Knowledge-Based Systems

Also known as artificial intelligence, knowledge-based systems are so named because they accumulate knowledge and, with it, the ability to make decisions or predict the future based on knowledge of historical data.

Two well-known types of knowledge-based systems are expert systems and neural networks. We explain each in the following sections.

Expert systems

Expert systems build a database of past events in order to predict outcomes in future situations. An inference engine analyzes the past events to see whether it can find a match between a past event and the current problem. For instance, if a stock-picking program knows that IBM always goes up two points when the Mets are in town under a full moon, then it tells you to buy IBM when the Mets are in town and the moon is full.

Expert systems are designed to work with degrees of uncertainty, and they do so in one of two ways:

Fuzzy logic: Breaks down the factors influencing a decision or outcome into its components, evaluates each individual component, and then recombines the individual evaluations in order to arrive at the yes/no or true/false conclusion for the big question or problem.

Certainty factors: Operate on the numeric probability of yes/no, true/false, rain/snow, or whatever the expert system is working on. The individual probabilities are aggregated, and the final conclusion is reached. For example: Tomorrow it will snow in Buffalo.

Fuzzy logic is the component of an expert system that produces a quantitative result based on uncertainties.

Neural networks

Neural networks mimic the biological function of the brain. If the sum of a set of inputs exceeds a threshold, the neuron fires, or discharges. A neural network accumulates knowledge by observing events; it measures their inputs and outcome. Over time, the neural network becomes proficient at correctly predicting an outcome because it has observed several repetitions of the circumstances and is also told the outcome each time. Then, when confronted with a fresh set of inputs for a new situation, the neural network predicts outcomes with increasing reliability over time.

Neural networks learn that input components are weighted, which is to say that their specific degree of influence on the outcome is calculated.