Unit 3: Expert systems - Subjective Questions

Definition: An expert system is an artificial intelligence application designed to solve problems and provide advice in a specific domain by imitating the decision-making abilities of a human expert.\n\nMain characteristics:\n- It contains specialized domain knowledge.\n- It uses reasoning techniques to draw conclusions.\n- It can explain its recommendations or decisions.\n- It is usually limited to a particular problem area.\n\nDifference from a conventional program:\n- A conventional program mainly follows explicitly coded procedures and algorithms.\n- An expert system separates knowledge from the reasoning mechanism.\n- Expert systems can use heuristic rules to handle uncertain or incomplete information.\n- They can often explain how a conclusion was reached, whereas conventional programs generally provide limited explanations.

Major characteristics of expert systems include:\n- Domain specificity: They are developed for a well-defined field such as medicine, finance, agriculture, or equipment diagnosis.\n- Knowledge-based operation: They use facts, rules, cases, and relationships collected from experts and other sources.\n- Inference capability: They derive new conclusions from available knowledge.\n- Heuristic reasoning: They use practical rules of thumb to solve complex problems.\n- Explanation facility: They can explain the reasoning behind a question, recommendation, or conclusion.\n- Handling uncertainty: Some systems use confidence values, probabilities, or fuzzy logic.\n- Consistency: They can apply their knowledge consistently without fatigue or emotional bias.\n\nThese capabilities enable an expert system to support, rather than necessarily replace, human experts.

The architecture of an expert system generally consists of the following components:\n\n- Knowledge base: Stores domain-specific facts, rules, procedures, and relationships.\n- Inference engine: Applies reasoning strategies to the knowledge base and user-supplied facts to derive conclusions.\n- User interface: Allows users to enter information and receive advice or explanations.\n- Working memory: Holds facts and intermediate results related to the current problem.\n- Explanation facility: Describes how the system reached a conclusion or why it requested particular information.\n- Knowledge acquisition module: Helps collect, organize, validate, and update knowledge from human experts or other sources.\n\nThe user provides facts through the interface. These facts are placed in working memory, and the inference engine uses them with the knowledge base to produce a recommendation or conclusion.

The knowledge base is the central repository of information about the problem domain. It stores the knowledge required by the inference engine to reason about a problem.\n\nIts functions include:\n- Storing factual knowledge, such as symptoms, machine conditions, or financial indicators.\n- Storing heuristic knowledge in the form of rules or guidelines.\n- Representing relationships among objects, events, and concepts.\n- Providing domain knowledge for drawing conclusions.\n- Supporting explanations of the system's recommendations.\n- Allowing knowledge to be modified, expanded, or updated.\n\nKnowledge may be represented using production rules, semantic networks, frames, ontologies, cases, or logical statements. The quality, completeness, and accuracy of the knowledge base strongly influence the performance of the expert system.

An inference engine is the reasoning component of an expert system. It applies rules and other reasoning methods to facts stored in working memory and knowledge contained in the knowledge base.\n\nTypical operation:\n1. It receives facts supplied by the user or generated during reasoning.\n2. It searches the knowledge base for rules whose conditions match those facts.\n3. It selects one or more applicable rules using a control strategy.\n4. It fires the selected rule and adds its conclusion to working memory.\n5. It repeats the process until a solution, recommendation, or stopping condition is reached.\n\nThe inference engine may also manage conflicts among rules, handle uncertainty, request additional information, and provide explanations for its conclusions.

Forward chaining:\n- It is a data-driven reasoning method.\n- It begins with known facts.\n- It repeatedly applies rules whose conditions are satisfied.\n- It continues until a conclusion or goal is reached.\n- It is useful when many possible conclusions may be derived from available data.\n\nBackward chaining:\n- It is a goal-driven reasoning method.\n- It begins with a desired conclusion or hypothesis.\n- It searches for rules that could establish that goal.\n- It then attempts to prove the conditions of those rules by requesting or deriving supporting facts.\n- It is useful when the system has a specific goal to verify.\n\nMain distinction: Forward chaining moves from facts to conclusions, whereas backward chaining moves from a goal to the facts needed to support it.

Consider the following rules:\n\n- Rule 1: If an engine does not start and the battery is weak, then the battery may need replacement.\n- Rule 2: If the battery is weak, then the headlights may be dim.\n\nSuppose the initial facts are:\n- The engine does not start.\n- The battery is weak.\n\nForward-chaining process:\n1. The inference engine places the initial facts in working memory.\n2. It identifies that the conditions of Rule 1 are satisfied.\n3. It derives the conclusion that the battery may need replacement.\n4. It also identifies that the condition of Rule 2 is satisfied.\n5. It derives the conclusion that the headlights may be dim.\n\nThus, forward chaining begins with known data and progressively derives new conclusions by firing applicable rules.

