In April, 1962, a meeting was held at 38, Belgrave Square under the chairmanship of Sir Keith Murray to consider how far this country might take a lead in the development of computer science on broad, new and inter-disciplinary lines. The initiative in this connection had been taken by Mr. Perkins and Mr. Holland-Martin of International Computers and Tabulators Ltd. who had drawn up proposals for a school of Computer Science or Computational Philosophy. Mr. Perkins had earlier had informal discussions with the Minister for Science (Lord Hailsham) and with Sir Alfred Pugsley of Bristol University whose report on the possibility of developing a university centre on the lines proposed had been somewhat discouraging.
2. Mr. Perkins was present at the meeting together with a Mr. McRea, Sir Harry Melville, Sir Willis Jackson and Mr. Copleston were also present. It was agreed that the next step would be to arrange further consultations on an informal basis with representatives of various disciplines with a view to working out a clearer formulation of the concepts involved and of the academic content of the subject to be studied. Since D.S.I.R. might eventually. be involved financially, Sir Harry Melville. agreed that his Department should organise those consultations.
3. A meeting was duly held in July, 1962, under the chairmanship of Sir Harry Melville at which it was decided to appoint a Working Party under the chairmanship of Professor Howarth. The Working Party came to the conclusion that D.S.I.R. should be invited to consider establishing a fairly substantial research group under Dr. M.V. Wilkes at Cambridge and that D.S.I.R. should continue to examine this field with the aim of encouraging more research and giving it support. A copy of the Working Party's report is attached.
4. It is understood that discussions are at present proceeding concerning the establishment of the research group at Cambridge; and a new Computing Science Sub-Committee of the Research Grants Committee, under the chairmanship of Lord Halsbury, has been set up to give continuing encouragement and support in this subject.
On 31st July, 1962, a meeting on Computer Science which had been called by Sir Harry Melville to review the need to develop new scientific uses of computers decided to appoint a Working Party with the following terms of reference:-
The original membership was :-
Dr. J. Howlett, N.I.R.N.S., and a representative of the National Physical Laboratory, (Variously Dr. J. V. Dunworth, Dr. E. T. Goodwin and Dr. A. M. Uttley) were added to the membership. Mr. C. Jolliffe, Mr. P. D. Greenall, Mr. F. Rock Carling and Mrs. M. E. E. Neighbour of D.S.I.R., H.Q. attended all meetings of the Working Party.
2. The Working Party agreed at their first meeting that the field of advanced computer science was one of great scientific interest and also one which was likely to be of considerable value in leading to applications of computers in scientific research and in industry. It therefore seemed to the Working Party that effort spent in identifying and encouraging research into new, or recently developing, ways of using computers, such as for instance, their application to non-numerical analysis, (problem-solving and theorem proving) would be well repaid. The Working Party decided that its field should be the study of these ways of using computers rather than the application of computers in already known ways to larger problems or to the more rapid methods of solution of problems of existing type.
3. The Working Party therefore decided to begin by gathering information on the chief work in progress in the United Kingdom on computers, together with some indications of the most significant work in progress in the United states. They therefore invited Mr. Strachey to act as their consultant for this purpose and to undertake a review. This he did, writing to some 63 university workers asking them to give him details of researches on which they were engaged. In summarising the replies to this letter, Mr. Strachey told the Working Party that, in his view, there were only one or two substantial research projects in the Working Party's field of interest, though there were a few small projects. There was more interest in making computers work usual problems quickly in preference to finding unusual ways of using computers. A small amount of work on non-numerical uses of computers was being done in the biological sciences, but very little in other disciplines. There appeared, however, to be a growing interest in this field and in the field of computer theory, and there was scope for this work to be developed now that larger computers were becoming available in the United Kingdom. The Working Party also received valuable information from other sources, especially through Professor MacKay.
4. The Working Party therefore decided to see whether work in this field could be stimulated. To achieve this successfully, it was agreed that it could best be undertaken in a university setting where it would receive help and encouragement and stimulation from contact with a variety of scientific disciplines. Furthermore the need to have access to a very large computer for a substantial part of its working time, with freedom to make modifications to the computer when necessary was self-evident. The Working Party also considered whether support would best be given to one substantial research group or spread over a number of smaller research teams. It was felt that this could only be answered in the knowledge of the resources which might be fully available for the research, and also upon the character of the existing teams on which developments might be based. The Working Party therefore decided in the first instance to examine a number of proposals for expanding researches in this field which had arisen in the course of their enquiries.
5. These proposals were as follows:-
6. After considering these proposals the Working Party came to the conclusion, that at any rate as a first step they preferred to recommend that support should be given for the establishment of one new computer research group, but that this should not preclude assistance on a smaller scale to other researches, or the establishment of a second research group at a later stage. After considering the various proposals the Working Party decided that a research group at Cambridge University with a programme based jointly on the proposals of Dr. Wilkes and Mr. Strachey offered the greatest promise, taking into consideration the intrinsic merits of the proposals, their relevance to the Working Party's field of interest, and the conditions in which the group could be established at Cambridge University.
