Building intelligence

Artificial intelligence - the ability of machines to emulate human thought or behaviour - is created using two distinct elements: knowledge and reasoning. Considerable computing space and speed is needed to impart knowledge to your intelligent system, so you d better set aside some cash for that.

You ll also need to program your computer to have the reasoning abilities to make use of the library of knowledge you ll give it and to make guesses about new situations, based on that past knowledge. So be sure to book off more than a few weekends for some heavy-duty programming.

Language of A.I.

What s involved in that programming, you might ask? In order to perform human-level tasks and understand the world around them, computers must be programmed with mathematical algorithms - sets of rules used for problem solving - and advanced logic that allows machines to think outside the binary world of yes or no.

These rules allow intelligent machines to make logical decisions, deal with new situations and perceive the world around them. You ll need to load up your system with such rules - and occasionally bend them, even break them when appropriate. It s that "when appropriate" part that s so difficult and which enables a machine to be able to think and learn more like a human.

In modern computer science, there are basically two ways to provide a computer with the ingredients necessary for A.I.

Artificial brains
One way to achieve A.I. is called the bottom-up approach. This involves building intelligence and then harnessing that intelligence to behave appropriately for the task or environment at hand. Such a learning-from-scratch approach doesn t try to develop human behaviour and maintains that all that is required for A.I. is simply being able to interact effectively with the environment.

An example of one way to do this is to make an electronic replica of the brain s complex network of neurons. These artificial "minds" - called neural nets - are like small sections of brain that are able to relay information between computers in the network. Early computer scientists first got this idea after observing how electrical signals passed between neurons in the human brain.

Once you ve got the brains - in theory - you ve just got to load them up with information and the ability to use it. This teaching process is the same as what happens to humans as they grow up. But for a machine, that process is easier said than done.

One piece at a time
The top-down approach to building A.I. involves attempting to mimic the brain s behavior with computer programs - essentially setting your computer up with an A.I. apprenticeship. This sort of A.I. is most-used today in "expert systems" - programs that try to simulate one element of human thought in greater detail than is possible with more general A.I.

You can already find these task-specific systems at work in medicine, industry and communications. They use existing A.I. capabilities to take over very narrow areas of tasks usually performed by humans.

Building an expert system wouldn t let you recreate all the thought processes needed to have an intelligent conversation, but it could help you kick some serious posterior in a game of chess.

Many experts agree that true artificial intelligence will come from combining top-down and bottom-up approaches, like using neural nets as expert systems.

Filling-in the blanks

All that s still just for starters, though. Knowledge, reasoning and logic are the beginning of A.I., not the end. A vast amount of the human brain still has to be mapped and the science of how to recreate it in machines is in its infancy.

So don t fret if you can t build your own artificial mind even after reading this handy guide. The experts have been at it for years. Knowing how it all works doesn t hurt though, just in case you think you ve got the million-dollar answer for how to make a truly complete thinking machine.