The Hype associated with early AI research

‘Did the Hype Associated with Early AI Research Lead to Alternative Routes Towards Intelligent Interactive Computer Systems Being Overlooked?’

By Chris Reynolds, chris@codil.co.uk 

Introductory statements for panel  session "AI – Future Realities" chaired by John Handby at "Archives of IT Forum on the Histories of the Internet,"  London, 9 January 2024

Artificial Intelligence research has involved chasing one heavily funded and overhyped  paradigm after another, with intervening AI winters. A study of the commercially unsuccessful projects can tell you a lot about the economic and political environment that decided which projects should get funded and which would be abandoned, 

As the long-retired project leader I am looking through the archives of a project which closed because it did not correspond to the currently fashionable paradigm.

 

I am neurodiverse and, perhaps as a result, have worked in a number of projects involving complex information. Probably because I have aphantasia I have concentrated on projects which involve text and avoided projects which involved images.

Complex Systems I have Worked With

Exeter University	59-62	Modelling with quantum uncertainty
Cooper Technical Bureau	62-65	Documenting R&D research
Shell Mex & BP	65-66	Study of complex invoicing system
English Electric Leo	66-70	Future Large Commercial Integrated Management Information Systems
Linesman/Mediator	70-71	User interface to military system
Brunel University	71-88	Designing an intelligent electronic clerk
DSIRO, Sidney	90-91	Planning interactive environmental systems
[in retirement]	now	Reassessing intelligent electronic clerk research

 

In 1962 I worked as a graduate-level clerk handling technical mail in an international research organisation. In effect I was acting as a human chatbot. I felt computers might help (along the lines of Bush's 1946 article "As we may think") and decided to find out how computers could be used to provide better management information.

 

I moved to what I now know was probably the most complicated batch sales accounting systems anywhere in the UK. I suggested that the opaque and very complicated invoicing systems (about a million ever changing sales contract "rules" and 5000 products) could be reformatted into a salesman friendly transparent form when terminals became available.

 

I was head-hunted to work on future top-of-the-range commercial management information systems. The CODIL (Content Dependent Information Language) project was a generic attempt to produce a transparent interactive electronic clerk which could handle a wide range of commercial and database oriented tasks.  The research was based in a division that was axed when International Computer Limited was formed. Significant unfunded research was done up until 1988 at Brunel University involving a wide variety of tasks. 

 

While at Brunel I was also involved in the BLEND interactive publishing project and was the online book review editor of the pre world wide web HICOM bulletin board.

 

The CODIL project ended in 1988 after an experimental educational version had got very favourable reviews. It was decided by a newly appointed head of department that the approach did not agree with the AI paradigm which was currently favoured during the late 1980s. This was at a time when the currently fashionable AI network models were out of fashion

 

 The current  reassessment of the archives highlights the differences between the Current AI models and the CODIL model.

 

AI Approach	Intelligent Clerk Model
Turing	Test
Involves a human playing a game with a “black box.” Systems designed to pass the test normally end up as incomprehensible opaque black boxes	Involves a computer working symbiotically with humans on complex real world tasks. Top priority is transparency  ‘and  mutual understanding
Big Data & Large	Language Models
Uses the vast power of computers and complex algorithms to process many billion variables.	Works on a human scale with a “short term memory” which normally handles less than  a dozen variables at a time.