From the Chronicles of Scruffy Versus Neat: the Lighthill Affair

The 1973 Lighthill Affair was an Affair in the sense of the Dreyfus and Profumo Affairs. And although it was scaled down to teacup size, it was big enough to make it into a textbook published twenty years later:

… the Lighthill Report, which formed the basis for the decision by the British government to end support of AI research in all but two universities. (Oral tradition paints a somewhat different and more colorful picture, with political ambitions and personal animosities that cannot be put into print.) [1]

In this article I will put in print some of the things hinted at here, and elaborate on the issues that have remained topical.

I first heard the word “Lighthill” in November 1972, soon after I started my research fellowship with the Department of Machine Intelligence at the University of Edinburgh. As Jean Michie said to me: “What, you haven’t heard of Lighthill? That man who wants to do away with us?” I hadn’t, but it wasn’t before long I learned that Jean was right about his wanting to do away with the Department of Machine Intelligence and, what was more, about his having the clout to bring this about.

The “Lighthill” referred in the conversation with Jean came to mean the publication of a report, followed by a televised debate at the Royal Institution in London. The proceedings gave the impression of a movie Crown Prosecutor presenting of his case. Sir James Lighthill, FRS, Lucasian Professor at Cambridge University pronounces from behind a lectern on a platform elevated above the other participants of the debate, seated in the pit below: Donald Michie, John McCarthy, and Richard Gregory. The debate was actually only the second half of the proceedings; the first half consisted of an oration by Lighthill standing on his platform. Science policy or theater?

At the time the Department of Machine Intelligence, Michie’s creation, had become a world centre of AI. To give you an idea, I’ll list (as far as memory serves) of visitors to the department I met over the years I was there [2]. A year later the Department of Machine Intelligence at the University of Edinburgh was reduced to three people: Donald Michie, professor of Machine Intelligence, one secretary, and one technician. The remainder joined the newly created Department of Artificial Intelligence or left for jobs elsewhere. All this on the basis of the report of a single person, ignorant of the research area. It is extraordinary that a body dispensing public funds, such as the UK Science Research Council, could have proceeded in this manner.

If a report were necessary, it should have been written by an American expert. Only in the US was a suitable depth of expertise available. Moreover, the search for a British expert would have been complicated by the difficulty of finding someone not connected via the old boy network. Not only was the reporting “expert” British, but also the old boy connection was present with a vengeance: Lighthill was one of, what I shall call, “The Winchester Four”, described in an interview with Freeman Dyson:

IOPScience: You excelled in mathematics at school. Were those happy days?
Freeman Dyson: Yes, on the whole. We were very lucky because everything was screwed up by the war [World War II] — I remember in my last year at Winchester having only seven hours of classes a week. It was wonderful — we were free to get our own education. The teaching was fairly good, but didn’t make much difference — we learned much more from each other than we did from the teachers. There were four of us, who were about the same age, who became fellows of the Royal Society — the Longuet-Higgins brothers, Sir James Lighthill and I.

The Winchester Four stayed close in their interests: all became mathematical physicists. All four became Fellows of the Royal Society. If there would have been something like Nobel Nominees, then Freeman Dyson and Christopher Longuet-Higgins would have been among those; Dyson in quantum electrodynamics and Longuet-Higgins major (public-school style, to avoid confusing him with the younger brother) in quantum chemistry.

Longuet-Higgins major joined Donald Michie and Richard Gregory in founding the Department of Machine Intelligence in Edinburgh. Soon after this, Gregory left Edinburgh. By the early seventies an acrimonious dispute had developed between Longuet-Higgins (from now on all references to L-H will be to Longuet-Higgins major) and Michie. The latter, who had taken the initiative in the department’s founding, believed that it was time for Longuet-Higgins to assume the admittedly onerous duties as the head of the department. Longuet-Higgins denied that this could be expected of him.

The dispute with Michie was not the only reason for Longuet-Higgins to be dissatisfied with the Department of Machine Intelligence. Artificial Intelligence (from now on “AI”) suffered under the tension between two schools of thought, briefly denoted as the Neats and the Scruffies, described as follows by Russell and Norvig [1]:

“… the neats — those who think that AI theories should be grounded in mathematical rigour — versus the scruffies — those who would rather try out lots of ideas, write some programs, and then assess what seems to be working.”

His mastery of mathematics had given Longuet-Higgins a deep understanding of quantum mechanics, which, in turn made it possible for him to explain certain chemical phenomena that had been missed by earlier followers of Linus Pauling’s program of explaining the nature of the chemical bond in terms of quantum mechanics [8]. In AI or out, Longuet-Higgins was the epitome of Neat. By the early seventies Scruffy had become the dominant mode in AI. To the distress of Longuet-Higgins, Machine Intelligence in Edinburgh was not all Neat.

Such was the situation when it transpired that the UK Science Research Council had selected for its grand review of AI not an AI researcher, nor even a computer scientist, but, astonishingly, one of the Winchester buddies of Longuet-Higgins. One can imagine that we (that is, those whose jobs were about to be terminated) were not unbiased readers of the, to us, notorious Lighthill Report.