7. The Working Party therefore invited Dr. Wilkes and Mr. Strachey to prepare a fuller statement of the lines on which such a research group might begin work and an outline of its organization. Notes by Dr. Wilkes and Mr. Strachey on the work of the group are attached to this report.
8. This proposal was, the Working Party felt, the most substantial and promising project so far discovered, and the one which covered the most aspects of the Working Party's fields of interest. Although it did not meet in every respect the hopes with which the Working Party had set out, examination of the field had led to the conviction that one single centre tackling all aspects of advanced computing science was neither practical nor desirable. This project held out the best hopes for a major impact on the subject and should enable the U.K. to make a substantial contribution. It offered prospects of a rate of growth which was substantial but realistic.
9. Mr. Perkins offered to discuss With I.C.T. and other industrial firms, to what extent they would be prepared to help to support a group in computing science.
10. It was proposed that the group should consist of 2 senior research staff, with 13 scientific, programming and other assistants. Provision would also be made for visiting senior scientists from abroad. When built up to its full strength, the cost of this group would be of the order of £44,000 a year. Dr. Wilkes felt that to ensure the degree of stability necessary for work in this field the group would require to be assured of support for a period of 8 years. This the Working Party accepted, but felt that in the first instance D.S.I.R. might consider offering a grant up to the end of the University quinquennium, namely to July 1967, with the expectation that a renewed grant would be made to support the group up to 1972 if it were making satisfactory progress. The Working Party felt, however, that the University of Cambridge should be invited to assume some financial responsibility for the group by 1967.
11. The Working Party therefore recommend that the D.S.I.R. be invited to give further consideration to a grant of £116,000 be made to Cambridge University for the establishment of a research group in computing science, under Dr. M.V.Wilkes, the group to include senior staff of high calibre. The grant is intended to support the group up to July 1967. If, by that date, the group is well-established and making good progress, it should be supported, at least up to 1972, partly by D.S.I.R. and partly by the University of Cambridge.
12. As a result of their studies, the Working Party were convinced that research in this field of advanced computing science was timely and promising, and merited special consideration and support. They felt, therefore, that there was a need for a. continuing body to advise D.S.I.R. on applications for grants received in this field, and also on ways in which new research could be stimulated. They did not, however, feel that they were the right body to fulfill this function but they welcomed the decision of the D.S.I.R. to establish a Computing Science Sub-Committee of its Research Grants Committee, with terms of reference which would enable the new Sub-Committee to carry out both these functions. The Working Party therefore recommend that they should now be dissolved.
I regard the object of such a unit as being to make a co-ordinated attack on certain problems in an area which has not yet received a generally accepted name, but which is sometimes referred to as non-numerical analysis, and sometimes as symbol manipulation.
This programme is especially timely, and particularly so at Cambridge where we shall have, at long last, a fast computer with a sufficiently large high-speed store. Computers are now in general use for doing arithmetic and one of the tasks of the next decade is to bring them equally into general use for non-numerical work such as formal mathematics. I personally wish to devote as much of my energies as I can make available to this programme. In addition, as Mr. Strachey pointed out in his memorandum dated 26 June 1962, there are now clear signs that a mathematical theory of information processing is beginning to develop. Mr. Strachey went on to say This theory seems to have no direct Connection with numerical analysis but grows more naturally from formal algebra and symbolic logic. It is based on a desire to understand the fundamental ideas and processes involved in a mechanical application of an elaborate set of rules. This is essentially a mathematical pursuit and involves continual abstraction, simplification and generalization. This would represent the pure side of the unit's activities, and I would attach much importance to such studies being carried out in close contact with more applied work, since in Computer Science, as in other branches of science, the interplay between theory and practice is at the very basis of progress.
I am convinced that the scale of the problems in the area outlined is such that real success cannot be achieved by isolated workers or by groups of twos or threes, but needs the co-ordinated effort of a substantial group working on inter-related projects over a period of years. As a first step in the creation of the unit, I would establish a nucleus of people working on programming languages and time-shared memory schemes whereby a user can be brought into immediate and direct contact with a powerful computer via a keyboard. This work would grow naturally out of work being done for the Titan project, but would no longer be primarily aligned towards providing a computing service. Throughout its life the unit would remain active in this area, and one would hope that it would produce results that were important in their own right and would, at the same time, provide the advanced programming technology for other projects, both pure and applied.
