Can Machines Think?

In considering this question it is very easy to be coerced by the rapid progression of Technology. As Fifth Generation computing pushes back the boundaries of our knowledge of machines and their capabilities, and the Human Genome Project maps out our knowledge of the mechanisms' of the human organism it is easy to lay aside scientifically specific detail and commit ourselves firmly to the belief that eventually the two will meet up; knowing what makes us human we will be able to reproduce these characteristics in mechanistic form. But this approach is very superficial and tenable by only the most optimistic of technocrats. Whereas to tip to the opposite extreme one would turn to ideological philosophy and attempt to answer this empirical question in the manner of Seventeenth Century philosophers who searched introspectively for the truth'. Balancing between these two diametric approaches we have the benefit of the resource of contemporary thought and opinion, as long as we are clear about the relevance of this information to our needs. To do this we must redefine and expand the question and its implications to give a clearer path to take.

A machine is a tool, its purpose and design are determined by its designer and as such it is rational and analytical. Machine range from the simple, such as a wheelbarrow which fulfils the need to move bulky objects and is designed to achieve this aim, to the most complex of computers such as the chess computers whose creators intend for them to win against the best human opposition and have this purpose designed into them in the printed circuits and algorithms by which they operate. Machines do not have a self created aim nor do they fulfil a purpose beyond their design. In the opinion of Alan Turing, an innovator in the field of Artificial Intelligence during the forties and fifties, a machine in its most refined form can be made as a Turing Machine; a device which can be fed instructions in its own algorithmic language to carry out a series of steps and achieve an overall aim. Taken in this definition, as a closed and finite series of steps, we shall see that a machine can achieve the status or make the discoveries comparable to that of human thought.

Thought and intelligence are things of which we are all aware and must accept (think about the consequences of denying this). We have the assertion of Descartes cogito'; "I think therefore I am. Or more usefully; I am therefore I think. However, this personal confirmation of thought does little to give a qualitative definition. Thought is creative, but not only that, it involves a self-awareness and introspection that allow it to be meaningful. To have many monkeys working away on typewriters eventually one of them will produce Shakespeare's sonnets, but this is not how Shakespeare himself wrote them. A thinking being creates and determines its' own purpose [as is central to Existentialism.] In as far as this definition of thought goes it is not contentious but when considering the relation of human thought to the physical world, and the mental activity of the mind in relation to the physical activity of the brain one arrives at the Mind/Body' problem. This problem, since first encroached by Descartes, has been tackled by various philosophers in differing ways, the broadest classification of which give Dualist/Pluralist and Monist theories.

The Dualists believe that mind and body have incompatible natures, and it is the relation between them that differentiates the various opinions under the heading of Dualism, and is usually associated with a concept of God. The philosophy of Descartes provides the first instance of the dualistic nature of mind and body; all that we can be certain of is that we are and God is good and through God we experience the physical world. How this occurs is left unanswered by Descartes but this link is explored by Malebranche in Occasionalism. When we have intention or volition in our minds to do something in the physical world God, aware of our wish, causes it to happen. Our willing something to occur is not the direct cause but rather the

occasional' cause by determinism, Leibnitz imagines dualism in the image of two clocks, the mental and the physical, and it is by God's divine harmony that they chime synchronistically. Finally there is the Epiphenomenalism of Huxley in which the determinism of physical laws is not only the cause of physical phenomena but also of mental phenomena of our consciousness. Clearly from a Dualist's point of view we cannot construct thinking machines because thought is separate from the physical world and takes place in the soul'. However, the conclusions of the Dualists are reached by ideological supposition and do not take into account anything more than pure opinion.

