Scruffies and Neats in Artificial Intelligence

In a previous essay [0] I traced the Lighthill Affair to the tension between the scruffies and the neats in Artificial Intelligence. As a reminder the official [1] definition of these terms:

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

For a few lines this is a pretty good characterization. But I think it only scratches the surface. In this essay I will explore the contrast in temperament and attitude that exists along several dimensions and is found elsewhere in science.

One of these dimensions is the one in which pious contrasts with naughty. The choice of these terms bears explanation. In reading a scientist like Dijkstra I get the impression that for him science is the highest, most noble calling that a human can aspire to. Hence, “pious”. The other type can be no less excellent as a scientist, but does not exude the pious aura. Consider Richard Feynman who started to wonder why, or whether, the human sense of smell is so much less sensitive than a dog’s. After some encouraging results that showed he was a better sniffer than humans are generally believed to be, he formed the hypothesis that dogs are only better at sniffing trails because they have their nose closer to the ground. So he crawled around the rug on his hands and knees, sniffing, to find out whether he could tell the difference between where he had walked and where he hadn’t [2]. Because of this episode I don’t classify Feynman among the pious scientists. (by the way,  the hypothesis was rejected, even when he had done the walking barefoot. In line with popular belief, the dog was a better sniffer than its owner.)

I might have used “prankish” rather than “naughty”, were it not for a passage in an essay by my favourite venture capitalist. The essay I have in mind [3] discusses what to look for in a start-up company in search of funding. Graham lists five points, among which number four is naughtiness.

Though the most successful founders are usually good people, they tend to have a piratical gleam in their eye. They’re not Goody Two-Shoes type good. Morally, they care about getting the big questions right, but not about observing proprieties. That’s why I’d use the word naughty rather than evil. They delight in breaking rules, but not rules that matter.

It is this that I recognize in accounts of Feynman. As another example in science, consider Irving John (“Jack”) Good. He was the General Editor of a 413-page volume with the title The Scientist Speculates: an Anthology of Partly-Baked Ideas [4]. Many of the 123 entries have intriguing titles. A sampling:

  • pbi #48 Explosive Telepathic Fields
  • pbi #85 Life in the Sun [i.e. on the surface of]
  • pbi #79 Multipurpose Plants
  • pbi #76 Steak from Sawdust
  • pbi #67 A Problem for the Hedonist
  • pbi #67 A Method for Encouraging Clairvoyance in Rats
  • pbi #40 Self-Organizing Pumps and Barges
  • pbi #39 A Splendid National Investment.

Would any self-respecting person deign to be associated with such a prankish enterprise? Look at this partial list of contributors: Bruno de Finetti, Dennis Gabor, Wassily Leontief, J.E. Littlewood, N.W. Pirie, John Maynard Smith, C.H. Waddington.

And, not surprisingly, we find contributions from Marvin Minsky and Donald Michie. The latter started off the present article indirectly via its predecessor about Michie and Longuet-Higgins, two very different temperaments, of whom it was hoped that their complementary talents would result in fruitful collaboration.

Michie was an experimentalist, but an unusual one. His lack of computer access spurred him in 1965 to invent an experimental set-up named MENACE (“Matchbox Educable Nought-And-Crosses Engine”), a demonstration of the BOXES learning algorithm in the form of a contraption that, on closer inspection, turns out to be an agglutination of 256 salvaged matchboxes requiring the operator to transfer coloured beads between boxes [5].

As I argue in [6], scruffy in AI got its imprint from the MIT hackers as described by Steven Levy [7]. They were not only AI pioneers, but also computer pioneers as well as pranksters. This last aspect showed in their allergy to passwords and locks (of the hardware kind). Back to Feynman, who was their soul brother in this respect [2].

Enough about naughty versus pious. In physics there is a lively animosity between experimentalists and theorists. Here follow some illustrations.

When Edsger Dijkstra was asked how he got into computing, he would tell that he was a student in “theoretical physics”. I suspect that the University of Leiden did not have a theoretical option in its physics program, but that some students had already differentiated in their minds.

When Christopher Longuet-Higgins descended into the Department of Machine Intelligence in Edinburgh, he had earned the freedom to do so by important results in theoretical chemistry.

One of my favourite pop-physics books is the one by Leon Lederman [8]. I recommend it, in spite of the god-awful title (no doubt a concoction of the publisher’s). I learned some physics from it. I also enjoyed his characterization of theorist versus experimentalist. Lederman keeps tracks of points scored by both sides: theorists anticipated the positron, the pion, the antiproton, and the neutrino, particles detected soon after in experiments. These predictions fed the already considerable arrogance of the theorists (“They need us to tell them what they are seeing.”). Theorists failed to anticipate the tau-lepton, the upsilon and the muon. As regards the latter, Isidor Rabi is famous for having remarked, in arch-theorist mode, “Who ordered that?” Lederman, after a dutiful homily about theorists and experimentalists using their complementary skills to uncover the secrets of the universe, is more convincing about the more mundane aspects of the differences between the two tribes:

Today we have two groups of physicists both with the common aim of understanding the universe but with a large difference in cultural outlook, skills, and work habits. Theorists tend to come in late to work, attend grueling symposiums on Greek islands or Swiss mountaintops, take real vacations, and are at home to take out the garbage more often. […] They tend to worry about insomnia. […] Experimentalists don’t come in late: they never went home. […] Sleep is when you can curl up on the accelerator floor for an hour.