Many years have passed. Most of those forced out found other jobs, mostly far away from Edinburgh. I trust that for the others these exciting events have become a distant memory, as they have for me. A memory consisting of Sir James Lighthill as a pompous idiot who lent himself to produce a flaky report to serve as a blatantly inadequate cover for a hatchet job.

Last Christmas I was reading the autobiography of Laurent Schwartz (1915–2002), the great French mathematician. Schwartz was an amazing phenomenon. As a mathematician he was unusually versatile, spanning the entire range from probability and applied mathematics to Bourbaki (he even was a Bourbaki). For his efforts in opposing the Algerian and Vietnam wars I regard him as a French version of Pauling, or Chomsky.

Schwartz is mainly remembered for his discovery of distributions, a theory that mathematically modelled certain hitherto pseudo-mathematical objects that were found indispensable in physics; the Dirac delta function, for example. Schwartz made the discovery in 1944, in recently liberated Paris. It was a few years before the new theory gained a following. As usual it was among the best young mathematicians that it was first appreciated. Schwartz mentions one James Lighthill.

This, and the passage of time, caused me to review my 1973 impression of Sir James as a “pompous idiot”. I found the book by Lighthill [4] on distributions, purchased in 1958 by Victoria College (now University of Victoria). It is a gem: brief, elegant, accessible; a great way not only to learn the advanced concept of distribution, but also a good introduction to basics like Fourier series and integrals. A biography I dug up outlined an impressive career including practical applications in aerodynamics and directing the Royal Aircraft Establishment.

And I got the [Lighthill Report]. Did I find valuable insights that I missed in 1973? Alas not. Yet, with the wisdom of hindsight, and knowing how Neat versus Scruffy has played out over the decades, it was worth re-reading the report. What I now read is a tract by a Neat on the attack, setting out to root out Scruff.

Allow me to elaborate on Neat and Scruffy as described by Russell and Norvig. Yes, the Neats believed that AI theories needed to be grounded mathematically. But I think it is more enlightening to say that to be Neat was to be stuck in cybernetics. “Cybernetics” is the term coined by Norbert Wiener and used as the title of his 1948 book introducing the concept. What inspired the new word was the recently discovered parallel between feedback control in electrical circuits and animal nervous systems. That Wiener mentions digital computers at all in a 1948 book shows that he was unusually well informed and perceptive (at the time “computer” meant, if not a person, an analog electronic computer).

In 1948, Wiener’s book was a conceptual breakthrough. The problem was that by 1970 the next conceptual breakthrough — the computer as new universe — had already happened, but the majority of mathematicians and scientists were still living in a world bounded by cybernetics. Admittedly, by then computers (now unambiguously digital) were vastly more useful than they were in 1948, but they still played a minor role in cybernetic research. For the Neats, the significance of improvements in computer technology was that electrical engineering systems could be automated in a more sophisticated way. This is laid out in the Advanced Automation section (“area A”) of the Lighthill Report. As befits a cybernetician, Lighthill applauds the use of computers for studying natural nervous systems (area C in the Lighthill Report). The report places some AI activities in areas A or C. The Neats had been imaginative enough to include post-Wiener additions to the mathematical arsenal, like resolution theorem-proving as method for problem-solving and planning. Not surprisingly the report was forced to include a third category, which it named area B. Here went everything that didn’t fit the cybernetics framework. Like robotics.

The Neats continued to regard the computer as an instrument useful only for the advancement of Wiener’s vision of cybernetics. I think the Scruffies deserve more credit than their characterization by Russell and Norvig. What distinguishes them is their sense of the computer being more than an instrument subservient to existing disciplines. To them the computer represents a phenomenon in its own right, one that promises to shed new light on endlessly debated conundrums like the nature of intelligence, thought, and consciousness. As Archimedes is reputed to have said: “Give me a lever long enough and a fulcrum on which to place it, and I shall move the world,” so the implied motto of the Scruffy was “Give me a PDP-10 with Lisp and I will blast wormholes through the space of the traditional sciences.”

McCarthy disdained decades of psychology research in learning. “I am interested in the phenomenon where you teach, not by multiple training actions, but by telling, once, and the subject then knowing it forever — something that only happens in humans.” Michie exulted in bypassing control theory, much approved of by Lighthill as a gloriously successful and sophisticated branch of applied mathematics, to produce controllers for unstable systems via the admittedly stupid BOXES algorithm.

As a whole, AI at MIT was Scruffy and particularly so when it came to representing and using knowledge in computers. In this respect Stanford and Edinburgh were Neat: they embraced mathematical logic as a formal framework. They welcomed resolution because it was within mathematical logic as well as easy to implement on a computer. MIT viewed mathematical logic as a distraction. They believed that the resolution system could not accommodate the necessary pragmatics. Accordingly they advocated “procedural embedding of knowledge” by bypassing mathematical logic and creating a new language [6].

