Thinking machines

Article metrics

The Cambridge Quintet: A Work of Scientific Speculation

Little, Brown/Addison-Wesley: 1988. Pp.181 £16.99, $23. US publication date, April 1998

Goodbye Descartes: The End of Logic and the Search for a New Cosmology of the Mind

Wiley: 1997. Pp.301. $27.95, £17.99

John L. Casti's The Cambridge Quintet ends with a languid “The thinking machine conundrum is not going to be solved in one evening”. Both this book and Keith Devlin's Goodbye Descartes give the impression that the conundrum continues to defy ‘solution’, but this need not detract from the enjoyment of reading these works, each of which has a novel way of approaching the discussion.

Snow.

Casti uses the device of a fictional dinner party in Cambridge set around the middle of this century. C. P. Snow is the host. His task as a civil servant is to advise the government on the feasibility or otherwise of building a machine that “thinks like a human being”.

Wittgenstein.

The four guests, with much literary licence, represent the main pillars of the debate. Alan Turing is there, issuing the mathematical challenge that universal computation machines, being universal, must be capable of thought. Erwin Schrödinger, as a confident physicist, believes that thought in living objects and machines may be subject to the same laws of physics. The third diner is J. B. S. Haldane who, in his role as a geneticist, argues that the biological and evolutionary make-up of living beings gives them a character that cannot be emulated on computers. The last guest is Ludwig Wittgenstein, who will have none of it. For him the idea of a thinking machine is an absurdity, an oxymoron. Casti's conversational exposure of these extreme views, which are widely held today, suggests that they may be simply irreconcilable.

With greater hope, Devlin holds that ‘thought’ has a character that transcends mathematics, particularly logic. This is not in the same spirit as those who believe that science is not yet sufficiently developed for the task, as Roger Penrose argued in Shadows of the Mind (Oxford University Press, 1994); rather, Devlin argues that mathematics is too advanced, too precise, to match the fluidity of human thought. This gives him the opportunity of writing with great clarity on logic and its useful but limited role in computation. Devlin concludes that a ‘soft’ form of mathematics might do the trick. Unfortunately we never find out much about it: as Devlin himself admits, he is only hoping this might be so.

Haldane.

Despite the qualities of these books, they may be too ready to accept the lack of progress on the ‘thinking machine’ question posed by Turing more than 60 years ago. Are the hurdles better understood now? The first barrier to progress comes from the word ‘machine’ in the context of ‘thinking’. It is generally assumed that such a machine would be a conventional computer which, when given an appropriate program, would behave in a way akin to a thinking human. Unfortunately much of the argument comes to a halt with the functional emptiness of this model: the computer itself does nothing; some programmer must work out every step of what the machine is to do.

Schrödinger.

By contrast, the brain is clearly a functional machine of a kind, as it does not rely on anything like a program. It is beautifully structured to be driven by both evolved structure and learned function, neither of which may be available to a programmer. Understanding how this mechanism contributes to the experience of sensation and thought is top of the current agenda.

Turing.

It helps that evolution and learning can be modelled and analysed in an appropriate artificial domain: that of neural systems. Parallelism, adaptation and modularity are needed to make real machines work in emergent, seemingly intelligent ways. Models need these self-organizing features. That the modelling is often done on a conventional computer is a red herring; the appropriateness of the model is important, not its box.

Another hurdle is the assumption that thinking machines will work by performing endless manipulations of simple symbols at ever-increasing speeds. The symbols of a neural machine, whether real or artificial, are neural firing patterns which are wonderfully extravagant, rich, diverse and sufficiently expressive to be capable of directly supporting subtle sensations that could distinguish between the thought of a good Burgundy and plonk, or between loving and liking. The notion of the brain as a manipulator of parsimonious symbols or, indeed, as some kind of data-processing machine is a customary but sterile starting point for a science of the mechanics of thought.

Although both Casti and Devlin hint at the possibility of evolving and adaptive machines, they largely avoid the topic. Perhaps they do not wish to offend the many who, over the past half century, have believed in the supremacy of an empty computing box that begs for someone to tell it what thinking is.

Author information

Rights and permissions

Reprints and Permissions

About this article

Comments

By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.