Monists, on the other hand, do not accept any difference between mental and physical, taking several opposing stances to Dualism. The Idealists believe that the physical world is part of the symbolism of language. Philosophers like Berkeley and Hegel assert that everything is essentially mental. This means that physical objects, such as machines (and brains), are subordinate and irrelevant to thought. Russell, by contrast, put forward a theory of Neutral Monism suggesting that there is a more fundamental concept underlying both the physical and mental arenas. He does not say what this might be, and so moves the argument no further forward. Thirdly, there are the Materialists who proffer a direct correlation between physical and mental states. The Behaviourism of Watson and Ryle looks at humans in a physical context and accounts analytically for the psychological notions associated with situations by which mental behaviour can be explained. Thus Ryle identifies what he sees as Descartes Dualistic category mistake in separating mind and body as the dogma of the ghost in the machine , there is, in his opinion, no such thing as the soul. Following on from where the Behaviourists leave off the Materialists; Armstrong and Stuart think that such things as the flow of electrical pulses in the nervous system and the intricate patterns of neuron firings in the brain as the physical context of mental states. We have similar mental states in similar physical situations because of physical laws that relate the two, though there is no contingency for determining specific correspondence. Materialism offers a positive view-point from which one can fruitfully consider the possibility of being able to quantify the laws' that relate physical to mental.

An answer to the question of whether or not machines can think in the context of Materialism would be argued along the following lines:

"We have a complete description of the human being in the form of the gene, and we know that from this single finite chain of amino pairs comes the information, decoded further into proteins by the genetic code, to produce a thinking human being. So all the information which embodies the mechanisms of thought, and the ability to articulate it, is contained in this finite genetic language. This gives us a full qualitative definition of thought, and then it is theoretically possible to design and build machines to emulate this, using suitable language and design we ought in theory be able to build a thinking machine. "

There are, however, serious failings in the answer of Materialism, for though it explains identity' in thought , it does not give any clear reason as to why we are aware of the identity between mental states, or what we might call memory (rather than instinct'). There is no aspect in Materialism of the self-referencing and self-aware nature of human thought, nor any explanation of the genuine creative ability of thought. Indeed, this analytical approach will never break out from the law-like foundations it has created for itself to be able to account for human discovery of the structure of mathematics which has been found by Godel, Chaitin, and Cohen not to be analytical but rather to be an open' system. Most notably Godel showed in his Incompleteness Theorem that a closed' analytical or axiomatic system of mathematics is impossible, because such a system if it is wide enough to contain all mathematical truths will also contain a statement of its own falsity. To avoid this contradiction we must accept open' mathematics. Also this makes it clear that if humans were analytical as asserted by Materialism then it would have been impossible for Godel to conceive of this fact, this gives Human thought the elusive properties of self-referencing and openness. Similarly self-referencing could never fully work in a machine, because even with programmes and algorithms which can update and rewrite themselves there would have to remain unchanging hardware and language that lay down boundaries and confines to the machines abilities, preventing it from ever being able to think. Thus Turing's machine falls short of the Materialists conception that given a finite qualitative definition of human thought this could be reproduced in mechanistic form. Machines (after Turing) can never think.

This still leaves the possibility of Artificial Intelligence, an avenue that ought to be considered in a full appreciation of this question. The chance that not by any deliberate design or intention, but by sheer fluke, man could happen upon a cogitator', (the thinking equivalent of a computer,) which could not only think but would be willing to communicate its' thought with us. It may after all decide it does not desire interaction with humans. But if by chance a cogitator' were to exist it would be something other than a machine. [ In order that we could genuinely believe it to be capable of thought we must have some test with which we can assess our supposed cogitator'. Such a test was devised by Turing when faced with this problem and I quote his own explanation of the Turing Test' answering the doubts of an Oxford don sceptical of Artificial Intelligence: "I am sure Professor Jefferson would be quite willing to adopt the imitation game as a test. The game is frequently played under the name of viva voce' to discover whether someone really understands something or has merely learnt it parrot fashion. In our case the interviewee would be the cogitator' and the test would settle whether it was intelligent or merely pretending' to be so.]

