Artificial Intelligence Technologies

1. Artificial Intelligence and Commercial Knowledge Base Systems
2. Artificial Intelligence Technology Definitions
3. Implementing a Knowledge Base System
4. Business Benefits of Knowledge Base Systems

Artificial Intelligence and Commercial Knowledge Base Systems.

Knowledge base systems, also known as expert systems, are a facet of Artificial Intelligence (AI). AI is a sub-field of computer science that focuses on the development of intelligent software and hardware systems that emulate human reasoning techniques and capabilities. Knowledge base systems emulate the decision-making processes of humans and are one of the most commercially successful AI technologies. These systems are used in a variety of applications for business, science and engineering. Business applications capture a company's critical business knowledge and utilize it for decision support.

Artificial Intelligence Technology Definitions

A knowledge base system may employ any number of approaches to knowledge representation and manipulation from the AI world including:

• Rule-based Systems capture knowledge in the form of structured if-then statements.
• Model-Based Reasoning uses software models to capture knowledge or to emulate real processes.
• Neural Nets are a network of nodes and connections used to capture knowledge, they can "learn" by using examples.
• Fuzzy Logic is used to represent and manipulate knowledge that is incomplete or imprecise.
• Decision Trees capture decision-making knowledge that can be expressed as sets of order decisions.

Rule-based Systems

A rule based system uses "rules" as the knowledge representation for knowledge coded into the system. Rules typically take the form of if-then statements. This is a popular and intuitive knowledge representation. Constraint knowledge which identifies a set of conditions or a limit is easily represented using rules. Another form of knowledge, pattern matching, is also a good candidate to be implemented using rules.

The term business rules and rule-based system are often confused. Business rules typically refer to knowledge important to operating a business, in contrast a rule-based system refers to a type of knowledge representation. A rule based system may be an effective way to capture certain types of "business rules" (i.e. business knowledge) although, depending on the type of knowledge, other representations may be more effective. [back to top]

Model Based Reasoning

Model based reasoning was initially developed to support industrial processes such as oil refining or chemical processes. This technology uses a mathematical model that mimics the real process. Possible control actions can be applied to the model and the resulting effects can be observed. The model is used to predict the outcomes of various control actions thus providing a basis for selecting the best control action.

The model-based technique is a very powerful knowledge representation. This concept can also be applied to the business domain. Models can be constructed to capture the gist of business processes. These models can then be manipulated to predict the effects of various actions. Built as part of a knowledge base system, the models can predict outcomes based on different business scenarios. This type of reasoning is very useful as part of a sophisticated decision support system.

One of the key challenges with this technique is ensuring the model has the proper fidelity and captures the important characteristics of the process being modeled. [back to top]

Artificial Neural Nets

Artificial neural nets were developed from experiments to model the behavior of brain tissue using software. These experiments were some of the earliest forms of artificial intelligence software.

Neural nets are good at associative problems. Given partial information, an associative problem is to find items that "fit with" (i.e. are associated with) the given information. For example, birds are small animals with feathers that fly. Given a feathered animal that flies, we can find associated information - namely that this animal is likely to be a bird and is probably small.

A key advantage of neural nets is that they can be trained by example. To encode knowledge into a neural net many examples of the desired information can be presented to the neural net. Each example causes the neural net to alter its structure and store the new information. After training, the knowledge is stored in the neural net as a pattern of weights distributed across all the connections between individual neurons. These connected neurons make up the neural net. This easy training is off-set by the difficulty in identifying the knowledge stored in the neural net. There is no descriptive form of the knowledge captured in a neural net. The knowledge is only a distribution of connection weights. [back to top]

Fuzzy Logic

Fuzzy logic has its roots in set theory. It was developed to handle situations where membership in sets is not clearly defined.

This technique is very useful for handling imprecise information. For example, what if we are looking at the P/E ratio of an internet company and wish to assess if the ratio is "high"? In this case, we might consider a ratio of 500 to be high, but what about a ratio of 200? Fuzzy Logic could use a number to represent the membership of our specific P/E ratio in the set of high P/E ratios -- a P/E ratio of 200 might be considered 0.5 high. This technique helps avoid problems with hard constraints. A hard constraint with the lower bound for high P/E ratios of 200 would disqualify a P/E ratio of 199 from the high category. Using fuzzy logic this same P/E ratio would be considered about 0.5 high.

Fuzzy logic can often be combined with other knowledge representations. For example, rules can use fuzzy logic expressions to allow them to more effectively handle imprecise information. [back to top]

Decision Trees

Decision trees predate computer-based artificial intelligence. This technique has been used for many years to lay out the conditions and steps required for decisions.

Decision trees are useful for capturing structured decision-making processes. This technique is useful for troubleshooting and configuration applications. The knowledge for these applications is often structured into a set of steps and decision points.

One problem with this technique is lack of flexibility. Decision trees must be defined ahead of time thus limiting their flexibility. It is possible to combine decision trees with other AI techniques to lessen this problem. Despite this limitation decision trees can be very effective representations for specific types of knowledge. [back to top]