The Dream of the Mechanical Brain: The Rise and Fall of AI

Herbert G. Klein

Enculturation, Vol. 3, No. 1, Fall 2000

About the Author
Table of Contents

The history of computing is the history of the attempted mechanisation of the human mind. The field of Artificial Intelligence which came into being as such in the 1950s has roots which go back to the very beginnings of the theory of computing. The original driving force was, as will be shown, less the wish to make calculating less boring or more efficient, but rather the desire to include human beings in the effort to interpret all natural phenomena as fully amenable to mathematical description. Not only the human body, as had been propagated by René Descartes, but also the human mind was to be regarded as a machine. One of the indispensable conditions for the mechanising of mental faculties is their formalisation: any process which can be formalised, can be mechanised. Logic, the obvious means to this end, had been around since antiquity, but had to be updated in order to meet modern requirements. The human ability to reach conclusions from premisses which do not explicitly contain these conclusions--a form of creativity--could be formalised, if there was a set of abstract symbols and methods, which made "automatic" deduction possible. The German mathematician Gottfried Wilhelm Leibniz (1646-1716) had evolved the idea of a symbolic logical representation and he had hoped to construct a "universal grammar" with which all possible logical sentences, including scientific proofs, could be constructed. This would have made human creativity largely superfluous, because many tasks could have been managed just as well (or even better) by a machine. Conversely, formalisation would turn the human mind into a machine--just like Descartes had turned the body into a machine. This is why Norbert Wiener sees Leibniz as the father of the mechanisation of the mind and therefore of the modern computer:

If I were to choose a patron saint for cybernetics out of the history of science, I should have to choose Leibniz. The philosophy of Leibniz centers about two closely related concepts--that of a universal symbolism and that of a calculus of reasoning. From these are descended the mathematical notation and the symbolic logic of the present day. Now, just as the calculus of arithmetic lends itself to a mechanization progressing through the abacus and the desk computing machine to the ultra-rapid computing machines of the present day, so the calculus ratiocinator of Leibniz contains the germs of the machina ratiocinatrix, the reasoning machine. . . . It is therefore not in the least surprising that the same intellectual impulse which has led to the development of mathematical logic has at the same time led to the ideal or actual mechanization of processes of thought.

As Leibniz did not progress beyond the construction of very elaborate calculating machines, the next important step was taken by George Boole in 1854 with the publication of his book The Laws of Thought. The formal logic developed in this work indeed provides the ground for the mechanisation of thought, as it forms the basis for modern computing: already in 1903, Nikola Tesla was granted a patent for the AND-switch, which is still used in practically all computers today. (This is true for all conventional computers, i.e. those built upon the lines of the von-Neumann architecture.) As far as the theoretical and intellectual conditions are concerned, the computer age could have started there and then--if hardware technology had been advanced enough.

Boole himself was of the firm conviction that it was possible to describe the working of the human mind with the laws of formal logic, although he did not claim to have found a causal explanation:

It may, perhaps, be permitted to the mind to attain a knowledge of the laws to which it is itself subject, without its being also given to it to understand their ground and origin, or even, except in a very limited degree, to comprehend their fitness for their end, as compared with other and conceivable systems of law.

Boole's formal language uses symbols as variables, i.e. they are assigned particular meanings for specific operations. As Johnson-Laird points out, there are no fixed rules for this assignation of meanings, but only for the manipulation of the symbols. Nevertheless, it is claimed that this formalism is a sufficient basis for deciding whether statements about the world are "true" or "false". However, there is no possibility of "objectively" verifying their premisses, since there is no description-independent reality: formal languages can only talk about symbolic worlds. We are therefore dealing with a closed system which only permits certain kinds of statements. The original aim of making statements about reality is given up in favour of the inner consistency of the system. Thus it becomes important to prove that such consistency is possible. The German mathematician David Hilbert asked for just such a proof in his famous "Hilbert program". Hilbert was not merely interested in the scientific aspects, rather he was moved by a deep need for epistemological reassurance, comparable to Descartes': "If mathematical thinking is defective, where are we to find truth and certitude?"

