made a major contribution to the development of the subject, it is poised to play a unique role in the advance of Information Technology which is seen to be a national necessity. The Department collaborates closely with other departments within the University in two distinct groupings. Departments concerned with Cognitive Science, namely A.I., Linguistics, Philosophy and Psychology all participate in the School of Epistemics, which dates from the early 70's. A new development is an active involvement with Computer Science and Electrical Engineering. The 3 departments form the basis of the School of Information Technology. A joint MSc in Information Technology began in 1983. A.I. are involved in collaborative activities with other institutions which are significant in that they involve the transfer of people, ideas and software. In particular this involves MIT (robotics), Stanford (natural language), Carnegie-Mellon (the PERQ machine) and Grenoble (robotics). Relationships with industry are progressing. As well as a number of development contracts, A.I. have recently had a teaching post funded by the software house Systems Designers Ltd. There, however, is a natural limit to the extent to which a University Department can provide a service to industry: consequently a proposal to create an Artificial Intelligence Applications Institute has been put forward and is at an advanced stage of planning. This will operate as a revenue earning laboratory, performing a technology transfer function on the model of organisations like the Stanford Research Institute or Bolt Beranek and Newman. Research in A.I. A.I. is a new subject so that there is a very close relationship between teaching at all levels, and research. Artificial Intelligence is about making machines behave in ways which exhibit some of the characteristics of intelligence, and about how to integrate such capabilities into larger coherent systems. The vehicle for such studies has been the digital computer, chosen for its flexibility. A.I. Languages and Systems. The development of high level programming languages has been crucial to all aspects of computing because of the consequent easing of the task of communicating with these machines. Artificial Intelligence has given birth to a distinctive series of languages which satisfy different design constraints to those developed by Computer Scientists whose primary concern has been to develop languages in which to write reliable and efficient programming systems to perform standard computing tasks. Languages developed in the Artificial Intelligence field have been intended to allow people readily to try out ideas about how a particular cognitive process can be mechanised. Consequently they have provided symbolic computation as well as numeric, and have allowed program code and data to be equally manipulable. They are also highly interactive, and often integrated with a sophisticated text editor, so that the iteration time for trying out a new idea can be rapid. Edinburgh has made a substantial contribution to A.I. programming languages (with significant cross fertilisation to the Computer Science world) and will continue to do so. POP-2 was designed and developed in the A.I. Department by Popplestone and Burstall. The development of Prolog has been more complex. Kowalski first formulated the crucial idea of predicate logic as a programming language during his period in the A.I. Department. Prolog itself was designed and first implemented in Marseille, as a result of Kowalski's interaction with a research group there. This was followed by a re-implementation at Edinburgh, which demonstrated its potential as a practical tool. To date the A.I. Department have supplied implementations of A.I. languages to over 200 laboratories around the world, and are involved in an active programme of Prolog systems development. The current development in languages is being undertaken by a group supported by the SERC, led by Robert Rae, and supervised by Dr Howe. The concern of the group is to provide language support for A.I. research nationwide, and to develop A.I. software for a single user machine, the ICL PERQ. The major goal of this project is to provide the superior symbolic programming capability of Prolog, in a user environment of the quality to be found in modern personal computers with improved interactive capabilities. Mathematical Reasoning. If Artificial Intelligence is about mechanising reasoning, it has a close relationship with logic which is about formalising mathematical reasoning, and with the work of those philosophers who are concerned with formalising every-day reasoning. The development of Mathematical Logic during the 20th century has provided a part of the theoretical basis for A.I. Logic provides a rigorous specification of what may in principle be deduced - it says little about what may usefully be deduced. And while it may superficially appear straightforward to render ordinary language into logic, on closer examination it can be seen to be anything but easy. Nevertheless, logic has played a central role in the development of A.I. in Edinburgh and elsewhere. An early attempt to provide some control over the direction of deduction was the resolution principle, which introduced a sort of matching procedure called unification between parts of the axioms and parts of a theorem to