Logic and artificial intelligence 


Last week I finished a couple of essays on the apparently unrelated subjects of Bertrand Russell's Theory of Definite Descriptions, regarding the relationship between formal logic and language, and John Searle's Chinese Room Experiment, about whether machines could ever exhibit true understanding or intelligence in the sense that human beings presumably do. Oddly enough, I found making almost almost opposite arguments in each. The first was that purely formal logic is insufficient to capture the richness and meaning of natural language, and the second that a formal system, such as an appropriately programmed computer, could in theory possess the sort of cognitive states that characterise complex self-aware organisms, such as ourselves. The notion of levels of abstraction also crept into both essays, indicating that these subjects may not be as unrelated as I first thought. 

One interest nugget of information that I came across while researching the Searle essay was an article in the Canadian Journal of Philosophy about how computers don't actually follow instructions, which came as something a surprise to me given that I've spent most of the last 20 years programming the damn things on just this assumption. However, after giving the matter some more thought, I realised that the author was right, and that computer programs are merely a way of describing what a machine does, or to put it another way, specifications for a machine, rather than a list of instructions that the machine literally follows in the way that you or I might follow a list of directions.

That computers behave as if they are following a list of instructions is a feature of their construction, but when you submit a program to a computer, you are in fact creating a machine that is configured to perform a particular set of tasks. The actual tasks take place by virtue of electrons whizzing through tiny various wafers of specially prepared metals in a way that can conveniently be described as the machine executing a particular series of instructions, but in reality nothing of the sort actually takes place (unfortunately, even the physical description is just another level of abstraction, as according to quantum mechanics, there's no such thing as ‘an electron’, but you get the basic idea). If the computer was made of cogs and levers, or ping-pong balls and pieces of string, then it would be obvious that any apparent instruction-following behaviour is merely an abstraction of what the machine is actually doing. But somehow, the fact that it all happens by way of invisible forces moving through solid state circuitry makes it all the more plausible that there really is some kind of central processing daemon that ‘interprets’ each instruction and ‘carries out’ what is asked of it. The fact that every introductory textbook or computer programming that you can find says that this is indeed exactly what happens is also somewhat misleading.

This kind of anthropomorphising (my new word for the day) is typical of our way of thinking and interpreting behaviour. Indeed, it forms the foundation of the Turing test, which attempts to replace the question of whether computers could really think with the more manageable one of whether such a machine could fool a normal human being into thinking that they were conversing with another person, when they are in fact talking to a machine. Perhaps in the end there isn't much difference between the two, and I'm definitely finding that the more I go into these questions, the more I begin to question what is really meant by ‘thought’, ‘understanding’, and so on, and in what sense we ourselves posses these characteristics. Perhaps these too are merely convenient labels for particular forms of behaviour or functions of the body that capture a particular level of abstraction (that phrase again) within a complex organic system, but that can in turn be reduced to more fundamental descriptions concerning electrochemical activity in the central nervous system, of which the brain forms just a part. (In evolutionary terms, the brain is actually an outgrowth of the spinal column, and many early vertebrates got along just fine without one!)

Of course, these are all big questions, but so far I have not come across any evidence to suggest that purely man-made artefacts, even formally specified ones, such as digital computers and their programs, could not exhibit exactly the same sorts of behaviour and cognitive states that we humans enjoy. In the end, I think that Searle's Chinese room argument misses the point, and that by taking an oversimplified view of what a computer is and the types of programs it can embody, Searle throws out the baby with the bath water and fails to recognise that even machines could one day reach the level of sophistication required to exhibit complex intentional states, such thinking, knowing and understanding. Whether this is a good thing or a bad thing largely depends on our own level of mental development, and whether we are wise enough to know what to do with the new forms of intelligence that we ourselves have created. God knows that we haven't yet learned to look after each other and the other forms of life with which we share the planet, so will introducing yet more forms of artificial life really help? 

Posted on Thursday - July 07, 2005 at 10:00 PM            

© 1995-2006