Links-RuleBasedSystems

Updated Sept 7 2008: Found new link to Diane Cook lecture notes ...

Rule Based Systems and Artificial Intelligence

One of the better and more accessible rule-based system links is the student notes for a course on AI given by Jocelyn Paine in 1996.

The Jocelyn Paine definition of rule-based systems:

Consists of a rule-base (permanent data); an inference engine (process); and a workspace or working memory (temporary data). Not part of the basic reasoning process, but essential to applications, is the user interface.

The definition continues with an operational type of definition - ''its when you do ...'.

A good and complete introduction is the lecture notes from the Artificial Intelligence course ( in PDF format ) given by Diane J. Cook ( now at Washington State University ) and Manfred Huber at the University of Texas Arlington.

The lecture notes are not intended to be a fully-integrated document, they read more like a detailed outline of ideas in PDF form. The subject of rule-based systems is only a part of the content of the course. Some sections of the lecture notes are devoted almost entirely to rule-based systems, but some parts of the subject are scattered about in sections devoted to other aspects of artificial intelligence ( neural networks, etc. )l.

The three sections primarily concerned with rule-based systems are:

• Knowledge Representation - including reasoning, logic, inference and semantic networks. Note the section "Problems With Semantic Networks" on page 23.
  

• Expert Systems and Rule-Based Systems - examples of forward and backward chaining, graphs and search spaces and solving complex planning problems.

• Uncertainty Reasoning - including probability, distributions, fuzzy logic, and Dempster-Shafer Theory.

The section for Machine Learning and has some good sections pertaining to rule-based systems ( as software agents ), but is mostly devoted to neural networks. It is also large, over 300KB. See ... link to under the hood.

The ideas presented in the sections above are highly summarized and provide only the most basic concepts about rule-based systems, but they also provide a very good outline of the subject of artificial intelligence. The three areas above are, in a sense, being lifted out of their AI context so that they can contribute to the definition of rule-based systems, without drifting into the subjects concerned with Real Artificial Intelligence, such as neural networks, which are clearly not rule-based. ( see Defining Rule Based Systems ).

Rules and AI Games

A refreshing break from the usual business examples pf rule-based systems is the article on Rule-Based Systems at AI Game Developers. Note that they are employing functional classifications of rules, such as body movement and collision avoidance. In effect, these are examples of structure->function->behavior maps. For example the query might be "in order to avoid ( F ) the object two feet in front of me ( S ) , how do I need to move my body (B)". There might be additional sub-queries generated by backchaining during plan generation, such as "how fast am I moving toward the object ?" or "how tall is the object ?". Can I jump over it ?

It might look as if the example above is producing the mapping function->structure->behavior, but that is a different map. For example, if I were tired of tripping over things on the floor, I would ask "how could I rearrange the room so there would be less chance of tripping over things ?". The F->S->B query starts with the goal of avoiding collisions ( F input ) and produces a plan of how I would have to re-arrange the room ( B output ), given a description of the current arrangement of furniture in my room ( the assumed or given middle term S that produces the mapping from F to B ). The first example starts with the recognition that I am about to trip over something ( inferred from a structural description of the world ) unless I can generate a plan of action ( behavior ) within a few fractions of a second.

Matrix-Based Systems

Another refreshing break is Combining Causal and Similarity-Based Reasoning by a group at MIT, primarily Charles Kemp and Josh Tenenbaum who have worked extensively in this area. It might be called a "matrix-based" as opposed to a rule-based approach, although the result is much the same. It has strong fuzzy logic and abduction capabilities.

Tags:   Rule   Based   Systems