ATLAS Computer Laboratory at Didcot, Oxon has ordered an FR 80 precision computer output microfilm (COM) recorder from Information International of Los Angeles. The Laboratory provides a sophisticated computing and plotting service for a large number of universities and government establishments.
FR80 which costs £225,000 will be used for the high-speed recording of scientific data in graphical form.
Data is fed in to the FR80 on magnetic tape. A dedicated computer then buffers and processes this data and an electro-optical system records it as alpha-numeric and/or graphical information on a high-precision cathode-ray tube.
The use of colour on a precision CRT addressable to 16,000 points on each axis is of particular advantage when the output needs to distinguish between a number of variables. Colour highlights these parameters in, for example, a three-dimensional representation. If all these parameters vary with time then this may be recorded as an animated movie sequence. The Atlas Computer Laboratory plans to use and develop these techniques along with other applications such as grey-scale recording (the FR80 records 256 levels of grey), line drawings, technical reports etc. Plots which can take hours on conventional pen plotters are output to hardcopy in 10 seconds.
The Atlas system will have a disc file which means rapid handling of programs as well as the ability to define and store forms and logotypes. The use of the high-speed page printer features allows character recording at rates of up to 40,000 per second.
The interrogation of the Science Research Council members by the MPs on the appropriate Subcommittee of the Parliamentary Select Committee for Science and Technology must be a good thing. It is such a rare event for the public to gain a glimpse of the peculiar kind of gambling involved in science policy that the current opportunity is welcome anyway and especially timely as the university world faces insolvency and disruption on every front. The computer content of the parliamentary study is expected to be slight. However, it is potentially an occasion when a variety of knives and hatchets will be used to settle old scores. And there is the snag. Old scores are great fun to settle but they are likely to cloud the mind more than excessive alcohol or an overdose of enthusiasm.
Private gossip has often suggested that the SRC has had a fair share of both kinds of excess in the past, although all its current members are well known for their almost total sobriety and objectivity. One of the test cases for the case-makers in the SRC will be their elucidation of the function of the Chilton Atlas Laboratory. Explaining the name of the place is the easy part. The site is Chilton and the Atlas bit came from the famed machine first used to form its reputation in the number crunching scene. However, from then on things are liable to get tough.
Right from the start there was a feeling in some research circles in computing techniques that the Atlas Lab was simply another example of the high energy physicists stealing funds and machines from the computer scientists and thereby pretending to the throne of computing itself. Good things clearly came from Chilton. The molecular biologists would never have thrived so well if it had not been struggling to help them in the early sixties. But all the same the existence of Chilton meant that there was only the Manchester and London Atlas units to explore all kinds of topics. Eventually the Cambridge Titan complex became part of the thrust but funds were sorely strained and the facilities at London had to be split so that money could be made on a commercial service bureau operation. The mid-sixties were a time when computing research in UK universities could be said to be adjourned and aghast.
But then came the half-heat of technical revolution and the Computer Board. There was some expectation at the time that the Chilton complex would be turned into a Regional Centre for university computing. But London, Manchester and Edinburgh were set-up (it must be jelly, because jam don't wobble like that) and the SRC remained firmly in control of Chilton and added an ICL 1906A to the hardware and an assessment of Fortran compilers to its reputation for controversy. (At this point we can raise our glasses to Paul Bryant who ploughed all the way through that report and its lashback, and send a greeting to Mike Baylis who stuck it out till the first version had seen the controversy strike.)
From that point on the knives were out. One of the furious reactions to the Bryant-Baylis report was simply I do not care how good it is. They are simply empire building at Chilton. This feeling, that with the growth of on-site university computing centres and the provision of links to regional centres there was no longer a proper reason to keep Chilton alive, spread.
The SRC did not ignore the feeling. It was noticed that there was a special need for a concentrated body of expertise in certain types of number crunching and Chilton came to be that body. The real value of this option to universities can be debated. Certain members of university departments which do not have a strong voice in the local on-site centre have found the Court of Appeal at Chilton useful to save projects, refine research, and generally advise on the sensible use of computing techniques which are not always well understood on the local pack of hardware and software passions.
On the other side of the motion to strengthen university budgets by closing Chilton down, there are many universities now with such a wide range of support skills that absorbing another speciality would be a simple matter - it is claimed. Anyway, there are now too many organisational irons in a computing fire which is starved of that vital fuel money. There are Regional Centres and shared arrangements galore and the Computer Board and the departmental fights for the funds from the University Grants Committee and endowments.