be proved. While this principle was inadequate as a means of guiding a machine in the proof of significant theorems, it survives in Prolog whose equivalent of procedure call is a restricted form of resolution. A.I. practioners still regard the automation of mathematical reasoning to be a crucial area in A.I., but have moved from earlier attempts to find uniform procedures for an efficient search of the space of possible deductions to the creation of systems which embody expert knowledge about specific domains. For example if such a system is trying to solve a (non linear) equation, it may adopt a strategy of using the axioms of algebra to bring two instances of the unknown closer together with the "intention" of getting them to coalesce. Work in mathematical reasoning is under the direction of Dr Bundy. Robotics. The Department has always had a lively interest in robotics, in particular in the use of robots for assembly. This includes the use of vision and force sensing, and the design of languages for programming assembly robots. Because of the potential usefulness of fast moving robots, the Department has undertaken a study of their dynamics behaviour, design and control. The work of the robot group is directed by Mr Popplestone. A robot command language RAPT is under development: this is intended to make it easy for non-computer experts to program an assembly robot. The idea is that the assembly task should be programmed in terms of the job that is to be done and how the objects are to be fitted together, rather than in terms of how the manipulator should be moved. This SERC funded work is steered by a Robot Language Working Party which consists of industrialists and academics; the recently formed Tripartite Study Group on Robot Languages extends the interest to France and Germany. An intelligent robot needs to have an internal representation of its world which is sufficiently accurate to allow it to predict the results of planned actions. This means that, among other things, it needs a good representation of the shapes of bodies. While conventional shape modelling techniques permit a hypothetical world to be represented in a computer they are not ideal for robot applications, and the aim at Edinburgh is to combine techniques of shape modelling with techniques used in A.I. so that the advantages of both may be used. This will include the ability to deal effectively with uncertainty. Recently, in collaboration with GEC, the robotics group have begun to consider how the techniques of spatial inference which have been developed can be extended into the area of mechanical design, based on the observation that the essence of any design is the relationship between part features, rather than the specific quantitative details. A proposal is being persued for a demonstrator project to produce a small scale, but highly integrated "Design and Make" system on these lines. Work on robot dynamics, also funded by the SERC, has resulted in the development of highly efficient algorithms for simulating standard serial robots, and in a novel representation of spatial quantities, which greatly simplifies the mathematics. Vision and Remote Sensing. The interpretation of data derived from sensors depends on expectations about the structure of the world which may be of a general nature, for example that continuous surfaces occupy much of the scene, or specific. In manufacture the prior expectations will be highly specific: one will know what objects are likely to be present and how they are likely to be related to each other. One vision project in the A.I. Department is taking advantage of this in integrating vision with the RAPT development in robotics - the prior expectations are expressed by defining body geometry in RAPT, and by defining the expected inter-body relationships in the same medium. A robot operating in a natural environment will have much less specific expectations, and the A.I. Department collaborate with the Heriot Watt University to study the sonar based control of a submersible. This involves building a world representation by integrating stable echo patterns, which are interpreted as objects. Natural Language. A group working in the Department of A.I. and related departments in the School of Epistemics is studying the development of computational models of language production, the process whereby communicative intent is transformed into speech. The most difficult problems to be faced when pursuing this goal cover the fundamental issues of computation: structure and process. In the domain of linguistic modelling, these are the questions of representation of linguistic and real world knowledge, and the understanding of the planning process which underlies speaking. Many sorts of knowledge are employed in speaking - linguistic knowledge of how words sound, of how to order the parts of a sentence to communicate who did what to whom, of the meaning of words and phrases, and common sense knowledge of the world. Representing all of these is prerequisite to using them in a model of language production. On the other hand, planning provides the basis for approaching the issue of organizing and controlling the production process, for the mind seems to produce utterances as the synthetic, simultaneous resolution of numerous partially conflicting goals - communicative goals, social goals, purely