The English philosopher Bertrand Russell shared this longing for absolute certainty: "I wanted certainty in the kind of way in which people want religious faith. I thought that certainty is more likely to be found in mathematics than elsewhere." Together with A.N. Whitehead, Russell tried to found a non-contradictory philosophy, based on Hilbert's groundwork, in Principia Mathematica (1925), which should prove the possibility of objective knowledge. The basic idea was to reduce the problem of the consistency of mathematics to that of the consistency of formal logic. But only a few years later, in 1931, Kurt Gödel was able to prove the impossibility of logical consistency within closed systems by showing that there are always possible statements within such systems whose truth value is undecidable. This does not only mean that mathematics and formal logic are neither complete nor consistent, but even further that it is impossible to declare which statements are undecidable and which are not. This has implications that do not only seriously affect mathematics and philosophy, but also the question whether it is possible to formalise and thus computerise mental processes. The answer seems to depend on whether the mind can be seen as an open or a closed system. As J.R. Lukas puts it succinctly: "The Gödelian formula is the Achilles' heel of the cybernetical machine".

One way out of this dilemma is to deny that the human mind is anything but a rather simple machine. This road is taken by the English mathematician Alan Turing who asserted in 1936 in his essay "On Computable Numbers, With an Application to the Entscheidungsproblem" that Hilbert's problem of decidablity was itself undecidable: ". . . the Entscheidungsproblem cannot be solved." Turing shows this by means of a theoretical model, the so-called Turing-machine, which can do all kinds of computation, although in a very simple and time-consuming manner. This machine, as Turing went on to show, is (unsurprisingly) not able to make any predictions about its own behaviour, (i.e., it is not capable of self-reflexion). If, and only if, the human mind is analogous to such a machine, then indeed the human mind is not capable of self-reflexion either. Such a conclusion was very much what Turing had intended, for he was of the opinion that what is called intelligence in human beings was based on formalisms which could just as well be used by machines. In his essay he describes the working of the human mind while being occupied with arithmetical operations in terms which are directly applicable to a machine: "The behaviour of the computer at any moment is determined by the symbols which he is observing, and his 'state of mind' at that moment." The fact that the term "computer" in this description refers to a human being involved in a certain activity, and that only a few years later it will be used for a general purpose machine, shows the strong suggestiveness of the analogy. As Turing's biographer Andrew Hodges puts it: ". . . the Turing thesis is that the discrete-state-machine model is the relevant description of one aspect of the material world--namely the operation of brains."

Shortly after the end of WW2, during which Turing had participated in the cracking of the German ENIGMA-code, he declared that he wanted "to build a brain". This hubristic claim is only possible through a severe reduction of what is encompassed by the word "intelligence." This, indeed, is a procedure that can be traced from Hobbes to some of today's proponents of AI. Turing was aware that a realisation of his ideas was only possible through electronic means. Although at this time there were already computers in existence, they were far from being able to fulfil Turing's demands. Nevertheless, he invented a test (the "Turing-test") which should be capable of showing the comparability of human and machine intelligence, once such a technologically advanced machine should have been built. In his essay "Thought and Machine Intelligence" he develops a set-up, which has a (female) human being and a machine electronically communicate with a third party (a male human), who is then to decide which is the woman and which the machine. Turing prognosticates:

I believe that in about fifty years' time it will be possible to programme computers, ..., to make them play the imitation game so well that an average interrogator will not have more than 70 per cent. chance of making the right identification after five minutes of questioning. The original question, 'Can machines think?' I believe to be too meaningless to deserve discussion.

This neglect of the question whether what the machine does can be called thinking at all, is symptomatic for Turing's approach, as is also shown by the following claim: "May not machines carry out something which ought to be described as thinking but which is very different from what a man does?" Although man and machine should only be judged by their results, not by the means of reaching these results, Turing still considered it useful to think of the human mind as "a machine of this kind," (i.e., a digital computer. The question of consciousness is brushed aside as irrelevant):

I do not wish to give the impression that I think there is no mystery about consciousness. There is, for instance, something of a paradox connected with any attempt to localise it. But I do not think these mysteries necessarily need to be solved before we can answer the question with which we are concerned in this paper.