The computational physics of liquids and solids will be discussed at the fifth symposium organized by the SRC and the Atlas Computer Laboratory at Queen's College Oxford, 14-16 April. There will be five main topics: atomic liquids; molecular liquids; phase transitions; fluids consisting of charged particles: solids. The closing date for bookings is 28 February. Applications to Marjorie Sherwen, Conference Secretary, Atlas Computer Laboratory, Chilton, Didcot, Oxfordshire OX11 0QY
A national interactive computing facility is planned by the Science Research Council (SRC) as part of the future role of the Atlas Computing Laboratory at Chilton.
This service will be aimed at engineering oriented users who will be able to control the running of their programs.
The regrouping of all major establishments within the SRC is still under discussion and the April 16 meeting of the Council of the SRC is expected to make major decisions over regrouping matters. Last year the SRC considered moving the Atlas Laboratory from Chilton to Daresbury - another SRC site. This idea has now been dropped. However, the principle of establishing an interactive facility for engineers across the UK has been accepted and full proposals are likely around the middle of the year. Until these proposals are set out the exact nature of the new interactive facility cannot be described. But it is hoped to serve both those engineers who need mathematical model interaction as well as those who need graphics interactive features.
The robotics researchers may also be provided with useful facilities if enough flexibility can be built into the new network.
Hardware decisions will have to wait for months. The new role for the Chilton centre is seen as a direct outcome of studies which show that the usefulness of batch working is very limited.
Atlas held its fifth symposium on the subject of "Computational Physics of Liquids and Solids" at Queen's College, Oxford in April. While the majority of the hundred participants were from the UK there were speakers from the USA, France, Italy, Holland and other European countries, showing the high level of interest in this field.
By the end of the symposium it was clear that high speed electronic computers have made a tremendous impact on the theory of liquids and solids. As one speaker remarked "Computer simulation methods have not so much replaced theory, but have made a theory of liquids possible".
Peter Hunter has been awarded a Joint Research Fellowship by the Atlas Computer Laboratory of the Science research Council and St Catherines's College, Oxford. It is tenable for three years from the autumn of 1975, and will aid him in his research work at the Oxford Physiology Laboratory where his study of the working of the heart requires powerful computing facilities and access to computer graphics services.
AFICIONADOS of the arts of chess and computing gathered at Balliol College, Oxford, last month for the second conference on computer chess organised by the Atlas Computer Laboratory and the Artificial Intelligence and Simulation of Behaviour study group.
A number of experts were invited to give papers on their work, including Dr Mikhail Donskoy of the USSR, one of the authors of the world champion program, KAISSA. Unfortunately Donskoy was unable to attend, a disappointment only mitigated by the presence from the US of Dr Hans Berliner, of Carnegie-Mellon. Berliner is the world correspondence chess champion and has also spent the last eight or nine years programming computers to play the game. To most people in the practical, model building side of the subject, ie making a program play the computer game, Berliner is THE expert. Although full of good, implementable ideas he has no illusions as to the limitations of such chess programs.
This was clearly demonstrated on the Sunday evening before the conference proper when Berliner was invited to play MASTER, the British No 1 program developed by AERE, Harwell, and the Science Research Council Atlas/Rutherford computing complex.
Berliner has beaten more chess programs than he has had cold buffets with a good wine. When he arrived at Atlas his dinner and MASTER were waiting for him, and the following are extracts from the log of the game, (for M read MASTER, for H read HANS, for O read operator).
M: Right, who's first for a beating?
O: OK. Hans is Black.
M: G1 F3; N-KB3.
O: Hans is eating at the moment.
M: Surely not food for thought?
H: Munch, . Munch. B8 C6;N-QB3.
M: Crunch crunch - That's not in my book. Silence for 24.3 seconds while MASTER ruminates.
M: C2 C4; P-QB4 Nodes 24000 value 25.
H: E7 E5.
M: (trying to transpose back into book and failing) Crunch Crunch Crunch. . . .
M: D2 D4; P-Q4 Nodes 57000 value 25.
(interval)
M: Is Hans munching or crunching?
H: Yes, E5 E4.