linguistic goals - all variously determined and related. The potential for dramatic change in the study of human language which is made possible by this injection of dynamic concerns into what has heretofore been an essentially static enterprise is vast, and the A.I. Department sees its work as attempting to realise some of that potential. The study of natural language processing in the department is under the direction of Dr Thompson. Planning Systems. General purpose planning systems for automatically producing plans of action for execution by robots have been a long standing theme of A.I. research. The A.I. Department at Edinburgh had a very active programme of planning research in the mid 1970s and was one of the leading international centres in this area. The Edinburgh planners were applied to the generation of project plans for large industrial activities (such as electricity turbine overhaul procedures). These planners have continued to provide an important source of ideas for later research and development in the field. A prototype planner in use at NASA's Jet Propulsion Laboratory which can schedule the activities of a Voyager-type planetary probe is based on Edinburgh work. New work on planning has recently begun in the Department and is mainly concerned with the interrelationships between planning, plan execution and monitoring. The commercial exploitation of the techniques is also being discussed. The Department's planning work is under the direction of Dr Tate. Knowledge Based and Expert Systems. Much of the A.I. Department's work uses techniques often referred to as Intelligent Knowledge Based Systems (IKBS) - this includes robotics, natural language, planning and other activities. However, researchers in the Department of A.I. are also directly concerned with the creation of Expert Systems in Ecological Modelling, User Aids for Operating Systems, Sonar Data Interpretation, etc. Computers in Education. The Department has pioneered in this country an approach to the use of computers in schools in which children can engage in an active and creative interaction with the computer without needing to acquire abstract concepts and manipulative skills for which they are not yet ready. The vehicle for this work has been the LOGO language, which has a simple syntax making few demands on the typing skills of children. While LOGO is in fact equivalent to a substantial subset of LISP, a child can get moving with a very small subset of the language, and one which makes the actions of the computer immediately concrete in the form of the movements of a "turtle" which can either be steered around a VDU or in the form of a small mobile robot. This approach has a significant value in Special Education. For example in one study an autistic boy found he was able to communicate with a "turtle", which apparently acted as a metaphor for communicating with people, resulting in his being able to use language spontaneously for the first time. In another study involving mildly mentally and physically handicapped youngsters a touch screen device invoked procedures for manipulating pictorial materials designed to teach word attack skills to non-readers. More recent projects include a diagnostic spelling program for dyslexic children, and a suite of programs which deaf children can use to manipulate text to improve their ability to use language expressively. Much of the Department's Computers in Education work is under the direction Dr Howe. Teaching in the Department of A.I. The Department is involved in an active teaching programme at undergraduate and postgraduate level. At undergraduate level, there are A.I. first, second and third year courses. There is a joint honours degree with the Department of Linguistics. A large number of students are registered with the Department for postgraduate degrees. An MSc/PhD in Cognitive Science is provided in collaboration with the departments of Linguistics, Philosophy and Psychology under the aegis of the School of Epistemics. The Department contributes two modules on this: Symbolic Computation and Computational Linguistics. This course has been accepted as a SERC supported conversion course. In October 1983 a new MSc programme in IT started. This is a joint activity with the Departments of Computer Science and Electrical Engineering. It has a large IKBS content which is supported by SERC. Computing Facilities in the Department of A.I. Computing requirements of researchers are being met largely through the SERC DEC-10 situated at the Edinburgh Regional Computing Centre or residually through use of UGC facilities. Undergraduate computing for A.I. courses is supported by the EMAS facilities at ERCC. Postgraduate computing on courses is mainly provided through a VAX 11/750 Berkeley 4.1BSD UNIX system within the Department. Several groups in the Department use the ICL PERQ single user machine. A growth in the use of this and other single user machines is envisaged over the next few years. The provision of shared resources to these systems in a way which allows for this growth in an orderly fashion is a problem the Department wishes to solve. It is anticipated that several further multi-user computers will soon be installed - one at each site of the Department - to act as the hub of future computing provision for the research pursued in Artificial Intelligence.