Read more: “Instant Expert: Artificial intelligence“
We have long suspected that intelligence is not exclusively a human quality, and that it is possible to build machines that can reason and learn as well as a human can. But what seemed straightforward at first has turned out to be surprisingly difficult
What is AI?
The field of artificial intelligence (AI) is the science and engineering of machines that act intelligently. That raises a vexing question: what is “intelligent”? In many ways, “unintelligent” machines are already far smarter than we are. But we don’t call a program smart for multiplying massive numbers or keeping track of thousands of bank balances; we just say it is correct. We reserve the word intelligent for uniquely human abilities, such as recognising a familiar face, negotiating rush-hour traffic, or mastering a musical instrument.
Advertisement
Why is it so difficult to program a machine to do these things? Traditionally, a programmer will start off knowing what task they want a computer to do. The knack in AI is getting a computer to do the right thing when you don’t know what that might be.
In the real world, uncertainty takes many forms. It could be an opponent trying to prevent you from reaching your goal, say. It could be that the repercussions of one decision do not become apparent until later – you might swerve your car to avoid a collision without knowing if it is safe to do so – or that new information becomes available during a task. An intelligent program must be capable of handling all this input and more.
To approximate human intelligence, a system must not only model a task, but also model the world in which that task is undertaken. It must sense its environment and then act on it, modifying and adjusting its own actions accordingly. Only when a machine can make the right decision in uncertain circumstances can it be said to be intelligent.
Philosophical origins
The roots of artificial intelligence predate the first computers by centuries. Aristotle described a method of formal, mechanical logic called a syllogism that allows us to draw conclusions from premises. One of his rules sanctioned the following argument: Some swans are white. All swans are birds. Therefore, some birds are white. That form of argument – Some S are W; All S are B; Therefore some B are W – can be applied to any S, W, and B to arrive at a valid conclusion, regardless of the meaning of the words that make up the sentence. According to this formulation, it is possible to build a mechanism that can act intelligently despite lacking an entire catalogue of human understanding.
Aristotle’s proposal set the stage for extensive enquiry into the nature of machine intelligence. It wasn’t until the mid-20th century, though, that computers finally became sophisticated enough to test these ideas. In 1948, Grey Walter, a researcher at the University of Bristol, UK, built a set of autonomous mechanical “turtles” that could move, react to light, and learn. One of these, called Elsie, reacted to her environment for example by decreasing her sensitivity to light as her battery drained. This complex behaviour made her unpredictable, which Walter compared to the behaviour of animals.
In 1950, Alan Turing suggested that if a computer could carry on a conversation with a person, then we should, by “polite convention”, agree that the computer “thinks”.
But it wasn’t until 1956 that the term artificial intelligence was coined. At a summer workshop held at Dartmouth College in Hanover, New Hampshire, the founders of the nascent field laid out their vision: “Every aspect of learning or any other feature of intelligence can in principle be so precisely described that a machine can be made to simulate it.”
The expectations had been set for a century of rapid progress. Human-level machine intelligence seemed inevitable.
A field of fragments
In the 1960s, most leading artificial intelligence researchers were confident that they would meet their goal within a few decades. After all, aeronautic engineering had gone from the first jet aircraft to an astronaut on the moon in 30 years. Why couldn’t AI take off in a similar way?
The difference is that there are no simple formulas for intelligence; the discipline lacks its own F = ma or E = mc2. By the 1980s, AI researchers realised that they had neither sufficient hardware nor knowledge to simulate everything a human can do and the field fragmented. Instead of working towards a single human-equivalent computer intelligence, research groups splintered off to investigate specific aspects of the larger problem: speech recognition, for example, computer vision, probabilistic inference – even chess (see timeline).
Each of these subdisciplines saw significant successes. In 1997, IBM’s Deep Blue computer beat the world chess champion, Garry Kasparov. Deep Blue could evaluate 200 million chess positions per second in its search for the right move. This allowed it to quickly look ahead at many different sequences to see where they might lead.
Deep Blue scored an impressive victory in a game that demands intellectual rigour. However, the machine had a very narrow range of expertise. It could win a game of chess, but it could not discuss the strategy it had employed, nor could it play any other game. No one would mistake its intelligence for human.