Consider the following rules:\n\n- Rule 1: If a patient has a fever and a sore throat, then the patient may have an infection.\n- Rule 2: If a patient has an infection, then medical examination is recommended.\n\nSuppose the goal is to determine whether medical examination is recommended.\n\nBackward-chaining process:\n1. The system starts with the goal that medical examination is recommended.\n2. It searches for a rule that concludes this goal and finds Rule 2.\n3. It identifies the subgoal that the patient has an infection.\n4. It searches for a rule that can establish this subgoal and finds Rule 1.\n5. It asks whether the patient has a fever and a sore throat.\n6. If both facts are confirmed, the system proves that the patient may have an infection and therefore recommends medical examination.\n\nBackward chaining starts with a hypothesis and works backward to verify the facts needed to support it.

The user interface provides communication between the user and the expert system. Its quality directly affects the usefulness, acceptance, and reliability of the system.\n\nImportance of the user interface:\n- It enables users to enter facts, symptoms, observations, or requirements.\n- It presents conclusions, advice, warnings, and recommendations.\n- It can ask questions in a logical and understandable sequence.\n- It allows users to request explanations about the system's reasoning.\n- It reduces misunderstanding by using clear terminology and helpful instructions.\n- It can display confidence levels, alternatives, and supporting evidence.\n- It improves accessibility for users who are not familiar with artificial intelligence.\n\nAn accurate inference engine is not sufficient if users cannot communicate with the system easily or understand its output.

An effective user interface should make interaction clear, efficient, and trustworthy. Important features include:\n\n- Clarity: Questions, options, and results should be expressed in simple and unambiguous language.\n- User guidance: The interface should explain what information is required and how it should be entered.\n- Consistency: Similar actions and terms should be presented in the same way throughout the system.\n- Error handling: It should identify invalid or incomplete input and help the user correct it.\n- Explanation support: Users should be able to ask why a question was asked or how a conclusion was obtained.\n- Flexible interaction: It may support menus, forms, natural language, voice input, or graphical displays.\n- Useful output: Results should include recommendations, reasons, confidence information, and possible alternatives where appropriate.\n- Accessibility: The design should accommodate users with different levels of technical knowledge and ability.

Knowledge acquisition is the process of collecting, organizing, validating, and encoding expertise for use in an expert system.\n\nMain steps include:\n1. Identify the domain: Define the problem area, objectives, users, and system boundaries.\n2. Select knowledge sources: Obtain information from human experts, manuals, databases, case records, and observations.\n3. Elicit knowledge: Use interviews, questionnaires, observation, protocol analysis, and document analysis.\n4. Represent knowledge: Convert the information into rules, frames, cases, semantic networks, or other suitable structures.\n5. Validate knowledge: Check whether the encoded knowledge is accurate, complete, consistent, and relevant.\n6. Test the system: Compare its conclusions with those of qualified experts using representative cases.\n7. Maintain knowledge: Update the knowledge base when regulations, procedures, or domain conditions change.\n\nA major challenge is the difficulty experts may have in expressing their intuitive or experience-based knowledge explicitly.

Production-rule representation:\n- Represents knowledge as condition-action or condition-conclusion rules.\n- A typical form is: IF conditions are true, THEN perform an action or infer a conclusion.\n- It is suitable for diagnostic and decision-making tasks.\n- Rules are relatively easy to understand and modify.\n- Large rule collections may become difficult to manage and may contain conflicts.\n\nFrame-based representation:\n- Represents knowledge using structured objects called frames.\n- A frame contains slots for attributes and values.\n- It is suitable for representing entities, situations, and their properties.\n- Frames can support inheritance, allowing one concept to receive properties from another.\n- They are useful for organizing hierarchical and descriptive knowledge.\n\nComparison: Rules emphasize reasoning and actions, while frames emphasize the structured description of objects and concepts. Many expert systems use both approaches together.

Real-world information is often incomplete, imprecise, or uncertain. Expert systems can handle such information using several techniques:\n\n- Certainty factors: Attach a numerical confidence value to a fact or rule.\n- Probability: Represent the likelihood of events or conclusions using probabilities.\n- Bayesian reasoning: Update the probability of a hypothesis when new evidence becomes available.\n- Fuzzy logic: Represent gradual concepts such as high temperature or moderate risk using degrees of membership.\n- Default reasoning: Use a typical assumption unless contrary evidence is found.\n- Evidence-based reasoning: Combine support from different sources while considering their reliability.\n\nThe system should communicate uncertainty clearly. For example, it may provide a recommendation together with its confidence level and the evidence supporting it. Incorrect handling of uncertainty can lead to misleading or unsafe conclusions.

The explanation facility enables an expert system to describe its reasoning process to the user. It generally answers two important questions:\n\n- Why was a question asked? This explains which rule or hypothesis required particular information.\n- How was a conclusion reached? This shows the facts and rules used to produce the result.\n\nSignificance:\n- Builds user confidence in the system.\n- Helps users identify incorrect or missing input.\n- Assists experts in validating the knowledge base.\n- Supports training and learning.\n- Makes decisions more transparent and auditable.\n- Helps diagnose errors in the inference process.\n\nA system that provides advice without any explanation may be rejected, especially in sensitive fields such as medicine, law, finance, and safety. Explanation quality is therefore an important part of expert-system design.

