Video summary
Lec 22 Applications of Expert Systems
Main summary
Key takeaways
Summary
The lecture reviews certainty factors in expert systems, then discusses their advantages, limitations, applications, and basic structure.
Certainty Factors and Combining Evidence
A certainty factor represents confidence in a hypothesis using two measures:
- Measure of belief: How strongly the evidence supports the hypothesis.
- Measure of disbelief: How strongly the evidence argues against it.
The two measures are combined to obtain a certainty factor. The lecturer notes that, for a given hypothesis, belief and disbelief are generally not both positive at once: evidence either supports the hypothesis or weighs against it.
When combining evidence from two observations, the system uses rules for combining their belief measures and, separately, their disbelief measures.
For two hypotheses:
- Belief in H1 AND H2 is the minimum of the two belief values.
- Belief in H1 OR H2 is the maximum of the two belief values.
Advantages of Expert Systems
- Economical and widely deployable: Once developed, a system can be copied and used in many locations, reducing the cost per user.
- Available and fast: It can be accessed when needed and generally responds faster than a human expert.
- Consistent and reliable: It applies its encoded facts and rules consistently, without human fatigue, distraction, or emotional influence.
- Can explain its reasoning: Users or human experts can inspect how the system reached a conclusion.
- Preserves expertise: Knowledge remains available even when human experts leave, and knowledge from several experts can be combined.
- Processes information systematically: It can consider the facts and rules encoded for its domain.
- Can be maintained and improved: A knowledge engineer can update the knowledge base to improve results.
The lecturer also describes expert systems as efficient, free from human memory limits, and potentially useful for securely handling information.
Limitations and Risks
- Narrow, shallow knowledge: A system knows what has been encoded for its specific domain; it lacks deep understanding, common sense, and knowledge of related domains.
- Limited ability to handle mixed knowledge: Combining expertise from different domains can be difficult.
- Errors can propagate: A faulty rule or system bug may affect many decisions or diagnoses.
- Development and maintenance can be difficult and expensive.
- Some apparently simple tasks may be computationally costly for a computer, even if a human can handle them easily.
- Users may lack trust in computer-generated decisions, especially for critical matters.
- Legal and ethical accountability is unresolved: If an automated decision is wrong, responsibility may be difficult to assign. The lecturer illustrates this with a hypothetical system that incorrectly grades students’ exam papers.
Areas of Application
The lecture lists applications across many sectors:
- Finance: Asset management, robo-advice, market and financial-indicator predictions, stock recommendations, credit granting, and loan analysis.
- Mechanical engineering: Troubleshooting complex electromechanical machinery.
- Telecommunications: Decisions about network technologies.
- Healthcare: Assistance with medical diagnosis.
- Agriculture: Forecasting crop damage.
- Other areas: Customer service, transportation, law, information management and retrieval, product-embedded AI, hospital and facility layout, help desks, and employee performance evaluation.
Structure and Example of an Expert System
Expert systems are knowledge-based systems that represent and process domain knowledge. Their main components are:
- Knowledge base: Stores facts and rules.
- Inference engine: Applies those rules to derive conclusions.
These components are kept separate, unlike in many ordinary programs. Different inference engines—for example, one using forward chaining and another using backward chaining—can potentially work with the same knowledge base.
The lecturer demonstrates the idea with a family-tree system in Prolog:
- Encode facts such as gender and parent relationships.
- Add rules to derive relationships such as siblings, brothers, grandparents, nieces, nephews, or uncles.
- Ask questions about relationships; Prolog’s built-in inference engine searches the knowledge base and applies the rules to produce answers.
A small, clearly defined domain can therefore be relatively straightforward to encode, though larger or more complex systems may be difficult and costly to build.
Speakers and Sources
- Speaker: One unnamed lecturer, teaching an NPTEL lecture associated with the Indian Institute of Science, Bengaluru.
- Other speakers: None identifiable in the subtitles. Prolog and the family-tree example are discussed, but are not separate speakers.
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.