Berliner later went through the game, shown complete in Figure 1, explaining some of his early moves. The first one (N-QB3) is standard practice for good players - don't let weaker opponents (particularly machines) play book openings.
| Move | Master | Berliner | Secs | value |
|---|---|---|---|---|
| 1 | N - KB3 | N - QB3 | 0 | 0 |
| 2 | P - QB4 | P - K4 | 24 | 25 |
| 3 | P - Q4 | P - K5 | 57 | 27 |
| 4 | N - K5! | Q - B3 | 24 | 31 |
| 5 | N × N | QP × N | 22 | 28 |
| 6 | N - B3 | B - KB4 | 56 | 34 |
| 7 | B - K3 | O - O - O | 163 | 60 |
| 8 | Q - R4 | P - QR3 | 105 | 88 |
| 9 | O - O - O | N - R3 | 65 | 85 |
| 10 | P - B3! | Q - K2 | 58 | 46 |
| 11 | Q - R5 | P - KN3 | 17 | 99 |
| 12 | Q - K5? | Q × Q | 21 | 90 |
| 13 | P × Q | R × R+ | 2 | 57 |
| 14 | K × R | B - K3 | 10 | 51 |
| 15 | N × P | N - KB4 | 45 | 43 |
| 16 | B - B4 | B × P | 40 | 38 |
| 17 | P - QN3 | B - Q4 | 41 | 52 |
| 18 | N - B6 | B - K3 | 24 | 63 |
| 19 | P - K4 | N - Q5 | 47 | 40 |
| 20 | B - Q2?? | B - QR6! | 43 | 15 |
| 21 | B - QB4 | B × B | 38 | 10 |
| 22 | P × B | R - Q1 | 29 | 29 |
| 23 | K - K1 | P - KR4 | 74 | 19 |
| 24 | K - B2 | N × P | 50 | 34 |
| 25 | B - B1 | B × B | 38 | -29 |
| 26 | K × N | R - Q6+ | 1 | -35 |
| 27 | K - K2 | R - Q7+ | 7 | -56 |
| 28 | K - B3 | B - N7 | 22 | -84 |
| 29 | K - K3 | R × P | 103 | -110 |
| 30 | K - B4 | R - QB7 | 27 | -111 |
| 31 | R - Q1 | R - B6 | 39 | -109 |
| 32 | K - N5 | R × P | 26 | -100 |
| 33 | R - Q7 | B × P | 10 | -97 |
| 34 | R × P | B × N+ | 6 | -89 |
| 35 | K × B | R × P | 5 | -95 |
| 36 | R - B8+ | K - Q2 | 6 | -113 |
| 37 | K × P | R - KR5 | 105 | -121 |
| 38 | R - B7+ | K - Q3 | 6 | -71 |
| O | A couple of moves and resign ...OK? | |||
| M | Fine by me - this is a difficult one to win | |||
| 39 | R - B6+ | K - Q4 | 25 | -103 |
| 40 | R - B5+ | K - Q5 | 174 | -104 |
| 41 | R - B2! and resigns | 189 | -123 | |
Berliner tested the program twice to try tangling up its development; there were little traps, which he knows programs like CHESS 4.0 (the American program) are liable to fall into. MASTER avoided the traps and developed quite well; in fact by move 11 it had a slight advantage, threatening P-Q5, and Berliner had to play chess for a while rather than continue the interrogation.
The fact that MASTER lost the game eventually was no surprise to anyone. Berliner has a rating of 222 (British) whereas MASTER only rates about 170 in the opening and middle game. This drops to about 30 for end games which, in this case, it hardly got into. The program's authors were very pleased with its performance against a player of Berliner's calibre.
The conference itself lasted one and a half days and the papers (depending on the listener) ranged from the profound to the puerile. I have no desire to repeat my own personal views, indeed as one of the organisers I helped to provide a platform for speakers whose views I found astonishing. What did seem evident to me was that the majority of the audience fell into one of three categories. One group is the artificial intelligentsia. They fully understand the difficulties of the problem and are still thinking about how to solve it. The second group are the let's get on and program something crunchers, the model makers, the people who can make big computers float round the room whistling God save the Queen. These people absolutely refuse to put anything remotely resembling knowledge, chess or otherwise, into their programs if they can avoid it, believing that if the result plays good chess then it can be more easily adapted to attempt other, more useful, decision making problems.
The third group is the most important, these are the people who are new to the subject, the people who say, It sounds like a fascinating problem and I'd like to know what's going on.
