If you want to talk about artificial intelligence, and you want to get a name for yourself, then make sure you paint it black. A recent BBC article epitomizes this trend – the doomsayers speak louder and more stridently than the more cautious, muted voices. The article asks if we should fear AI? The consensus opinion among the ‘experts’ quoted is yes. To quote one representative voice:
‘Elon Must, founder of Tesla motors and aerospace manufacturer Space X, has become the figurehead of the movement, with Stephen Hawking and Steve Wozniak as honorary members. Mr. Must who has recently offered 10 million pounds to projects designed to control AI, has likened the technology to “summoning the demon” and claimed that humans would become nothing more than pets for the super-intelligent computers that we helped to create’.
His apprehension is nothing new. Over thirty years ago, Marvin Minsky of the Massachusetts Institute of Technology claimed that the next generation of robots will be so intelligent that we will be lucky if they decide to keep us as house hold pets. The quote is from John Searle’s book ‘Minds, Brains and Science’, which we will discuss shortly. Must, I think, is just as mistaken as Minsky was. But maybe this time it’s different?
I don’t think it is different this time and the reason I think that is re-reading John Searle’s collection of essays from 1984, ‘Minds, Brains and Science’, which were also delivered as the BBC’s Reith Lectures in the same year. The second lecture, ‘Can Computers Think?’ could well have been written to answer the sort of nonsense that the doom peddlers are espousing today. He wrote at a time when many assumed that it is ‘only a matter of time before computer scientists design the sort of hard and soft ware that are the equivalent of human brains.’ Pundits then, as they do now, assume that the human brain is a kind of digital supercomputer, and that developments in IT and AI are analogous to closer and closer approximations to it, and will eventually surpass it.
But we are not comparing like with like. Searle writes:
‘The reason that no computer programme can ever be a mind is simply that a computer programme is only syntactical, and minds are more than syntactical. Minds are semantical … they have a content.’
To illustrate this, he imagines a machine is programmed to simulate the understanding of Chinese. You feed in questions in Chinese and it gives answers in Chinese. It’s programmed so well that it looks like that it actually understands and speaks Chinese. But this impression misleads. Searle asks you to imagine that you are locked in a room with a basket of Chinese symbols. Someone outside the room feeds in Chinese symbols and in return you hand them symbols, according to a rule book, but you have no idea what the symbols mean. You are just following the rule book. You are in fact answering questions in Chinese but you neither speak nor understand Chinese. And neither does the computer. All you have, writes Searle, is ‘a formal programme for manipulating uninterpreted Chinese symbols’.
A computer has a syntax but no semantics – all form but no content:
‘Understanding a language, or indeed, having mental states at all, involves more than just having a bunch of formal symbols. It involves having an interpretation, or a meaning attached to those symbols.’
Let’s think of another example of our own to develop this insight further. Some have suggested that AI will even displace lawyers. Now, we know that in law, the words do not speak for themselves. Take the 2nd Amendment of the United States constitution:
‘All well-regulated militia, being necessary to the security of a free state, the right of the people to keep and bear arms, shall not be infringed.’
In the United States, that phrase has split not only ink but blood. What is the argument about? Surely not over the formal, dictionary definition of each word in that phrase. It’s about the ideas that each word expresses, either standing alone or in combination with others. Different human minds scanning those words do not ‘see’ the same ideas in phrases like a ‘free state’. They mean different things to different people.
Now, imagine trying to settle this issue by referring the matter to two robot lawyers who try to argue the case in the presence of a robot judge. The robot concerned could simulate the formal structure of a legal argument by ‘knowing’ the formal meanings of words but underlying these words are mental concepts that no robot could ever know. How can you programme a robot lawyer to know what a ‘free state’ looks like? And how you programme a robot to ‘know’ that a law for gun control (or its absence) does or does not ‘violate’ the 2nd amendment? We are back to Searle’s distinction between syntax and semantics. Words like ‘this law violates the 2nd amendment’ are expressions of semantics, not syntax.
It is not only that the law does not speak for itself. It has to be interpreted. We also have to decide when it applies in certain cases. Armed robbery and shoplifting are both forms of theft. But do we apply the same rules in both cases? Of course we don’t. But why don’t we? Because we have different ideas about what constitutes a ‘just’ punishment each case. Try programming a robot judge to ‘let the punishment fit the crime’. Again, this is a phrase that is not reducible to its component parts. You cannot explain it merely by attaching a dictionary definition to each word in the phrase. A robot could certainly be programmed to utter such words but it would have no more content than a parrot’s mimicry. It could simulate words but it cannot duplicate the mental states words generate. Semantics again!
Much of the confusion around this issue rests on ignorance between the concepts of syntax and semantics. But it also overlooks something else. Information and Communications Technology is often conflated with Artificial intelligence. Being able to look up information quickly, as per Google, is not the same thing as AI, with robots or machines able to gather and process their own information independently of any human intervention or influence – like the way Skynet can do in the Terminator movies or like the malevolent Hal in 2001. Improvements in calculation and processing power speed does not necessarily lead to improvements in intelligence and sentience. Cockroaches are superior to any robot in terms of their ability to learn from and adapt to their environment.
Good public speakers – or rabble-rousers – know that it is not the words they use that inspire other minds to act. It is the ideas that words generate that do that. Ideas move people. Robots do not have ideas because they do not have mental states. Therefore, they are not motivated to do anything. Indeed, as a great fan of the Terminator movies, I have often asked the question that the plot never answers: why do the machines bother to fight the humans? Why do they care so much? But we don’t need to ask these questions about the humans’ motivations for fighting the machines. The answers are obvious.
If you have got this far, then you may be wondering what this has to do with moral conflict and politics. Hopefully, the answer has already become clear, with the discussion about the 2nd amendment of the US Constitution a clue: moral and political conflict is generated not by arguments over syntax but over semantics. The resolution of such conflicts depends on the political use of semantics to smooth out these conflicts, sometimes by using weasel words and half-truths. A robot would have no idea about the complex mental concepts involved when two warring parties down weapons and ‘agree to disagree.’