To illustrate how it is the case that despite the existence of a closed' qualitative definition of thought man cannot even theoretically design or build a cogitator' it is useful to look at the case of a form of computer creation already in use, that of Neural Networking. This process involves an evolutional heuristic approach to the design of networks, which are sets of interlinked processors able to interact with one and other. The connections between the processors are tuned' and the output of the unit changes. By tuning the processors to give an output closer to the ideal determined by the intention of the designer. Eventually it will function as required and the operator can then list the settings of all the processors and the connections between them, giving all the information necessary for someone to produce copy of the network that behaves identically to the original. But no longer can the engineer responsible be called its' designer for it has been created though dynamic heuristic feedback from the performance of previous versions of the network, rather than an analytical consideration of the final aim. Were you to ask the neural networker why the network is configured in that particular way, then the only truthful answer she can give is; "Because it works . It is the case that for us to try to create, by applying the theories of Neural Networking, a cogitator' is ridiculous and illogical, because there is no way of applying heuristic feedback to the comparison of the quantitative closeness' to real thought of on configuration of one network to another. There do not exist degrees of thoughtness'.

It would seem then that the only hope for the creation of a cogitator' is through the working of chance, and a very, very slim chance at that . But then it is also the case of evolution of our world and us in it that the probabilities involved were small; so small in fact that the Strong Anthropic principle uses them as proof of God's existence. This leaves little other than faith upon which to base the eventual creation of a cogitator'. There is the escape clause provided by the Weak Anthropic principle; the question "Why are we here given that the odds are so heavily against our existence? is meaningless because if we were not then no-one would be able to ask the second part of the question because it is fundamentally contingent upon the world being the way it is. Unlike in the case of the Strong version of the principle the Weak cannot be applied to the question of cogitators' because it is not contingent upon their existence. We can extricate ourselves from this seeming dead end by turning to another recent advance, the discovery of Chaos.

Chaotic behaviour is characterised by its lack of predictability, the classic example being Weather systems; for although weather patterns stay within certain wide confines, e.g. it will never snow in the Sahara, it is impossible to predict with any detail or accuracy beyond the most short term of scales, (a couple of days.) But within Chaotic behaviour there is the possibility of an Emergent Causality; from a chaotic state can emerge' at a macroscopic level an ordered pattern. An example of this is found in the heating a layer of water sandwiched between two plates of glass, normally there is a messy chaotic arrangement of convection currents and eddies, but suddenly at a specific difference in temperature between the top and bottom a regular pattern of flat hexagonal currents emerges.Another example is of Evolution emerging from ever more disordered state of Entropy. The emergence of unpredictable and incomprehensible degrees of order in a chaotic systems is always due to the existence in phasespace' of Strange Attractors, (an attractor' can be thought of as a geometric basin in which the aspect of its sides forces the behaviour of the system to follow the pattern of the basin, and it is the character of strange attractors that they have fractal' basins.) Investigations of the chaotic nature of the brain have revealed strange attractors, which suggests an explanation of how the qualitative description of thought can be stored in the genetic code of DNA: There exist simple functions which describe complex fractal structures e.g. the Mandelbrot Set; an infinitely complicated structure is described by the iterative function: XA2 + C ---> X ) It can be seen that the cogitator' would similarly have strange attractors determining its thought, so the best hope for guiding future Artificial Intelligence in its naively heuristic course would come from a better understanding of chaos and fractals. However, I believe that this guessing game would take so long to complete with any success that it is best left to God or at least an immortal with huge reserves of patience.

Having reached the end of this investigation, in which I have strived for an affirmative answer to the original question, I have come up against brick wall after brick wall: The failure of classical philosophers to prove that machines can think through their inquiry into the Mind-Body problem, which led to consideration of the impossibility of a machine that had the ability of self-referencing, or that was able to resolve Godel's Incompleteness theorem. When the boundary was crossed from machine to cogitator', which sealed the answer to the question, it was still impossible for anyone less than God to produce one. Or if, by a huge fluke, one were produced it would be, like the evolution of thinking man, through the emergent causality of Nature and not due to the creativity or intelligence of man.

Caspar Addyman, November 1991