As a matter of fact, Turing thought it possible that the mechanical imitation of cognitive processes might be a way of solving "these mysteries." He did not doubt that there would be such machines some day, but he had only vague ideas about their construction. One way, he thought, might be an approach through the solution of abstract problems like playing chess, another teaching the computer various things as one would a child. Whichever way might eventually lead to the desired goal, there is no question about the goal itself: "We may hope that machines will eventually compete with men in all purely intellectual fields."

An important step towards this goal was taken by Shannon and Weaver, who published their groundbreaking work The Mathematical Theory of Communication in 1949. As both authors were at pains to make clear, they did not intend to make claims that went beyond purely technical processes. The first part of the book is a technical article, in which Shannon describes his work on electronic communication. He had limited himself to the mathematically formalisable aspects of information, deliberately excluding their meaning: "Frequently the messages have meaning; . . . These semantic aspects of communication are irrelevant to the engineering problem. The significant aspect is that the actual message is one selected from a set of possible messages."

The most important feature of Shannon's work turned out to be his idea of declaring all regularities of the received signal as redundant, whereas its irregularities were said to be the carriers of "information". He created a formula for this of which he says: "The form . . . will be recognized as that of entropy as defined in certain formulations of statistical mechanics. . . ." Weaver took it upon himself to make Shannon's findings accessible to a wider public and to show some of their consequences. First of all he makes clear that their approach is a very general one: "The word communication will be used here in a very broad sense to include all of the procedures by which one mind may affect another." Thus it is applied not only to machines, but e.g. also to the arts. At the same time, Weaver insists that the theory is only concerned with the description of formal aspects of communication, not with its contents: "The word information, in this theory, is used in a special sense that must not be confused with its ordinary usage. In particular, information must not be confused with meaning".

These are clear words, but they did not prevent others to misapply the technical use of the term "information" to the field of human communication, which thereby could supposedly be described in a rather simplified technical manner. This then permits the shift from the content to the form of the communication which in turn makes any difference between man and machine disappear, as in the following definition by J. L. Massey: "An information source is purely and simply any device (man, machine, or other) whose output symbols are not perfectly predictable in advance." Meaning is thereby replaced by the formalisable and quantifiable term "information," and language comes to be seen as a set of symbols which can be manipulated without regard to their meaning. This in turn makes thought processes appear as nothing but the production and manipulation of content-independent symbols. Indeed, this is already inherent in Shannon's approach, and it is therefore fitting that he was also considering formalising intellectual tasks so that they could be carried out by a machine. Weaver explicitly points towards this connection in his commentary:

. . . the ideas developed in this work connect so closely with the problem of the logical design of great computers that it is no surprise that Shannon has just written a paper on the design of a computer which would be capable of playing a skillful game of chess. And it is of further direct pertinence to the present contention that this paper closes with the remark that either one must say that such a computer "thinks," or one must substantially modify the conventional implication of the verb "to think."

In the above mentioned article, "A Chess-Playing Machine," which was published in Scientific American in 1950, Shannon does indeed argue that computers, as they had then been developed, were already close to rational thought. The advantage of teaching machines to play chess was, in his eyes, that on the one hand the goal as well as the means to reach it were clearly defined, on the other it was neither too trivial nor too complex a problem. Not the least advantage lay in the fact that a machine could directly play against a human being. He then reaches the following conclusion: "From a behavioristic point of view, the machine acts as though it were thinking. . . . If we regard thinking as a property of external actions rather than internal method the machine is surely thinking". This conviction is shared by Norbert Wiener who re-applies the mechanical concept to living organisms: "Now that the concept of learning machines is applicable to those machines which we have made ourselves, it is also relevant to those living machines which we call animals, so that we have the possibility of throwing a new light on biological cybernetics". Wiener's thinking resembles Turing's very much in the easy equation of biological and mechanical processes, although Wiener saw this more as an analogy than as sameness. This is then again the Cartesian concept of the animal as a machine, with the difference that this time it is not used to explain physiology as mechanics, but rather cognition as formalism--and Wiener leaves no doubt that he includes humans in the category of animal. The theoretical groundwork for these ideas was laid in Wiener's programmatically titled book Cybernetics or Control and Communication in the Animal and the Machine (1948). The term "cybernetics" is derived from the Greek for "steersman" and was coined by Wiener and Arturo Rosenblueth in 1947. It was to become the key term of a new paradigm.

