Current Approaches

What is meant by current approach is that, in what form is AI implemented currently. Although people tend to imagine AI to be something as the sort of the giant talking computer character HAL in Sir.Arthur.C.Clarke's "The Space Odyssey", AI is implemented in a totally different manner. It is, in most cases, implemented in the existing models of computers by performing just a few modifications to the hardware, if required. Certain types of AI exist just conceptually. Implementation has not begun yet for such types. Lets look at the three current major approaches:

Expert Systems

Neural Networks

This is a basically a type of computer but quite different from the conventional computers with the Von-Newmann architecture. A neural network is based upon the structure of the human brain, which consist of many billion cells called "neurons". Each "neuron" in our brain uses chemical energy for its operations and is equivalent to a single processing computer. The neural network is very similar using electrical energy from one point to another. The most important feature of neural networks is that they could learn after some initial training. There are various training methods and learning methods associated with it. Neural networks are unimaginably expensive to build causing it development to be very slow.

Fuzzy Logic

Fuzzy logic is a superset of the conventional Boolean logic that has been expanded to handle the concept of partial truth - truth values between "completely true" and "completely false". Fuzzy theory is now being regarded as a methodology known as "fuzzification", rather than a single theory, resulting in the generation of continuous (fuzzy) theories from crisp (discrete) theories. Some products of "fuzzification" are the "fuzzy calculus", "fuzzy differential equations" and so on. Fuzzy logic is implemented directly in a very little number of applications like the "Sony Palmtop Kanji Character Recognition Algorithm". But it has been successfully implemented along with expert systems and neural networks. An interesting case of fuzzy logic in action is where it was used along with neural networks to control a chopper with a missing rotor blade. An interesting point in this case is that although a pilot cannot by any means control such a chopper, he is required to train the program initially!