23 September 2026 · History · about 20 minutes

The Professor, the Robot and the Report: How Britain Cancelled Artificial Intelligence in 1973

A fluid dynamicist, a Bletchley Park veteran, a robot that took sixteen hours to build a toy car and a toy boat, and a televised argument at the Royal Institution. The first AI winter was a British invention, and it was decided by one man.

On 9 May 1973, in the lecture theatre of the Royal Institution on Albemarle Street, where Faraday had demonstrated electromagnetic induction and where the Christmas Lectures still happen, a Cambridge professor of applied mathematics stood up in front of television cameras to explain why Britain should stop paying for artificial intelligence.¹ He had never worked on it. That was rather the point. The Science Research Council had gone looking for a distinguished outsider to give what the archive that now holds the report calls an unbiased view of a field the Council no longer trusted itself to judge. It had found Sir James Lighthill, holder of the Lucasian chair that Newton had held and that Stephen Hawking would hold next. He was the greatest fluid dynamicist of his generation, a man who had spent the last two years of the Second World War on the aerodynamics of aircraft and the years after it on the theory of why jet engines are so loud.² Against him, in a debate the BBC filmed and broadcast under the title Controversy, sat three men who had worked on it a great deal: Donald Michie, who had broken German teleprinter ciphers at Bletchley Park alongside Alan Turing and now ran the machine intelligence department inside Edinburgh's School of Artificial Intelligence; John McCarthy of Stanford, who had coined the phrase artificial intelligence in 1955 and had flown over to defend it; and Richard Gregory, the Bristol neuropsychologist who had once been Michie's colleague at Edinburgh and had left it.¹

Michie brought a film. It showed a machine called Freddy: a gantry, a camera, a pair of adaptive grippers, wired to two computers whose combined memory would not today hold a single photograph. Freddy could be shown a heap of parts, a toy car's worth and a toy boat's worth mixed together and tipped out on a table in no particular order, and it could pick through them, work out which was which, and put both toys together. It took about sixteen hours.³ Michie showed the film as evidence. Lighthill, and everyone in the audience, could see it the other way. The question the debate turned on, and the question the report had already answered, was whether sixteen hours to assemble two toys was the beginning of something or the end of it.

The outsider's brief

The Science Research Council's problem in 1972 was money, and specifically a computer. Edinburgh wanted to buy an American machine, a DEC PDP-10, the same model on which most of the American work was being done, and it was expensive.⁴ And the Council knew, because by then everybody knew, that the Edinburgh department was at war with itself. Its own historian says the "high level of discord had become known" to the Council, and that "its reaction was to invite Sir James Lighthill to review the field".¹⁰ This was the decade in which British science funding was learning to say no. The Council's way of saying it was to commission a survey, and its choice of surveyor was itself the verdict: not a computer scientist, not a psychologist, not anyone from the three or four departments in the country that did the work, but a mathematician of unimpeachable standing from a different field entirely, whose findings could not be waved away as the losing side in a departmental feud.

Lighthill went about it as a mathematician. He spent two months, he said, reading the literature and discussing it by letter and in person with people who worked on it, and did not, Michie later complained, talk to the Americans who had founded the subject: McCarthy, Marvin Minsky, Nils Nilsson, Bertram Raphael, Alan Robinson.⁵ He produced a short paper, Artificial Intelligence: A General Survey, which the Council published in 1973 bound together with four replies, under the title Artificial Intelligence: a paper symposium.⁴ The pairing was meant to look fair. It was fair in the way a trial is fair when the sentence has already been passed.

The survey's method was to divide the field in three and to name the parts with letters, a device that gave the whole thing the air of a school timetable. Category A was Advanced Automation, "the clear objective of this category of work being to replace human beings by machines for specific purposes": the machine reading of printed characters, guided missiles, the automation of design and assembly.⁶ Category C was "Computer-based CNS research", the use of computers to model brains, "theoretical investigations related to neurobiology and to psychology".⁶ Both of these Lighthill was prepared to fund, because both were, in his view, real subjects with real customers: engineers on one side, biologists on the other. Between them he placed Category B, and B stood for Bridge, and also for Building Robots, and here the report found its target. The whole concept of Building Robots, he wrote, "is, indeed, seen as an essential Bridge Activity justified primarily by what it can feed into the work of categories A and C".⁶ His argument was that the bridge had been promised for twenty years and had not been built, and that the traffic across it was flowing in neither direction.