Norbert Wiener, a pupil of Bertrand Russell, had participated in the war effort like Turing and Shannon. His original aim had been the development of a self-regulating steering device for anti-aircraft missiles. Although this endeavour was not successful, it led to the idea of cybernetics as a theory of self-regulating systems. As Wiener kept emphasising, he saw technical and biological systems as governed by the same principles: ". . . it became clear to us that the ultra-rapid computing machine, depending as it does on consecutive switching devices, must represent almost an ideal model of the problems arising in the nervous system." The common denominator is the transmission of information, which can be described independently of the material basis:

. . . it had already become clear . . . that the problems of control engineering and of communication engineering were inseparable, and that they centered not around the technique of electrical engineering but around the much more fundamental notion of the message, whether this should be transmitted by electrical, mechanical, or nervous means.

Wiener claims that he had conceived the idea of a statistical information theory at the same time as Shannon. But Wiener emphasises the role of information even more and turns it into the measure of all things, claiming that it should make possible the description of all aspects of reality, material as well as mental. All intelligent systems are therefore considered as automata, whose behaviour can be explained through the means of communication theory:

In short, the newer study of automata, whether in the metal or in the flesh, is a branch of communication engineering, and its cardinal notions are those of message, amount of disturbance or "noise"--a term taken over from the telephone engineer--quantity of information, coding technique, and so on.

Wiener makes it clear that he is especially interested in the imitation of human thought processes and that his aim is to replace them--at least in part--by mechanical means. Like Turing and Shannon he believes that the construction of chess automata might be an important step towards this goal. These various approaches were eventually taken up by a group of scientists who participated in the famous "Dartmoor Conference" of 1956, among them Allen Newell, Herbert Simon, Marvin Minsky, Oliver Selfridge and John McCarthy. They pioneered the field that came to be known as Artificial Intelligence (AI), which is described by Minsky in a well-known definition in the following manner: "The field of research concerned with making machines do things that people consider to require intelligence". The most promising approach to this end seemed to be the use of symbolic logic, which permits the application of the same formalism to different contents. Based on this idea, Newell, Shaw and Simon developed the concept of a "General Problem Solver," "designed to model some of the main features of human problem solving." Their aim was the development of a program that should be capable of solving all kinds of problems on the basis of a few general rules without regard to their specificity. The complexity of reality would thereby be reduced to a formula. The foreseeable failure of this enterprise led to the restriction of later projects to severely limited "knowledge domains", like the program "Dendral" for chemical analysis or "Mycin" for the diagnosis of certain infectious diseases. Systems like these, which were first called "expert systems" and later "intelligent assistants" were indeed--and are--successful to a certain degree, but they can hardly be called intelligent. This led to the development of two diverging paths in AI: the one, usually called "soft AI", has given up any pretensions of emulating the human mind and is content with making computer programs more flexible, the other, called "hard AI", still clings to the idea that there is no intrinsic difference between human and machine intelligence. Thus Marvin Minsky maintains:

There is not the slightest reason to doubt that brains are anything other than machines with enormous numbers of parts that work in perfect accord with physical laws. As far as anyone can tell, our minds are merely complex processes.

Perhaps the latter is true, and minds are "merely" complex processes, but, as I have tried to show, the attempt to formalise them has so far been hardly successful. On the contrary, the specious analogy between the mind and the machine has led to severe simplifications in thinking about the mind. In the wake of pioneers like Wiener and Minsky, AI research really started booming with the large-scale production of ever faster computers. However, the prognostication that more CPU-power would lead to more intelligent machines was sadly disappointed. Even though the famous chess-playing problem seemed to have finally found a triumphant solution with "Deep Blue's" victory over Garry Kasparov in 1997, the cries of victory rang hollow: as Raymond Kurzweil had suggested before: "When this happens, . . . we shall either think more of computers, less of ourselves, or less of chess. If history is a guide, we will probably think less of chess." As a matter of fact, IBM has announced that it will discontinue research in this direction. Other, less spectacular endeavours, however, have reached more lasting success: expert-systems are now working with good results in certain limited areas. However, the claim that they are intelligent in any meaningful way has been quietly dropped along the way. Instead, great things are now being promised for other approaches: Neural Networks and Artificial Life are said to constitute the high-road to the visionary goal. They do indeed differ from the old approach in that they are self-organising within certain given parameters. This means that the behaviour of these systems is not fully predictable or computable. However, it does not appear so far that these systems are any more "intelligent" than the conventional ones.

