Expert systems: how rule-based programs captured specialist knowledge
Expert systems were early AI programs that captured specialist knowledge in the form of rules. They relied on knowledge representation techniques to encode facts and heuristics. Their evolution showed the importance of separating knowledge from reasoning procedures.
Origins and purpose
Expert systems emerged from early AI research that aimed to replicate human reasoning in computers. The Computer History Museum's AI revolution page notes that attempts to make machines think like people have produced many useful products and services, even though true human-like thought remains elusive.
The field drew on logic, which provided techniques for analyzing the inferential properties of languages. According to the Stanford Encyclopedia of Philosophy's entry on logic-based AI, theoretical computer science developed out of logic, and most computer scientists are familiar with the idea that logic can specify what computations a program should license.
Early expert systems were built entirely from large sets of procedural rules, with no separate representation of background knowledge. The Stanford Encyclopedia of Philosophy explains that later generations of expert systems showed greater modularity, separating knowledge from the reasoning system.
How rules captured expertise
Rule-based expert systems encoded knowledge as condition-action pairs: if certain facts hold, then draw a conclusion or take an action. This approach allowed programmers to translate the heuristics of human specialists into a form a computer could apply.
The power of such systems depended on the quality and organization of the knowledge. The MIT CSAIL paper on knowledge representation describes a knowledge representation as a surrogate, a substitute for the thing itself, enabling reasoning about the world rather than acting in it.
Knowledge representation also involved ontological commitments, answering the question of what terms to use when thinking about the world. The MIT CSAIL paper emphasizes that these commitments begin with the earliest choices and accumulate in layers.
A representation served as a fragmentary theory of intelligent reasoning, specifying which inferences are sanctioned and which are recommended. The MIT CSAIL paper argues that keeping all five roles in mind provides a usefully broad perspective.
Key figures and contributions
John McCarthy's plan was to use ideas from philosophical logic to formalize commonsense reasoning. The Stanford Encyclopedia of Philosophy notes that McCarthy and his students pursued projects with a distinctly philosophical flavor, a theme that persisted and was absorbed into knowledge representation.
Nils J. Nilsson, a researcher at SRI International and later Stanford, co-invented the A* heuristic search algorithm and the STRIPS automatic planning system. His Stanford home page also mentions his work on the integrated mobile robot SHAKEY and his textbooks on artificial intelligence.
Nilsson's book 'The Quest for Artificial Intelligence: A History of Ideas and Achievements' is available online, and his home page describes it as written for both lay readers and researchers interested in the historical antecedents of current AI systems.
Evolution and legacy
The evolution of expert systems illustrates the importance of treating declarative representations as separate items with their own research problems. The Stanford Encyclopedia of Philosophy states that even limited-objective reasoning systems can call for large, complex bodies of declarative information.
Later expert systems showed greater modularity, separating knowledge from the reasoning mechanism. This shift allowed for more flexible and maintainable systems, as the Stanford Encyclopedia of Philosophy points out.
The principles of knowledge representation continue to influence modern AI. The MIT CSAIL paper concludes that representation and reasoning are intertwined, and that all five roles matter for research and practice.
The broader history of AI includes many milestones that shaped expert systems. The IBM history of AI page recounts early developments such as the Atanasoff-Berry Computer and the work of McCulloch and Pitts, which laid foundations for computational approaches to intelligence.
| Role | Question to ask | Example |
|---|---|---|
| Surrogate | Does the representation stand in for the real world? | A rule about engine failure represents the physical fault. |
| Ontological commitments | What terms are used to describe the domain? | Using 'patient', 'symptom', and 'disease'. |
| Theory of reasoning | Which inferences are allowed and recommended? | If symptom A and symptom B, then disease C. |
| Medium for computation | Is the representation efficient for the reasoning engine? | Rules indexed by condition for fast lookup. |
| Medium of expression | Can humans read and edit the knowledge base? | Rules written in a readable if-then syntax. |
Frequently asked questions
What is an expert system?
An expert system is a computer program that uses rules and knowledge from human specialists to solve problems in a specific domain, such as medical diagnosis or equipment repair.
How did expert systems represent knowledge?
They typically used if-then rules, where each rule encoded a piece of expertise. The knowledge base was separate from the inference engine that applied the rules.
What role did logic play in expert systems?
Logic provided techniques for analyzing reasoning and specifying what computations a program should perform, as noted by the Stanford Encyclopedia of Philosophy.
Why did later expert systems become more modular?
Later systems separated the knowledge representation from the reasoning procedures, making them easier to maintain and update, according to the Stanford Encyclopedia of Philosophy.
Are expert systems still used today?
The principles of rule-based reasoning and knowledge representation continue to influence modern AI, though many contemporary systems use other approaches like machine learning.