The explosion

The intellectual centre of the report is a single idea, and Lighthill was honest enough to say it was the idea of a mathematician rather than a computer scientist. He was "inclined, as a mathematician, to single out one rather general cause for the disappointments that have been experienced: failure to recognise the implications of the combinatorial explosion", which he defined as "a general obstacle to the construction of a self-organising system on a large knowledge base which results from the explosive growth of any combinatorial expression, representing numbers of possible ways of grouping elements of the knowledge base according to particular rules, as the base's size increases."⁷

Strip the sentence of its clauses and the point is plain, and it was correct. The programs of the 1960s worked by search. Faced with a problem, they considered the possible moves, and the possible moves after those, and so on, and pruned the tree of possibilities with rules of thumb. In a toy world with a dozen blocks on a table, the tree was small enough to climb. Add a hundred blocks, or the contents of an ordinary kitchen, and the tree grew faster than any computer then built or then imaginable could follow. The successes that had been advertised as first steps were, on this reading, not first steps at all but the whole of the walk, because the ground beyond them rose vertically. Lighthill's evidence was chess. Some twenty-five years after the work began, "the best programs play chess of only experienced amateur standard characteristic of county club players in England", and what progress there had been, he said, had come by heuristic methods in which the intelligence was the programmer's and the computer supplied only speed.⁸ His verdict on the field was that "in no part of the field have the discoveries made so far produced the major impact that was then promised", and on the bridge, that the "attempt to construct a true Bridge between categories A and C is not succeeding".⁸

It is worth pausing on how strong this was, because the temptation half a century later is to read Lighthill as a philistine. He was nothing of the kind. The combinatorial explosion was and is real; it is the reason that chess was not solved by search alone, and the reason the programs of 1973 could not be scaled up by buying a bigger machine. What he could not see, because nobody could, was that the answer would come from the category he had half-praised rather than the one he had condemned: from C, from systems that learned their own rules of thumb from enormous quantities of examples, running on hardware whose cheapness he would have thought a misprint. The problem he diagnosed was the right problem. The cure arrived by a road he had not marked on the map, and it arrived forty years late, which in the life of a funding council is the same as never.

The page nobody expected

There is one passage in the survey that its admirers do not quote and its enemies never stopped quoting. It comes in the discussion of Category B, where Lighthill turns from what robot-builders do to why they do it. He is speculating, and says so. "Incidentally," he writes, "it has sometimes been argued that part of the stimulus to laborious male activity in creative fields of work, including pure science, is the urge to compensate for lack of the female capability of giving birth to children."⁹ He follows it with an exclamation mark: "If this were true, then Building Robots might indeed be seen as the ideal compensation!" And then, a paragraph of evidence later, he adopts it: "the view to which this author has tentatively but perhaps quite wrongly come is that a relationship which may be called pseudo-maternal rather than Pygmalion-like comes into play between a Robot and its Builder."⁹

This is a Lucasian Professor, in a document commissioned by a research council to decide the allocation of public money, proposing that the men who build robots do so because they cannot have babies. One can only guess at the faces in Edinburgh when they read it. What the passage reveals is not that Lighthill was a fool, which he was not, but that the survey was two documents in one. The first was a rigorous argument about search, made by a man who understood exponential growth better than anyone in the country. The second was an outsider's impression of a tribe he found faintly ridiculous, and the second was allowed to sit inside the first with the same authority, because the whole point of commissioning an outsider was that his impressions counted. The Council had asked for a view untainted by the field's enthusiasms. It got one untainted by the field's knowledge as well, and it did not, in 1973, distinguish between the two.

The replies

The four replies bound with the survey were written by men who knew they had lost. Stuart Sutherland, the Sussex psychologist, and Roger Needham, the Cambridge computer scientist, took the report seriously and disputed its lines. Christopher Longuet-Higgins, the theoretical chemist who had moved to Edinburgh to found its cognitive science, argued for the science. Michie, whose department was the one most obviously on trial, wrote the reply of a man defending a life's work, and it is the best of them.⁵

His first move was to reject the alphabet. The three categories, he argued, were not a description of the field but a way of denying it existed, since each letter assigned the work to some older discipline, control theory, science fiction, neurobiology, and left nothing over. The subject, in the definition his reply set out, "is concerned with machines, and in particular computers, displaying characteristics which would be identified in a human being as intelligent behaviour", and that was a subject in its own right or it was nothing.⁵ His second move was on the bridge. Lighthill had made a robot the link between engineering and biology; Michie thought this confused a piece of apparatus with a body of theory. "It is surely more fruitful, if one seeks inter-disciplinary connections, to choose a common body of theory rather than to seize on a piece of laboratory equipment."⁵ Aerodynamics, he pointed out, was the bridge between aeroplane design and the study of birds; the wind tunnel was only where you tested it. On this view Freddy was a wind tunnel, and one might add what Michie was too careful to say: nobody had ever cancelled aeronautics because a wind tunnel could not fly.

His third move was a number. The American effort, he wrote, ran at "at least twenty times the scale" of the British.⁵ This was intended as a plea and reads today as a prophecy. A funding council that withdrew from a field in which its rivals were spending twenty times as much would not be saving money so much as choosing which country the results would belong to.

On the television, the same arguments were made with more heat and to more effect. McCarthy, who had the advantage of not being British and therefore of not needing the Council's money, said what the others were too polite to say, that artificial intelligence was a science and the report had not understood what science it was. Gregory, asked whether there was any reason in principle an artificial brain could not be built, gave the reply of a man who worked in an anatomy department: "I see no reason why we cannot, in fact, make brains because they exist physically."¹ Michie read out a nineteenth-century dismissal of the telephone and got a laugh.¹ Lighthill, unhurried, conceded nothing and gave the debate its one durable line: "The general purpose robot, then, is a mirage."¹ The laughter was Michie's. The decision was not.

What a winter looks like

The phrase AI winter arrived a decade later and in America, as the subject of a debate at the 1984 meeting of the American Association of Artificial Intelligence,¹⁶ and it is usually told as an American story of over-promising and defence-budget retrenchment. But the first winter was British, and it was not a change in the weather. It was a decision, taken by a public body on the advice of one man, and it can be dated.

At Edinburgh the effect arrived through the university's own machinery, and it is worth being careful here, because the department's own historian tells it differently. Jim Howe, who later ran the department, writes that the 1974 review was called because the School structure "had failed to reduce tensions between senior staff", and, as we have seen, that the Council had turned to Lighthill in the first place because it had heard of the discord.¹⁰ The feud and the report were entangled from the beginning, and which did more damage is this essay's reading, not Howe's. What is not in dispute is what the review did. The School of Artificial Intelligence, the federation of three departments and a unit the university had set up in 1970 when Gregory resigned, was "scrapped in favour of a single department", the Department of Artificial Intelligence, under Bernard Meltzer; and "a separate unit, the Machine Intelligence Research Unit, was set up to accommodate Michie's work".¹⁰ Accommodate is the word the history uses and it is the right one. The man who had persuaded the university to recognise the subject in 1965, with an Experimental Programming Unit, and had been given a Department of Machine Intelligence and Perception around him in 1966, was given a unit. Gregory had gone in 1970. Longuet-Higgins went to Sussex in 1974.¹⁷ The Chilton Computing history, written by veterans of the Atlas laboratory that served the Council, puts it plainly: the report "provoked a massive loss of confidence in AI by the academic establishment in the UK including the funding body. It persisted for almost a decade."⁴ Support for the subject was withdrawn from most British universities. Edinburgh, Sussex and Essex kept theirs, diminished.¹¹

Freddy was not switched off, not at once. It went on working, on what money there was, into 1981, and then it went to the museum.³ The robot that had been shown on television as the future of British engineering has stood in a glass case at the National Museum of Scotland since 2006, with its original Pye television camera still in place.³ Visitors who read the label learn that it could build a toy car and a toy boat. They do not usually learn that a professor of fluid dynamics decided this was not worth knowing how to do.

Michie did not stop. He ran his unit and edited the Machine Intelligence volumes, the field's British series of record since the 1960s. In June 1983, at fifty-nine, he founded the Turing Institute in Glasgow. The next year it was writing landing software for the American space shuttle under contract.¹² He died in 2007, in a car accident, at eighty-three, having outlived the winter and seen the thaw and the first days of the summer that followed.¹²

The thaw was bought abroad

What ended the winter was not a reconsideration of the argument. Nobody in Whitehall re-read Lighthill and decided he had been wrong about search. What ended it was fear of a foreign country. In 1982 Japan launched its Fifth Generation Computer project, a national programme to build the thinking machines that Britain had just decided were a mirage. The British state, which had been unable to find money for four departments a decade earlier, found £350 million for a five-year programme in advanced information technology, £200 million of it from the government and the rest from industry.¹³ The Alvey Programme, named for the British Telecom technology director who chaired the committee that proposed it, was launched in 1983.¹³¹⁴ At Edinburgh the academic staff in artificial intelligence went from four to fifteen.¹⁰ The subject that had been too speculative to fund in 1973 was, by 1983, too strategic to leave alone. Nothing about the subject had changed. The rival had.

And then, because this is how the story goes, the government "rejected proposals to create a follow-up project", and the years after Alvey are called the second winter, and in the 1990s "support for AI research stopped being systematic".¹³ The pattern is not a cycle of the science. It is a cycle of the state's attention, which turns on and off according to who else is spending.

Who decides

The interest of the episode now is not whether Lighthill was right. He was right about the explosion and wrong about where the exit was, which is the usual condition of a clever person forecasting a field they do not work in. The interest is in the shape of the decision. A public body wanted to stop funding something and did not want to be seen to have decided this itself. It found a man of standing from outside the field, gave him a few months, and published his opinion with the dissents attached so that the process looked like a hearing. A television debate supplied the appearance of a public that had been consulted. Then the departments were reorganised by the universities' own committees, so that no minister ever had to say the words. The people who had spent their lives on the subject were not overruled. They were accommodated.

Half a century on, the direction of the argument has reversed and the shape has not. The company that now builds the machines Lighthill called a mirage proposed this week, in a document of great politeness, that governments agree on technical standards which "would not be licenses, mandatory prerelease review, or approval requirements for AI models", and it added, fairly, that national governments would decide whether to write those standards into law.¹⁵ What that amounts to, in this essay's reading and not in the document's words, is a request that the deciding about how fast to go should stay with the people going. In 1973 the state took the decision away from the practitioners and handed it to one outsider. In 2026 the practitioners ask the state to agree the measuring sticks and leave the measuring to them. Both arrangements have the same defect. In neither does the person who will live with the result get a vote. The Edinburgh graduate student who watched the department reorganised around him in 1974 had no more say than the reader of this essay will have in whether a research loop closes without a human in it. The alphabet has changed. The absence is the same.

Lighthill died in 1998, aged seventy-four, of a heart valve that gave way while he was swimming round the island of Sark, a nine-mile circuit he had been the first person to complete, in 1973, the year of the report, at forty-nine, and had repeated half a dozen times since.² It is hard not to like him for it. He had, in the end, a mathematician's confidence that a problem with a clear structure would yield to a clear mind, and a swimmer's confidence that the tide could be read. The sea proved him wrong about the second. The machines, at some length, proved him wrong about the first. He was one of the few people in the history of the subject to be given the power to stop it, and one of the few honest enough to write down, in the same document, that he might "perhaps quite wrongly" have misjudged the people he was stopping. Nobody at the Council read that clause as an instruction. It was.