Hans Berliner fits into two of these categories, his talk covered the AI approach and the crunchers, but he is hardly a newcomer. He was the first speaker and the domain of his talk was chess tactics with emphasis on recognising situations and dealing with them explicitly.
He is full of good ideas and techniques which are relevant to the problem of selecting the right move in a game of chess and, more important, showed clearly how each idea and technique could be implemented in a computer.
The next paper by Ron Atkin, of Essex University, was more profound. Atkin has developed a mathematically valid approach to positional play in chess and also described a method of simulating the hierarchical method used by the chess master. His ideas have an intuitive appeal; one feels he must be right but the problem of implementing the ideas in a computer are enormous and, as yet, unsolved.
Another problem was that not a single person in the audience was sufficiently competent to stand up and say, This is all very well but your approach will not handle this situation in a chess game, a tactic which often shoots down the simple minded crunch programs.
The final talk on the first day was given by Professor Donald Michie of the Machine Intelligence Research Unit, Edinburgh, who dealt with mechanisation of the King, Rook versus King end game. The first machine to play this end game was built by Torres y Quevado c. 1900. Hardly anything was published about it and Michie made an interesting" reconstruction of the machine, and I have to admit that Michie had got most of it right.
The next day the speakers included Dr Soei Tan, also of Edinburgh, who dealt with pawn structures; John Birmingham and Peter Kent partially described how MASTER operates; and Mike Clarke, of Queen Mary College, talked about the King, Pawn versus King ending.
Clarke's approach is much more scientific than the crunch programs; he has enumerated all possible positions of the K, P v K end game (there are 906,545,760 of them) and discovered which are wins and which are draws. He has done a similar enumeration for the King, Rook v King ending and should, with a little more effort. be able to scientifically work backwards to the 16 pawns, 16 pieces end game.
As can be seen a great deal of time was spent on this King, Rook v King ending. This is interesting because if this ending can cause such difficulty then it is intuitively obvious that the full game must be at least an order of magnitude more difficult.
This is intuitively obvious but difficult to prove, nevertheless it appears to be true in the real world. I certainly have great difficulty obtaining a draw with the lone King.
I was intrigued by this interest, and knowing the ratings of many of the people present, I asked six of them the following question: Figure 2 is the corner of a board which is infinite in the x and y directions. The lone black King is somewhere out there on the board, can he be mated? (Forget the 50 move rule).
Five people, with ratings of 170 or better, said no; the two Kings must be brought together to effect the mate and this is impossible. The sixth person already knew the answer. I was impressed by the speed at which humans can answer this problem (about five seconds); they see immediately the hopelessness of the case whereas MASTER with its lack of chess knowledge would waste many minutes trying to solve the problem.
To underline this weakness of MASTER a game it played two weeks previously was shown at the conference. In this test the program was allowed to Queen a pawn (it took over four milliseconds to make this decision) and then attempt the Q,K v K mate. This it failed to do. Even searching nine ply deep it could only drive the King into a corner, and then waffled because the Kings were on opposite sides of the board and the program was too myopic to play correctly.
With extreme reluctance the knowledge or incentive to bring the Kings together was fed into the program. MASTER then mated the lone K with Q,K followed by R,K and finally B,B,K within the 50 move limit. It failed to do the B,N,K versus K because it did not know into which corner to drive the King. Therefore, playing with itself it only has a 50 per cent chance of success and nil per cent against a human who will run to the wrong corner.
So some chess knowledge is necessary and for this the crunchers must depend on the artificial intelligentsia; collaboration is vital. However, too much chess knowledge results in programs that can only play chess (this is true for humans, also).
Indeed when it comes to variants of chess, e.g. Monster chess and draughts chess, then the expert player is in great difficulties against a generalised program such as MASTER.
At the end of the conference a discussion was led by Rex Malik who brought up the pertinent questions, What use is it outside chess? Even inside chess, where are we going? And, why don't people collaborate more?
The collaboration question is very important particularly when one realises that only about seven people are working (in their spare time) on the MASTER program. This is not good enough, particularly as we practically invented the game in 1951.
I admit one problem, if more support is made available, the funders will have to decide, which chess program to support. I should be grateful to any reader who can suggest a method of making this decision, perhaps it requires managerial knowledge?
Finally, Max Bramer has offered to demonstrate his K,R v K and K,P v K programs to anyone who is interested. He can be contacted at the Open University.
