Imitations

I count myself among the scores of people who have come to admire Alan Turing as a captivating figure of the twentieth century and as something of a martyr for the secular cause. However, I dislike Turing’s ‘imitation principle’. The source of my discomfort is that I can’t help but associate Turing’s principle with the (overwhelmingly flawed) design arguments for the existence of God.

Turing insisted that if a machine is capable of reproducing intelligent behaviour such that we cannot distinguish the answers it gives from the answers of a human, then it is unreasonable not to describe that machine as ‘intelligent’.

Turing seems to get at this idea on account of the following line of thought. He asks: how is it that you know that other humans are conscious and intelligent? Turing claims that we infer this belief by studying the actions and words of other humans. We notice that people say and do the same things as ourselves. Thus it is reasonable to think that other people also think and experience the world the same way as ourselves. This, Turing might say, is the reason we believe other people are conscious. Extending this idea to machines we might conclude that if a machine says the same things as a human, then it must experience the world in a human-like way and ought to be considered intelligent.

But this is not the whole story! We certainly do observe our peers and note the essential similarity of their behaviour to our own. However, we have the additional knowledge that other humans also share physical similarities with ourselves: the physical structure of their brains is essentially the same to that of our own brains. This empirical fact is essential: without it we would not be justified in interring that our own subjective consciousness is also experienced by others. In order to draw a strong analogy between the internal life of a machine and human intelligence it is thus not sufficient to observe behavioural similarities. One must additionally demonstrate that the internal functioning of the machine has essential similarities with the functioning of the brain.

Suppose a machine whose operations are demonstrably different from those of the human brain was to successfully imitate intelligent human behaviour. To my mind, it does not seem reasonable to call such a machine ‘intelligent’ — in the sense that we employ this word when describing ourselves. It seems far more reasonable to say that such a machine is ‘imitating intelligent behaviour’.

In rhetoric, it is easy to argue by analogy whilst brushing over details that apply in one case but not the other. We might call this the fallacy of a false analogy and it is my objection to the imitation principle. This pattern of deceit is also found in intelligent design arguments. The parallel with Turing’s line of thought is hard to miss.

Before we proceed to intelligent design, it behoves us to mention that, in his writings, Turing only ever went so far as suggesting that we should use the word ‘intelligent’ to describe clever machines. He never, to my knowledge, made assertions beyond this suggestion. Sadly, the apologist for intelligent design does not wish merely to suggest that we use the word ‘designed’ to describe the universe.

In an intelligent design argument the writer might began with a man-made object like a book (which contains information) or a watch (whose precise movements serve a coordinated whole). We all know that books have authors and watches have watchmakers. But, like Turing, they ask the question: how do we know that? The apologist claims that we infer the existence of an intelligent author from the nonrandom information contained in the book. We infer, they say, the existence of an intelligent watchmaker from the fact that all the parts work together for one purpose.

But this is manifestly not how I have arrived at my belief in authors and watchmakers! I know that humans exist. I have observed people writing. I have seen people write things that have subsequently been printed. I have met people who have claimed to have written books. And so on. It is for these reasons — and not the book’s ordered contents — that I have empirical grounds to inductively infer the existence of an author when I see a book. And it is this type of direct empirical evidence that also means I am justified in believing in a watchmaker (or, these days, a watch manufacturing corporation, a team of watch engineers, and a whole slew of managers, accountants, lawyers and marketers). Books do contain ordered letters and watches do contain working parts: but these facts are incidental, and they are not the grounds for our empirical beliefs about books and watches.

Our apologist, however, does not recognise this error and will be quick to push false analogies with the natural world. Regard the DNA molecule, they say. (Or the eyeball. Or the laws of physics.) The DNA contained in living cells contains nonrandom, ordered information — just as books do. So, as for books, we should, they say, infer the existence of an author for this information.

If they are particularly bold, they may even plead that this ‘author of DNA’ must, in fact, be God himself. By this stage we are several fallacies deep. Even so, let’s drive home the point: I’ve seen people writing books, I’ve not seen gods writing DNA. Nor do I have any reason to suspect that gods were engaged in making DNA. In fact, all the evidence suggests no gods were involved. This is because one can say a great deal about the precise processes through which DNA is replicated and mutated. Moreover, we have every reason to believe that DNA has a non-intelligent origin on account of the common origin of the species. For instance, consider the very fact that all living organisms share the DNA machinery in all of its idiosyncrasies (right down to the chirality of the damn amino acids!). This is almost inconceivable were it not for the common ancestry of all organisms. In this way, the empirical details of biology point strongly to a biological narrative, just as empirical details about books point strongly to their human authorship.

On these grounds, then, we have found no reasonable grounds to presume that the universe has a ‘universe designer’ — whatever that means. It is laughable that some have the temerity to claim special knowledge about the nature of this designer. Even so, these fallacious arguments seem doggedly persistent. Why do people repeat them? Likely it is because they are invested in the conclusions. Apologists desperately want justification for their religions, while Turing desperately wanted to witness the birth of true artificial intelligence.

I can empathise with Turing. So let me close by returning to his imitation principle and suggesting an adapted version in light of the above criticism. In order to say that a machine possesses “human-like intelligence” or “consciousness” let us require the following two conditions:

  1. The machine must be behaviourally similar to humans: its responses to prompts are indistinguishable from human responses &c.
  2. The machine must be structurally similar to humans: the machine’s physiology should bear a certain resemblance (definition needed) to human brain systems.

The second condition is what needs further development. Perhaps we could find a macroscopic statistical quantity that characterises the ‘informational sophistication’ of a physical system. This might similar to the way that physicists use entropy to characterise how ‘jumbled-up’ a system is. I am thinking here of the spirit one finds throughout Max Tegmark’s paper on consciousness. A successful attempt at formulating a ‘strong imitation principle’ of this kind might suggest new strategies for building truly intelligent systems that go beyond the present design paradigm of imitating human behaviour.