Norbert Wiener may have had something along these lines on his mind, when he envisioned autonomous cybernetic systems, but he thought that they also constituted a grave danger: he believed that machines could reach a stage at which they might become so powerful that they would dominate humans. As Katherine Hayles points out, there is indeed a disparity between the idea of self-regulation considered as a precondition of individual freedom and as the basis for more or less autonomous machines - a disparity that Wiener saw but simplified by reducing it to the difference between "good" (flexible) and "evil" (rigid) machinery without really making clear what the difference was supposed to consist in. Since cybernetics is about "control", the tension is perhaps inevitable. The history of AI, however, has shown that it is less the dominance of all-powerful machines that is to be feared than a way of thinking that subjects humans to the machine metaphor: thinking of machines as working on analagous lines to the human mind easily leads to the reverse assumption that minds are nothing but machines. Not recognising the difference between man and machine is thus one of the reasons for the shift towards the "posthuman" that Hayles diagnoses. The question should be asked, however, what interests are served by such a facile elision of differences. If we believe that we are just machines, then the machines have won indeed - and Wiener's fears would have come true in a grimly ironic way.

Works Cited

Boole, George. The Laws of Thought. On Which are Founded the Mathematical Theories of Logic and Probabilities. 1854. New York: Dover, 1961.

Crosson, Frederick J. and Kenneth M. Sayre, eds. Philosophy and Cybernetics. Essays delivered to the Philosophic Institute for Artificial Intelligence at the University of Notre Dame. London: University of Notre Dame Press, 1967.

Davis, Philip J. and Reuben Hersh. The Mathematical Experience. 1980. Harmondsworth: Penguin, 1988.

Gödel, Kurt. "Über formal unentscheidbare Sätze der Principia Mathematica und verwandter Systeme, I." Monatshefte für Mathematik und Physik 38 (1931): 173-198.

Hayles, Katherine. "Liberal Subjectivity Imperiled: Norbert Wiener and Cybernetic Anxiety."

Herken, Rolf, ed. The Universal Turing Machine: A Half-Century Survey. Hamburg & Berlin: Springer, 1988.

Hodges, Andrew. "Alan Turing and the Turing Machine." In Herken, 3-15.

Hodges, Andrew. Alan Turing. The Enigma. London: Burnett, 1983.

Johnson-Laird, P. N. The Computer and the Mind. An Introduction to Cognitive Science. Cambridge, MA: Harvard UP, 1988.

Kurzweil, Raymond. The Age of Intelligent Machines. MIT, 1990.

Lukas, J. R. "Minds, Machines and Gödel." Philosophy 36 (1961), 112-127.

Massey, J. L. "Information, Machines, and Men." In Crosson, 37-69.

Minsky, Marvin. The Society of Mind. New York: Touchstone, 1988.

Newman, James R., ed. The World of Mathematics, Vol. 4. A Small Library of the Literature of Mathematics from A'h-mosé the Scribe to Albert Einstein. New York: Simon & Schuster, 1956.

Shannon, Claude E. and Warren Weaver. The Mathematical Theory of Communication. 1949. Urbana: University of Illinois Press, 1962.

Shannon, Claude E. "A Chess-Playing Machine." In Newman, 2124-2133.

Simon, Herbert A. The Sciences of the Artificial. 1969. 2nd ed. Cambridge, MA and London: Harvard UP, 1981.

Turing, Alan M. "On Computable Numbers, With an Application to the Entscheidungsproblem." Proceedings of the London Mathematical Society (Second Series) 42 (1937): 230-265.

Turing, Alan M. "Can a Machine Think?" In Newman, 2099-2123.

Wiener, Norbert. Cybernetics or Control and Communication in the Animal and the Machine. 1948. 2nd ed. New York and London: M.I.T. Press/Wiley, 1961.

Copyright © Enculturation 2000

Home | Contents 3:1 | Editors | Issues
About | Submissions | Subscribe | Copyright | Review | Links