Expert systems can provide several important advantages:\n\n- Availability: They can provide assistance at any time and in multiple locations.\n- Consistency: They apply the same knowledge and rules to similar cases.\n- Speed: They can analyze large amounts of information quickly.\n- Knowledge preservation: They capture valuable expertise before experienced personnel retire or leave.\n- Reduced cost: They can reduce the need for repeated consultation with scarce specialists.\n- Training support: They can teach less experienced employees by explaining decisions.\n- Improved decision-making: They provide structured recommendations based on stored knowledge.\n- Safety: They can be used in dangerous environments or for preliminary diagnosis without exposing people to risk.\n- Scalability: The same system can support many users once it has been developed.\n\nThese advantages are greatest when the problem domain is stable, well understood, and suitable for symbolic reasoning.

Despite their benefits, expert systems have important limitations:\n\n- Limited domain: They generally perform well only within the specific area for which they were developed.\n- Knowledge acquisition bottleneck: Extracting expert knowledge can be expensive, slow, and difficult.\n- High development cost: Building, testing, documenting, and maintaining a reliable system requires significant resources.\n- Lack of common sense: A system may not understand everyday knowledge that humans take for granted.\n- Dependence on knowledge quality: Incorrect, incomplete, or outdated rules can produce incorrect conclusions.\n- Difficulty with novel situations: The system may fail when a case does not match its stored knowledge.\n- Maintenance problems: Rules and facts must be updated as the domain changes.\n- Limited emotional and social understanding: It cannot fully replace human judgment, empathy, or ethical reasoning.\n- User resistance: Users may distrust recommendations that they cannot understand or verify.\n\nTherefore, expert systems should usually be treated as decision-support tools rather than complete substitutes for human experts.

Expert systems:\n- Provide fast and consistent responses.\n- Can store and apply large quantities of specialized knowledge.\n- Do not become tired or emotionally influenced.\n- Can preserve knowledge and make it widely available.\n- Are limited by their programmed knowledge and may fail in unfamiliar situations.\n\nHuman experts:\n- Can use common sense, intuition, creativity, and broad contextual understanding.\n- Can adapt to unusual situations and learn from new experiences.\n- Can consider ethical, social, and emotional factors.\n- May be expensive, unavailable, inconsistent, or affected by fatigue and personal bias.\n\nConclusion: Expert systems are valuable for repetitive, knowledge-intensive, and well-defined tasks. Human experts remain essential for novel, ambiguous, socially sensitive, and ethically complex decisions. The most effective approach often combines human judgment with machine-based support.

The problem-solving cycle generally occurs as follows:\n\n1. Problem presentation: The user describes the problem and provides initial facts through the user interface.\n2. Data validation: The system checks whether the information is complete, valid, and consistent.\n3. Working-memory update: The supplied facts are stored in working memory.\n4. Rule matching: The inference engine compares working-memory facts with conditions in the knowledge base.\n5. Rule selection: If multiple rules are applicable, a conflict-resolution strategy selects the next rule.\n6. Inference: The selected rule is fired and its conclusion or action is added to working memory.\n7. Additional questioning: The system may request further information if the available facts are insufficient.\n8. Conclusion: The process continues until a solution, diagnosis, recommendation, or stopping condition is obtained.\n9. Explanation: The system presents the result and, when required, explains the reasoning and supporting evidence.\n10. User feedback: The user may accept, reject, or provide feedback that supports later evaluation and improvement.

A conflict occurs when more than one rule is applicable at the same time. Conflict resolution is the process of selecting which rule should be fired first.\n\nCommon strategies include:\n- Priority or salience: Fire the rule assigned the highest priority.\n- Specificity: Prefer the rule with the most specific conditions.\n- Recency: Prefer rules that use the most recently added facts.\n- Rule ordering: Use the order in which rules appear in the knowledge base.\n- Complexity: Prefer rules that contain a greater number of matching conditions.\n- Utility: Select the rule expected to produce the most useful result.\n\nConflict resolution prevents unpredictable behavior and helps the system reason efficiently. The chosen strategy should match the requirements of the application and should be documented so that the system's behavior remains understandable.

A problem is suitable for an expert-system solution when several conditions are satisfied:\n\n- The domain has a clearly defined scope.\n- The problem requires specialized knowledge rather than only numerical calculation.\n- Qualified experts are available to provide and validate knowledge.\n- The knowledge can be represented using rules, cases, frames, or another formal structure.\n- The problem occurs frequently enough to justify development costs.\n- A consistent decision process is desirable.\n- The environment is reasonably stable or the knowledge can be updated regularly.\n- The consequences of errors can be controlled through human review when necessary.\n- Users need assistance, training, diagnosis, or decision support.\n\nAn expert system may be unsuitable when the task depends heavily on creativity, rapidly changing information, physical skills, broad common sense, or complex emotional and ethical judgment.