Copyright © by Mark Baker 1996
Return to MarkChrisSoft home page
But what is intelligence? Is it the ability to reason (argue)? Is it the ability to acquire and use knowledge? Is it the ability to perceive (see, understand) and manipulate things in the real world? Intelligence is a combination of many information representation and information processing skills.
One aim of trying to develop computers with AI is to make computers more useful. Another aim is to try and understand more about what intelligence is.
There are many different approaches to artificial intelligence, each one attempting to model a small part of what we think of as intelligence. Programs that use AI tend to focus on one narrow, specific task or area, but can still be very successful. However, AI is a relatively new area of study and there is a long way to go.
Approaches include neural networks, exploring alternatives and knowledge based systems. Neural networks attempt to model the way that the neurons of the brain work. Links are set up between "neurons" and the more that the links are used, the stronger they become. Links that are not used, become weaker. This enables learning to take place, as certain patterns are reinforced through repetition. Neural net programs need initial training, to get them going.
Many game playing programs, such as chess programs, explore many different possible combinations of moves and eventually choose one move. They may, for example, give each possible move a score and then choose the highest scoring option.
In 1996, chess champion Garry Kasparov beat Deep Blue, IBM's parallel computing system, 4-2 in the Supercomputing '96 chess challenge. The computer was based in New York and used the Internet to transmit its moves to where the chess match was being held. Clearly the computer was able to compete with an international chess champion. Deep Blue could search 50 to 100 billion positions in the three minutes which each player had to make a move. This is an incredible amount of processing power, which Kasparov could not begin to match. Kasparov was able to match the computer at playing chess however, because human players use a very complex method of selecting the best move, which includes intuition, experience and pattern recognition. Deep Blue on the other hand, used a much simpler search and evaluation algorithm, but combined it with enormous processing power. These two turned out to be a good match for each other.
Deep Blue's development team hoped to apply some of the lessons they learnt in other fields, such as airport traffic control and molecular simulations for the drugs industry.
Expert systems are computer programs that have been written to help out in this kind of situation. The person who creates the expert system tries to take the knowledge of one or more experts and build it into a single computer program. This program can then be used by someone who is less skilled, to help them to complete a task that they could not do as well by themselves. If the system included knowledge from many experts, it should end up knowing more than any one expert. Using an expert system could be even better than using a single expert human. However, an expert system can only deal with a very narrow subject area.
Knowledge-Based Systems (also called Rule-Based Systems) can be used to create expert systems. The rules are often in the form IF...THEN..., for example:
IF the animal is a bird it does not fly it swims it is black and white THEN it is a penguin
An expert system will start by asking questions and then it puts the information gained into the rules, in order to make conclusions. It will tend to ask different questions depending on the information that is fed into it, so it does not behave like a simple questionnaire. It keeps on asking questions until it has collected all the information it needs, then it will suggest what should be done, explaining why it has reached that particular conclusion.
There are several expert systems written for doctors to use. For example, if a patient has a bacterial infection then the doctor usually needs to start treatment before they are sure which bacteria is causing the illness. This is because the most accurate lab tests take too long to produce results. If the doctor has had special training, they can often use the limited evidence available to make an accurate decision about what is needed and treat the patient with the best available drugs. However, if they have not had this training, they must use broad spectrum drugs, to cover all possibilities. These drugs are less effective.
MYCIN is a medical system that can help to deal with bacterial infections. It allows doctors without specialist training to make a better diagnosis. It starts by asking questions about the patient's symptoms, general condition, medical history and the results of the quick lab tests. MYCIN would also want to know the gender of the patient, their age, when symptoms first appeared, etc.
This expert system contains about 500 rules and 100 causes of bacterial infection and it will eventually reach a conclusion, depending on the answers it has been given. It can suggest which bacteria are likely to be the cause of the problem and give a course of drugs, explaining what each one is for.
XCON is an expert system that is used to configure (set up) VAX computers, which are made by DEC. These are large computers which are bought in modules (processor, memory, disk drives, terminals, etc) which must then be put together (configured). There are many different possible combinations of components and configuring the computers is a very skilled task. In 1984 XCON had 2500 rules and was growing. It could routinely handle computer setups involving 100-200 modules.
Oil and mineral exploration
Weather forcasting
Stock market predictions
Fault analysis
Some complex control systems
Author: Mark Baker, e-mail mbaker@rmplc.co.uk