In these last decades, a multitude of methods have been developed that create Artificial Intelligence for different types of problems and applications. One of those methods is the so – called expert systems.
Expert systems or knowledge-based systems are computer applications that solve problems with knowledge of human experts in a specific domain, that is, that expert systems emulate the ability to make decisions of a human expert.
These applications have two basic components: the reasoning system and a knowledge base formed by facts and rules. The reasoning system applies rules to facts to deduce new facts that are incorporated into the knowledge base. The most basic way to implement this system of reasoning is with conditional rules of the “If X then Y” style, and deduce new facts through inference.
A basic expert system consists of a knowledge base with facts and rules and an inference engine to obtain new facts.
Similar systems are already being used in hospitals to help the doctor diagnose and propose medical treatments. For example, England’s public health system uses a system of this style to make a first rapid diagnosis and thus help the doctor analyze a new patient with a list of possible diagnoses deduced by an expert system.
One type of expert system widely used today is that based on a Bayesian network. In essence, it consists of a graph that represents a set of known variables and the dependency relationships between them in order to infer, that is, estimate the probability, of the unknown variables. Given its characteristics, this model is ideal for classification, prediction or diagnosis.
Imagine that we have two variables so that you can determine that the grass in a garden is wet: that the sprinkler is activated or that it is raining (assuming that if it rains, the sprinkler goes out). Once we have these variables, we can create a Bayesian model in which the three variables have two possible values (true T, F false). The three variables are: G = Wet grass, S = Sprinkler activated, and R = Raining.
The model can answer questions like “What is the probability that it is raining since the grass is wet?”
However, expert systems also have their own inherent limitations:
- Common sense: For an Expert System there is nothing obvious. For example, an expert system on medicine could admit that a man has been pregnant for 40 months, unless it is specified that this is not possible since a man cannot have children.
- Natural language: With a human expert we can have an informal conversation while with an SE we cannot.
- Learning ability: Anyone learns with relative ease from their mistakes and from outside mistakes, that an SE does this is very complicated.
- Global perspective: A human expert is able to distinguish which are the relevant issues of a problem and separate them from secondary issues.
- Sensory capacity: An SE is meaningless.
- Flexibility: A human is extremely flexible when it comes to accepting data to solve a problem.
- Unstructured knowledge: An SE is not able to handle unstructured knowledge.
- An expert system does not have feelings or can understand certain emotions and human concepts such as marriage, morality, love or planning the future.