I have already mentioned the applied project which would be my own particular concern namely the use of a computer to aid a mathematician in performing algebraic and analytical manipulation. I have already had some experience with a programme for formal differentiation of algebraic and trigonometrical expressions, but this is only a beginning. There are challenging problems in partial differentiation, elimination, integration and so on. If he were to join the unit Mr. Strachey would, I know, wish to give particular attention to the theory of information processing and programming languages, as well as supervising other projects.
In writing this paper I have concentrated on the initial phases or the unit, so that readers might be able to visualize how it would take shape in the first instance. A unit building up to the scale envisaged in my previous memorandum dated 16 March 1963 would, however, be able to embrace a number or major projects, together with a variety of small ones. Subjects with which it should certainly concern itself as staff with appropriate interests and experience became available are heuristic programming, game playing, and theorem proving. One cannot look very far ahead, but I see the unit as providing an environment in which continually advancing programming techniques would act as a stimulus to, and benefit from, theoretical research in computer science, both in the new fields that are now seen to be emerging and in the newer ones that the future will undoubtedly bring forth.
There are two main ways in which a computer can be used by universities - as a calculator and as a stimulus to research in its own right. The first of these is by now generally acknowledged to be of great importance and many research workers in such fields as crystallography, astrophysics, nuclear physics, engineering and statistics are entirely dependent on the use of a computer to get their results. Many universities have set up a computer department to provide these calculating facilities for workers in other departments, and it is from this side of the work on computers that the demand for increased speed and numerical power arises.
This work is essentially applied; that is to say, that the interest in the work lies wholly in getting results; and if any research is done by the computer department it will generally be directed at developing better techniques for solving particular problems. Recently, however, it has become clear that there is considerable scope for pure research into computers and their uses - that is to say, research which is directed towards understanding the fundamental ideas and concepts which underlie the mechanical application of algorithmic processes and developing the mathematical techniques suitable for manipulating them. This Sort of research is pure in the sense that its aim is merely to understand and classify and not to produce any specific results. In the long run, of course, a basic understanding of this sort will inevitably lead to practical results, probably of great importance, but it is too early yet to forecast what form they will take.
There are now clear signs that a mathematical theory of computing (or information processing as it is now generally called) is beginning to develop. This theory seems to have no direct connection with numerical analysis, but grows more naturally from formal algebra and symbolic logic. It is based on a desire to understand the fundamental ideas and processes involved in the mechanical application of an elaborate set of rules; This is, essentially a mathematical pursuit and involves continual abstraction, simplification and generalization.
One point of basic interest which distinguishes it from existing symbolic logic, is an interest in the relationship between a concept and its representation; other points which seem to me important are the need for a deep investigation into the ideas of functions and variables and a considerable development of Church's calculus together with the use of recursive functions.
There is already a certain amount of work published which is in, or close to this area. The ideas underlying Algol 60 [1] - at least the part of it concerned with procedures and the block structure - are certainly relevant. McCarthy's LISP [2] is probably the most important single piece of work but a recent report by an American group studying business data processing languages [3] and papers by Gorn [4], van Wijngaarden [5] and Dijkstra [6] at the International Symposium on Symbolic Languages in Data Processing (Rome, April 1962) all show signs of the same general direction of development.
The ideas in this line of research seem to have come from people trained in computers who have become dissatisfied with the ad hoc way in which they are used and want to generalise. Classical mathematicians have been slow to appreciate that computers can stimulate any interesting new ideas - possibly because of the importance of the representation of ideas as opposed to the ideas themselves, which gives the subject a wholly spurious practical or applied appearance. In fact, the subject is only applied in that it requires the use of a computer in order to discover the more remote consequences of any theoretical idea.
There are a great many complicated systems of interacting objects which we live with but understand only very imperfectly. Even when we understand the rules which govern the behaviour of the component parts we can deduce very little about the behaviour of the system as a whole. Computers make it possible to construct a model of a system of this sort and to carry out experiments on the way the model behaves. This makes it possible first to test the validity of the model, and then to investigate the large-scale rules which govern the system as a whole.
Some work on these lines has already been done - partly on simplified models of very complicated physical systems, partly on systems which are more concerned with organizational problems. Neutron diffusion calculations in a reactor with a realistic geometry give a good example of the first type, road traffic simulation and scheduling problems are examples of the second.
There are two main areas of interest in this work: developing techniques for setting up models, and discovering the general rules which govern the behaviour of complicated systems and relate this behaviour to the rules governing the component parts.
The most interesting general work on these lines is probably Newell, Shaw and Simon's work on a general problem solver [7] and it is interesting to note that some of the most powerful programming techniques now in use (list-processing, IPL, LISP) were introduced by workers in this field who needed them to be able to manipulate the models they were working with.