The disagreement led to a confrontation at the 1970 Machine Intelligence workshop in Edinburgh. The workshop programme announced “The Irrelevance of Resolution”, a talk by Seymour Papert, one of the leaders of the AI group at MIT. Instead, one Gerry Sussman showed up with the message that he was sent by Papert to give the talk in his place. It looked like Sussman had no experience giving talks in faraway countries. He seemed nervous about going into the bastion of resolution theorem-proving and telling the assembled experts how totally misguided their work was. As a result he started by overdoing the confrontational aspect. A few minutes into the talk, Longuet-Higgins walked out of the room, pensively puffing his pipe.

In 1972 Minsky and Papert published what I view as the manifesto of Scruffy [AIM-252]. The summary on page 2 intones:

Thinking is based on the use of
SYMBOLIC DESCRIPTIONS
and description-manipulating processes
to represent a variety of kinds of KNOWLEDGE
— about facts, about processes,
about problem-solving, and about computation itself,
in ways that are subject to
HETERARCHICAL CONTROL STRUCTURES
— systems in which control
of the problem-solving programs
is affected by heuristics
that depend on the meaning of events.

The words are verbatim from AIM-252; I have been so free as to add ventilation [7], as called for by the solemnity of the occasion. Since that time we have learned that this is compatible with Neat; at the time it seemed the essence of anti-Neat.

The manifesto brings the computer beyond the role of enabler of cybernetics, which is where Neat had become stuck. It is a bold step putting the computer at the centre of a new world (let’s call it AI) where, for instance, it becomes possible to think (productively) about thinking. To establish AI as a subject in its own right it was initially necessary to repudiate heritage, even where it could be useful. In the early 1970s the Scruffies had to be unnecessarily confrontational, just as a teenager needs to be unnecessarily obnoxious. But the author of the Lighthill report did not have the strength of a wise parent. He was enraged by what he saw and constructed a report calculated to kill as much of Scruffy as possible.

Acknowledgments

Thanks to Alan Robinson for helpful criticisms, to Michael Levy for finding the Lighthill Report, and to André Vellino for finding the televised debate. The text benefited greatly by editorial input from Eva van Emden (http://editing.vanemden.com/)

References

[1] Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig. Prentice-Hall, First edition 1995, page 21.
[2] Here follows a list of AI researchers of whom I remember right now that I met them at the Department of Machine Intelligence between 1969 and 1975: Harry Barrow, Woodrow Bledsoe, Daniel Bobrow, Robert Boyer, Peter Buneman, Alan Bundy, Alain Colmerauer, John Darlington, Edward Elcock, Cordell Green, Patrick Hayes, Carl Hewitt, Gérard Huet, Robert Kowalski, Christopher Longuet-Higgins, David Luckham,  John McCarthy, Zohar Manna, Donald Michie, Robin Milner, Ugo Montanari, J Moore, Nils Nilsson, Seymour Papert, Ira Pohl, Alan Robinson, Laurent Siklossy, Aron Sloman, Gerald Sussman, Austin Tate, Richard Waldinger, and David Warren.
[3] I asked a friend whose life is in quantum chemistry, unsullied by AI, whether he had heard of H.C. Longuet-Higgins. His reply: “In my opinion Longuet-Higgins has made a very important contribution (1963) to theoretical chemistry by explaining that the symmetry of a molecule is not described by a point group, but by a permutation-inversion group. The latter group is in general unnecessarily large, and L-H has shown that only the subgroup of ‘feasible operations’ (a concept invented by L-H) is significant. This is a very subtle piece of theory that remains a mystery to most chemists. … The point of L-H is that you should see the nuclei of atoms as quantum mechanical particles, and most chemists cannot do that, they are so used to the structure of molecules that they cannot understand that nuclei can tunnel through a potential barrier (whereas most chemists can understand that for electrons). Chemists say: you have trans-butadiene and cis-butadiene and those molecules are different. L-H says, no they are two wave functions with maxima in the trans and the cis configuration and the wave functions overlap: trans-butadiene is a ‘little bit’ cis and vice versa.”
[4] An Introduction to Fourier Analysis and Generalised Functions by M.J. Lighthill. Cambridge University Press, 1958.
[6] See Wikipedia article.
[7] See Ventilated Prose.
[8] The Nature of the Chemical Bond by Linus Pauling, 1939.

2 Responses to “From the Chronicles of Scruffy Versus Neat: the Lighthill Affair”

  1. Andre Vellino Says:

    Great post Maarten!

    Have you seen this BBC live debate between Lighthill and Michie and McCarthy in 1973. The video is poor quality and the soundtrack is out of synch, but it’s worth a watch.

    http://www.aiai.ed.ac.uk/events/lighthill1973/1973-BBC-Lighthill-Controversy.mov

  2. Kragen Javier Sitaker Says:

    There’s a contemporary essay sort of advocating the Neat position in programming in general, although he calls it “Fortran school” (as opposed to “Lisp school”) at http://www.evanmiller.org/mathematical-hacker.html. Perhaps it may interest you.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s


%d bloggers like this: