A Memoir of AI Research in the 1980s

---

What first drew you to AI?

During my teens I was saturated in pulp SF: Asimov, Clarke, Heinlein. There were many, many examples of AI systems which fascinated me. I always wanted to work in that area. I also wanted to be a theoretical physicist, of course, but that ambition did not survive Warwick University.

When did AI become something you could actually work on?

My move to STL in March 1982 got me into a research lab, although initially I was working on formal methods for software development. AI was becoming a thing in the early 1980s, and the lab wanted to respond. I was asked to develop a research programme.

STL was Standard Telecommunication Laboratories, based in Harlow, Essex. Established by the American telecommunications group ITT, it conducted research for ITT’s European companies.

STC, Standard Telephones and Cables, was the British equipment and cable manufacturer within that family. STL’s remit had been broader than serving STC alone. When ITT reduced its STC shareholding to a minority in 1982, the laboratory had to move towards a closer relationship with STC and justify its role afresh.

The tension between research and manufacturing had an institutional basis as well as a cultural one. (STL history; STC history.)

Why did they choose you?

STL was mostly a hardware lab: electrical and optical engineers. We software researchers were a very distinctive subgroup, immersing ourselves in universal algebra, category theory, formal logic, VDM and CCS, and using tools like Lisp and Standard ML.

I was the only person in that group who was interested in AI.

The timing mattered. After the disappointments and funding cuts of the 1970s, AI was enjoying a commercial revival. Much of the momentum came from American expert systems: programs which encoded specialist knowledge as rules. Digital Equipment Corporation’s XCON, delivered in 1980, configured computer orders and became part of its manufacturing process.

Japan’s Fifth Generation initiative helped turn AI into an international industrial priority. Corporate laboratories had reasons to take notice.

You had the brief. What did you propose?

STL was very university-like. You didn’t have to show practical applications.

My first interest was learning. I demonstrated a learning system to the STL managing director which used a lattice of possible responses to a stimulus, pruning the lattice in response to errors. A kind of Darwinian learning system. I think the learning avatar was a simulated fish.

I have no idea what he made of it.

But your main interest lay elsewhere?

I was mostly interested in theory. What does it mean for a situated agent to be said to believe, desire and intend? I was trying to understand this mathematically and soon discovered modal logic as the correct approach.

Were you trying to build an agent, or understand what it would mean for a machine to have a mind?

I was influenced by Dennett’s ideas about the intentional stance.

Interpreting these logics of “agenthood” over models which were actually functioning agents was, for me, a stepping stone towards understanding consciousness in a materialist, operational way.

I also wanted to understand how we could create predictive social-science models, like Asimov’s psychohistory. That dovetailed with my Marxist theoretical foundation.

Dennett’s intentional stance means interpreting a system through the beliefs and desires which make sense of its behaviour. Nigel’s ambition was to give such descriptions a mathematical interpretation tied to what an agent actually did by construction, operationally. The further ambition was to connect this with consciousness and, beyond the individual agent, social behaviour.

What came out of that programme?

A number of published papers and my PhD thesis, developed while I was at STL. But the programme did not survive more than a few years.

What happened?

I put in a request to buy a robot arm, which I hoped could be controlled by a planning system based on an automatic resolution theorem prover. The request was denied, and the AI research programme was closed down as the eighties drew to a close.

What was happening to AI more broadly?

AI was stumbling. The practical applications weren’t materialising.

This was the onset of the second “AI winter”, usually placed in the late 1980s and early 1990s. There had been real successes, including XCON, but extending and maintaining expert systems demanded expensive specialist work. The market for dedicated Lisp machines also faltered as ordinary computers improved. Commercial expectations outran delivery. 

Research nevertheless continued: neural networks were reviving (very slowly), and probabilistic approaches were opening new directions.

STC had its own expert-systems project. Did you have much contact with it?

They were developing a system for configuring STC electronic equipment, such as multiplexers. I went to visit once and got the tour.

Their disdain was evident: practical guys versus some academic intellectual working on abstract irrelevance. I suppose my work was pretty esoteric even to the average AI researcher.

Did their practical approach fare better?

The expert system also failed to prove useful in the messiness of the ever-changing real world. That too was terminated.

But STC was incidental to my own history.

What happened to you after the PhD?

I was awarded my PhD and never did paid AI research again. BNR came in, and I was retasked into telecommunications network planning: SDH and all that.

In 1991, Northern Telecom acquired STC, and the Harlow laboratories became part of Bell Northern Research, its research organisation. SDH, Synchronous Digital Hierarchy, was the digital transmission technology central to Nigel’s subsequent network-planning work. (History of the Harlow laboratories.)

Did you feel you had abandoned the work?

Some of my colleagues thought I had sold out. But I couldn’t see how to extend the thesis work towards issues of consciousness.

I still don’t.

---