You get the flavor. Yet the poor experimentalists are not drop-out theorists. Nor do they leave for easier lives in the financial sector. It looks like experimentalists can’t help being what they are.

It will suffice to remind the reader of the notorious “Pauli effect” [9], the conclusive evidence of the vast difference between theoretical and experimental physicists.

Could it be that the difference is one of temperament and that a similar difference is found in computer science and AI? I think so, and I think that it explains the vast difference between the Longuet-Higginses and someone like Minsky, who not only did a lot of important things in AI, but also invented, built, and patented a new kind of microscope [10].

But, as one of my esteemed correspondents objects, Longuet-Higgins loved, and did, experiments; elegant little experiments. That illustrates my point: for a theorist experiments have to be elegant. If a question can’t be settled this way, the theorist thinks the time is not ripe. An experimentalist can’t wait. If nobody knows how to keep it elegant and little, then the expermimentalist tries to get the money, the facilities, and the people for a big messy experiment.

Sooner or later discussions about clashing temperaments will return to the Snow-Leavis Affair, alias the Two Cultures debate. It started with the 1959 Rede lecture in Cambridge University in which scientist, novelist, and civil servant Charles Snow described and deplored the fact that there seemed to be an unbridgeable gap between the scientific and literary scholars. The literary critic F.R. Leavis reacted to Snow’s lecture with a savage attack. Although fellow literary scholars thought the tone of Leavis’s article deplorable, they tended to agree that Snow’s analysis was shallow.

After half a century it is difficult to appreciate the furore arising from Snow’s lecture. Yet it is worth revisiting in the form of [11], especially for Stefan Collini’s preface, from which I quote:

The literary critic, habitually attending to the fine texture of verbal detail, can at times barely be persuaded that something is being said at all if it is being said badly. It is almost a truism of the critic’s working practice that the conventional distinction between form and content is misleading in literature: a work is those words in that order — one cannot blithely assume some “meaning” behind them which failed to get itself expressed properly but which is nonetheless the “message” of the text.

This fastidiousness has persisted into the twenty-first century. Many of the humanities lectures I have recently attended consist of the lecturer reading the paper, not “talking” to a rapid succession of Powerpoint slides. The verbatim record of the talk (if one would ever come to light) would be a jumble of words depending on a great deal of goodwill for any intelligibility. The text that was read is a joy to read after the occasion. In such fastidiousness I recognise the theorist, the neat: if an experiment cannot be done neatly, then it should not be done at all; we are not ready for it.

Dijkstra’s THE operating system [12], was an elegant little experiment. Of course it took half a dozen people a few years of hard work to write the assembler code for the Electrologica X8. But compared to contemporary operating systems (other than Unix) THE was both elegant and little. In physics there is a constant interaction between the experimentalists and theorists in the sense that a physics experiment only counts as such when theorists agree that it makes sense. In software the scruffies not only dominate, but there do not seem to be any neats around even to criticize, let alone restrain, the ubiquitous big, messy experiments. The reason is of course that, unlike in physics, some of the big, messy experiments generate big, neat piles of money.


[0] A Programmer’s Place February 18, 2011.
[1] Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig. Prentice-Hall, First edition 1995, page 21.
[2] “Surely, You’re Joking, Mr Feynman!” by Richard P. Feynman, as told to Ralph Leighton. W.W. Norton, 1985.
[3] What We Look for in Founders
[4] “The Scientist Speculates: an Anthology of Partly-Baked Ideas” Irving John Good, ed. Heinemann, 1962.
[5] “BOXES: an experiment in adaptive control” D. Michie and R.A. Chambers. In: Machine Intelligence vol. 2. Ella Dale and Donald Michie, eds. Oliver and Boyd 1968.
[6] The MIT style in Artificial Intelligence 1958 – 1985
[7] Hackers by Steven Levy. Doubleday, 1984.
[8] The God Particle by Leon Lederman with Dick Teresi. Dell, 1993.
[9] The Pauli Effect
[10] Memoir on Inventing the Confocal Scanning Microscope by Marvin Minsky.
[11] The Two Cultures by C.P. Snow, with introduction by Stefan Collini. Cambridge University Press, 1993.
[12] THE multiprogramming system

One Response to “Scruffies and Neats in Artificial Intelligence”

  1. Kragen Javier Sitaker Says:

    Perhaps Charles Moore, the FORTH guy, qualifies as a modern Neat? Or, at a less accomplished level, Darius Bacon, or Aristotle Pagaltzis, or me? Perhaps Rob “I have never written a program that used cursor addressing” Pike? Or Rob Harley? Or Dan J. “10 years of qmail 1.0” Bernstein? Or the people working on CompCert? Maybe the designers of the Elements of Computing Systems course?

Leave a Reply

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

You are commenting using your 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: