Edit History
Search Keywords
Simon Schaffer
Charles Babbage
Ada Lovelace
Difference Engine
Alan Turing
Artifical Intelligence

The Chinese Room
part 5

Babbage and his collaborators never found it very easy to teach their audiences just where the intelligence of the calculating engines really was. This became even more true when the Analytical Engine was launched in the 1830s. At the end of 1837, Babbage composed a long memoir on its powers, noting that 'in substituting mechanism for the performance of operations hitherto executed by intellectual labour, the analogy between these acts and the operations of mind almost forced upon me the figurative employment of the same terms'. Phrases like 'the engine knows', he confessed, were simply irresistible even though they might be misleading. And the lengthy Sketch of the Analytical Engine, a joint effort of the early 1840s by Ada Lovelace and Babbage's Piedmontese admirer, the mathematician Luigi Menabrea, urged that 'it is desirable to guard against the possibility of exaggerated ideas that might arise as to the powers of the Analytical Engine'. It was a slave which could only follow what it was ordered to do, could originate nothing, was not, in fact 'a thinking being, but simply an automaton which acts according to the laws imposed upon it'.

But there were real dangers in the apparently innocent word 'automaton'. Ada Lovelace, who in 1843 called herself 'the High Priestess of Babbage's Engine', famously explained how it worked by comparing it with the best-known programmable machine of the automatic weaving system, the Jacquard Loom, a device whose introduction had almost completely destroyed the jobs of silk-weavers in London's East End. 'We may say most aptly', she noted in the Sketch, 'that the Analytical Engine weaves algebraical patterns just as the Jacquard Loom weaves flowers and leaves'. Lovelace never raised the problem of the substitution of weavers' intelligence by a series of automatic program cards nor the consequent sufferings of London's skilled unemployed. Instead, she directed her polite readers to the Adelaide Gallery, where they would see the Loom at work. But such galleries were scarcely likely to resolve the problem of exactly whether the engines could think. In fact, just as in Babbage's reception rooms the Silver Dancer and the Difference Engine stood next door to each other, so at the Adelaide Gallery the Jacquard Loom 'in daily operation' stood in the next room to a splendidly automatic Chinese juggler. It was exactly in such places that the distinction between entertaining automata and rational engines was all too easily effaced, and it was there, too, that all the puzzles of mechanized intelligence were graphically put on show.

There is a tempting contemporary resonance to these stories of dancers, Turks, chess and calculating engines. The currently canonical way of telling whether a machine is intelligent explicitly involves deceiving an audience in a manner all too reminiscent of Maelzel's shows at Spring Gardens. The so-called Turing test requires the construction of a hidden device which can produce outputs capable of convincing human judges that it is one of them. Alan Turing, brilliant Cambridge-trained mathematician and veteran of the secret wartime campaign to crack the German Enigma code, was a keen reader of Babbage and Lovelace and much concerned with the problems of automating chess. 'One can produce paper machines for playing chess', Turing wrote in 1948, 'playing against such a machine gives a definite feeling that one is pitting one's wits against something alive'. In 1950, after a debate at Manchester University on the possibility of making intelligent machines, he wrote a paper proposing what he called an 'imitation game', in which a man and a machine would both feed typed answers to a judge concealed in a different room. Turing's own life was soon destroyed by homophobic persecution, so it is intriguing that his first version of the imitation game involved judging which of two invisible respondents was a man, which a woman. In the later test of human and machine, if the judge could not tell which was the man, then the computer would pass the intelligence test. Turing's 1950 paper explicitly discussed Lovelace's ideas about whether the Analytical Engine could be truly intelligent, while in an earlier version he proposed, like Babbage, programming random elements into the computer so as to increase its capacity for innovative intelligence. Turing over-optimistically predicted that by our century's end computers should have developed so much that they would win a five-minute game at least 3 times in 10, and public computer competitions, funded by a manufacturer of portable disco dance floors and directed by a behaviourist psychologist, are run nowadays in California on exactly the lines Turing set out.

The Turing test is about concealment and detection. Its appeal hinges on the place where intelligence is to be found inside a space to which access is forbidden. In a celebrated paper published in 1980, the philosopher John Searle argued against the conventional interpretation of the Turing test. He proposed a device of which von Kempelen would have been proud, a 'Chinese room' occupied by a human being completely devoid of intelligence about Chinese but supplied with a set of symbols and rules which would allow response to inquiries from outside the room. Searle envisaged that such a system might pass the Turing test by being indistinguishable from native Chinese speakers to anyone outside. But neither necessary nor sufficient conditions could be given for attributing intelligence to such a system. In response to the suggestion that while the room's occupant might not possess intelligence about Chinese, yet the entire system might be said to do so, Searle countered that the enterprise of artificial intelligence 'must be able to distinguish the principles on which the mind works from those on which non-mental systems work', and to judge that a system is intelligent just because of its imputs and outputs would force us to attribute intelligence to a wide range of non-mental systems. The parable of the Chinese room dramatises the spatial mode of such debates, by insisting on a definition of the place where intelligence might be said to reside.

Here the geography of intelligence is not simply a matter of mundane showmanship, but also relies on the exoticism of distance and the esotericism of concealment. In many western myths of mechanical intelligence, with Chinese or Japanese, Turks or Nazis as their protagonists, aliens are automata, mindless subjects of tyranny; they build automata, because they possess fiendish cunning; and they conceal what they have done, because they desire to master us. In the world of Babbage's Dancer, mechanical imitation seemed most at home in oriental climes of which rather little was supposed to be known and almost anything might be credited. In Turing's world, as his brilliant biographer Andrew Hodges points out, intelligence meant secretive messages passed within a guarded coterie and the cryptanalysis of enemy codes as least as much as computers' capacity to imitate human beings. At the end of Gibson and Sterling's fantasy, the cunning orientals would inherit the Earth by mastering machine programming and British engines fail through dark conspiracies. There is, perhaps, a long-term political and aesthetic relationship between intelligent automata, orientalism and the covert. The tale of the automata and their impresarios confirms that the places whence machines come, where machines are put on show and the places within machine-systems where intelligence is supposed to reside raised, and still raise, delicate political and philosophical issues.

Most of these issues hinge on the problem of work and its visibility. Babbage never reconciled himself to the workforce on which he relied; von Kempelen exploited preconceptions about the role of skill; Turing explicitly ruled out any computational task which required the use of a body. Much is made of the collaborative work required from human beings to make their machines look expert and intelligent in a recent book by the sociologist Harry Collins, Artificial Experts (1990). 'One of the reasons we tend to think a calculator can do arithmetic', Collins suggests, 'is the natural way we help it out and rectify its deficiencies without noticing. All the abilities we bring to the calculation - everything that surrounds what the calculator does itself are so widespread and familiar that they have disappeared for us'. This is the point of the tale of the Dancer and the Difference Engine. The intelligence attributed to machines hinges on the cultural invisibility of the human skills which accompany them. In Babbage's devices, the skills which surrounded automatic mechanization were systematically rendered invisible. Then and only then might any machines seem intelligent. The moral about the politics and geography of serious trickery is certainly worth remembering. If such machines look intelligent because we do not concentrate on where their work is done, then we need to think harder about the work which produces values and who performs it.

About the site