Have you ever had someone yell "Think fast!" and then toss something at you? It’s a dirty trick, but the amazing part is that you can usually manage to dodge, block, or catch the item, as you see fit. Not only is your brain fast, it can make judgements "on the fly," and adapt to changing situations.

In short, the human mind is one tough act to follow.

Traditionally, efforts to mimic human thought have centered around rule-based logic. The computer in front of you now uses rule-based logic: binary data is stored and manipulated according to a set of pre-programmed rules. Most robotic "brains" are rule-based, often contained on a single chip that functions as a computer–a microcomputer.

Rule-based systems can be used to create artificial intelligence, by programming vast amounts of information into a computer. Relying on this enormous set of data, such a computer is able to mimic intelligence, for example, helping to diagnose diseases by comparing symptoms to those in a database. Such "expert systems" can know more, fact-wise, than any single person, and yet they have only a very narrow range of useful function. Also, they can’t learn. They can only make connections they’ve been programmed to make.

Another approach to artificial intelligence is neural networks. Neural networks are modeled after the human brain, with the advantage that they are better at handling ambiguity than rule-based systems. A neural net "learns" by exposure to lots of inputs and corresponding outputs. Once trained, the neural net responds to an input with a likely output.

Unlike rule-based systems, a neural network doesn’t give definite answers, only most probable answers. (Some call this "fuzzy logic.") Sounds wishy-washy, perhaps, but many real problems–for example, "Will it rain today?"–don’t have definite answers.

A third and relatively new approach to robotic intelligence is something called a stimulus-response mechanism (also known as subsumption architecture), pioneered by Rodney Brooks at MIT. In a stimulus-response robot, there is no memory and no logical decision-making, only hard-wired responses to stimulation. For example, by linking light sensors directly to motors, it’s possible to make a light-seeking robot. Several stimulus-response mechanisms operating simultaneously in one robot can create elaborate behavior that seems intelligent.