Tuesday 21 June 2011

Contacting LEO, the Lyons King of the early computers

I came across the Leo Computer Society web site today – and decided to contact them to see if any of their members remembers the CODIL project or are interested in what became of it. I wrote:

In 1967 I decided I was not being paid enough as a Chemistry PhD working as an information scientist – and switched to working for Shell Mex and BP to get onto the computer bandwagon. They had, at the time, two large Leo III installations and I was in the first group of programmers to learn CLEO – and became something of an expert using it. Shortly afterwards they went to tender for replacement systems with online terminal input, etc. and I was asked to do some preliminary work on transferring the sales contract applications. I came up with what I considered a very obvious solution – which was deemed to be “research” and hence impractical. At that point it was announced that the contract would go to Univac – and I was offered a job with what was English Electric Leo – by which time they were already selling System 4 computers.

I joined the Sales department at about the time it moved to Euston Road, working with George Stern on what was, in effect market research for future large systems. This meant a rapid learning period ranging from discussions with CPU designers as to likely technical advances through to discussions with the future application “ideas” men in selected large customers.

After a few months I hesitatingly suggested to George that there was a fundamental problem in the design of the stored program computer when used for tackling logically complex tasks which required significant human interaction. With his encouragement I submitted a short report to John Aris – who passed it to David Caminer and John Pinkerton  Overnight I was assigned to Minerva Road to draw up detailed proposals as to how the idea might be tested, The result was that a “top secret” project was set up (funded £50,000) and work was started on the patents. The project was initially called DORIS but later changed to CODIL. I was supervised by John Meredith-Smith and for various reasons I continued working at the Euston Road offices, although reporting to Minerva Road.

Basically the idea was to build an information processor which, at the machine code level handled sets instead of numbers. Processing was simply by comparing sets using a simple algorithm and by using a simple set compare routine, associative addressing, and a very high degree of recursion the need for all the explicit conditional tests and address arithmetic vanished. The black box of the stored program computer paradigm was replaced by a white box information processing system, which processed sets and NOT conceptually separate program and data, and which ordinary people could use directly because it used sets in the way they found natural.

The “Pilot Project” to simulated the operation and demonstrate it on a wide variety of admittedly small test applications went better than predicted. Unfortunately the merger to form ICL blew the idea right out of the water – despite (or because of) Basil de Ferranti initially approving the idea. I was given permission to transfer to a university – but made an unsuitable choice – in part because I am naturally a quite retiring back-room boy type of individual who needs an actively supportive environment. Despite looking very promising the research was later abandoned for personal reason – and because all funders seem to think that because conventional computers were so successful an approach which asked serious questions about the stored program computer paradigm must be invalid!

Recently I decided to return to the subject and have started a blog www.trapped-by-the-box.co.uk which includes the history of the project and the key publications describing it. I have also been looking at recent research to see (a) whether anyone else has done anything else like it – the answer appears to be “No” and (b) whether there could be any further research interest. The answer to this one could be a very big “Yes.” Preliminary work suggests suggests that CODIL is a natural to implement on a neural net – and could be an admittedly crude model of a “symbolic assembly language” which occupies the memory between the brain's neurons and natural language (which is why it is easy for people to understand) and which can also process a wide range of information processing applications (as clearly demonstrated in it original experimental form). As far as I can gather no-one has made any progress on tackling this problem.

So what I have been doing could be of interest to your members for one or more of the following reasons

  1. I worked on a major LEO III installation for about 18 months and have memories of the machine and the work I did on it.
  2. It shows that after the switch to System 4 computers David Caminer and John Pinkerton were still very actively encouraging imaginative ideas into computer architecture.
  3. If my wildly speculative ideas turn out to be correct it could be that the chances of a major research and technological breakthrough were missed because of the formation of ICL and if this is so it is only fair that the origins of the ideas could be credited back to LEO.

I have not been in touch with anyone from the old days for many years and would love to hear from anyone who remembers me, or DORIS or CODIL

No comments:

Post a Comment