The computer has given us a new tool, not merely for doing complicated computations but, much more significantly, it enables us to carry out extended sequences of logical operations - that is to say, operations whose principal characteristic is that they are always completely determined. But the essentially new feature of computers is not the nature of the operations they perform but the fact that we can now consider prescribing rules which require several million steps for their execution. This means that it is no longer possible for a human being to foresee all the consequences of the rules he lays down any more than it is possible for a human being immediately to perceive all the characteristics of a solution of a differential equation. Both the lines of research I have described above have as their ultimate aim the better understanding and use of these complicated rules which involve very large numbers of steps in order to produce correspondingly complicated results.
The aim of the first line of attack is to establish the foundations of the theory on a sound basis; the aim of the second is to elaborate a constructive theory which will be based on these foundations. An analogy with another branch of mathematics may help to make the relationship between the two clearer. Classical analysis has proved an extremely powerful mathematical tool; I believe the use of extended algorithms will prove to be an even more powerful one. The investigations into the foundations of information processing theory can be compared with the investigations into questions of convergence, limiting processes and the like which underlie the whole of analysis. The investigation into the construction and use of models could then be compared with the introduction of the special functions of analysis (such as Bessel functions) and the development of the techniques which are appropriate to establish their properties.
It is a rather depressing fact that, so far as I know, no work on these lines is being done in any university in the country. There are various reasons for this, of which the most important are probably the unsuitable nature of all English computers for work of this sort and the narrow view of the function of computers adopted by the directors of most computing departments. Whatever the reason, the fact remains that all the work in this field has been originated elsewhere. Most of it has been done in America but there are significant contributions from Europe.
It is characteristic of the work I have been considering that, at any rate initially, the computer is used in an apparently extravagant way. This is because the focus of attention is no longer on getting the greatest possible amount of computation from the cheapest possible computer but has shifted to the problem of finding the simplest possible way to describe complicated operations. This inevitably means trying to make the computer do as much as possible of the donkey work of programming and in the early stages, at any rate, involves an apparently cavalier disregard of questions of space and speed. The problem of speed is not so vitally important, as the amount of computation carried out in these exercises is not usually large: but the question of space is completely inescapable. It is, in fact, quite impossible to do useful experiments along these lines on a computer which has an immediate access store of less than about 16,000 words. With the exception of Cambridge, which has in the last six months installed a 16,000 word store, no university in this country has a machine with an immediate access store of more than 1,000 words.
It is unreasonable to expect any significant amount of work to be done in this field by people who have no chance whatsoever of testing their ideas on a computer, and this fact, I think, is sufficient to explain the absence of any work in this field in English universities.
I believe that this work on the fundamental theories of information processing is of the utmost importance and that we cannot afford to neglect it. I am convinced, moreover, that it is, in every sense, a thoroughly proper subject for academic study and, indeed, that it is not likely to be pursued anywhere in this country except in a university or research institute.
The first requirement, therefore, is that one or more groups of academic workers should have access to a computer with a suitably large store. In the first instance it should be possible to hire time on a suitable machine in London and to start with the requirements would not be very great. A team of three to six research workers (which would seem a reasonable size) would probably need about three hours a week on an IBM 709 or EMIDEC 2400 or less on a faster machine. At the preferential rates available for universities this would probably cost £5,000-£10,000 a year. It would be desirable to make some arrangements for a block booking of time, as any arrangement which made the research workers continuously conscious of the cost of every minute spent on the machine would have a disastrously inhibiting effect on their ability to investigate new ideas.
There are various problems involved in setting up a research team for this sort of project which I do not feel qualified to discuss properly. Questions of finance, accommodation and supporting services in universities are outside my experience. It would clearly be desirable for any research team of this sort to be closely connected with other computer activities in the university and if possible with the mathematicians, but how far this would be possible and under whose jurisdiction the team might come I do not know.
26th June, 1962.
(1) Naur, P. (ed.) Report on the algorithmic language ALGOL 60 Comm. A.C.M. 3, 5 (May 1960) 299-314.
(2) McCarthy, J. Recursive Functions of Symbolic Expressions and their Computation by Machine. Comm. A.C.M. 3, 4 (April 1960) 184-195.
(3) Language structure Group, CODASYL Development Committee. An information Algebra. Comm. A.C.M. 5, 4 (April 1962) 190-204.
(4) Gorn, S. An axiomatic Approach to Prefix Languages. Symp. on Symbolic Lang. Rome 1962 (To be published).
(5) van Wijngaarden, A. Generalized ALGOL. Symp. on Symbolic Lang. Rome 1962 (To be published).
(6) Dijkstra, E.W. An attempt to Unify the Constituent Concepts of Serial Program Execution. Symp. on Symbolic Lang. Rome 1962 (To be published).
(7) Newell, A., J. C. Shaw and H. Simon. Report on a general problem-solving program. Proc. Internat. Conf. on Information Proc. UNESCO (1959) 256-264.
