The story of AI

Older than the machines

The idea of mechanical thinking is older than the machines. Long before computers, philosophers wondered whether reasoning itself could be reduced to steps. Aristotle laid out patterns of argument so regular they read like a recipe: if the first thing is true, and it leads to the second, then the second must be true as well. That is reasoning as a procedure, the seed of the whole idea.

The dream, in other words, came first. People imagined that thought might be broken into rules a machine could one day follow, long before there was any machine to try it on.

From idea to machine

Later, Alan Turing showed that a simple abstract machine, now called a Turing machine, could in principle carry out any computation that can be defined at all. That put a kind of outer boundary on what machines might ever do: if a task could be written down as clear steps, a machine could in theory work through it.

Then real computers arrived, and the hypotheses could finally be tested. For the first time the old questions about mechanical reasoning were not just philosophy. They were something you could build, run and watch fail or succeed.

Waves of hype and AI winters

Since then, excitement about AI has come in waves. There has been an early surge, then a later one, and the wave we are living through now. Each wave followed the same rhythm: early successes stirred up big promises, the promises outran what the technology could deliver, and funding dried up when reality fell short.

Those cold spells, when interest and money drained away and progress seemed to stall, came to be called AI winters. They are the reason many people greet each new burst of excitement with a raised eyebrow, having seen bold claims fade before.

Today we have new methods, enormous amounts of data and a great deal of computing power. Whether this wave breaks like the others, or finally reaches something like general intelligence, no one honestly knows. The pattern of the past is a caution, not a prophecy.

The imitation game

There is one famous attempt to cut through all of this. In 1950 Turing proposed a neat way to sidestep the endless argument over whether a machine can truly "think", and the idea is now known as the Turing test. Put a person and a machine on the far side of a wall, let an interrogator exchange written messages with both, and see whether the interrogator can tell which is which. If they cannot, the machine passes.

The test judges only behaviour, not what goes on inside. It does not ask whether the machine understands or feels anything, only whether it can act indistinguishably from a person in conversation. It is a test aimed at a general intelligence, it has real weaknesses, and yet it captured an idea that stuck.

Check yourself

• The story of AI is described as coming in waves. What rhythm did each wave tend to follow, and what is meant by an "AI winter"?
• The idea of mechanical thinking is older than computers. What did Aristotle's patterns of argument suggest about reasoning, and why is that called the seed of the whole idea?
• The Turing test judges a machine by whether an interrogator can tell it apart from a person through written messages. What does it actually judge, and what kind of intelligence is it aimed at?