© C. Kemke1Reasoning - Introduction COMP 4200: Expert Systems Dr. Christel Kemke Department of Computer Science University of Manitoba
© C. Kemke2Reasoning - Introduction Reasoning in Expert Systems knowledge representation in Expert Systems shallow and deep reasoning forward and backward reasoning alternative inference methods metaknowledge
© C. Kemke3Reasoning - Introduction Expert performance depends on expert knowledge! Experts and Expert Systems Human Experts achieve high performance because of extensive knowledge concerning their field Generally developed over many years
© C. Kemke4Reasoning - Introduction Types of Knowledge Knowledge Representation in XPS can include: conceptual knowledge terminology, domain-specific terms derivative knowledge conclusions between facts causal connections causal model of domain procedural knowledge guidelines for actions
© C. Kemke5Reasoning - Introduction Knowledge Modeling in XPS Knowledge Modeling Technique in XPS mostly rule-based systems (RBS) rule system models elements of knowledge formulated independently as rules rule set is easy to expand often only limited ‘deep’ knowledge, i.e. no explicit coherent causal or functional model of the domain
© C. Kemke6Reasoning - Introduction Shallow and Deep Reasoning shallow reasoning also called “experiential reasoning” aims at describing aspects of the world heuristically short inference chains complex rules deep reasoning also called causal reasoning aims at building a model that behaves like the “real thing” long inference chains simple rules that describe cause and effect relationships
© C. Kemke7Reasoning - Introduction Dilbert on Reasoning 1
© C. Kemke8Reasoning - Introduction Dilbert on Reasoning 2
© C. Kemke9Reasoning - Introduction Dilbert on Reasoning 3
© C. Kemke10Reasoning - Introduction General Technology of XPS Knowledge + Inference core of XPS Most often Rule-Based Systems (RBS) other forms: Neural Networks, Case-Based Reasoning
© C. Kemke11Reasoning - Introduction Rule-Based Expert Systems Work with a set of facts describing the current world state a set of rules describing the expert knowledge inference mechanisms for combining facts and rules in reasoning
© C. Kemke12Reasoning - Introduction Inference Engine Agenda Knowledge Base (rules) Explanation Facility User Interface Knowledge Acquisition Facility Working Memory (facts)
© C. Kemke13Reasoning - Introduction Architecture of Rule-Based XPS 1 Knowledge-Base / Rule-Base stores expert knowledge as “condition-action-rules” (or: if- then- or premise-consequence-rules) objects or frame structures are often used to represent concepts in the domain of expertise, e.g. “club” in the golf domain. Working Memory stores initial facts and generated facts derived by the inference engine additional parameters like the “degree of trust” in the truth of a fact or a rule ( certainty factors) or probabilistic measurements can be added
© C. Kemke14Reasoning - Introduction Architecture of Rule-Based XPS 2 Inference Engine matches condition-part of rules against facts stored in Working Memory (pattern matching); rules with satisfied condition are active rules and are placed on the agenda; among the active rules on the agenda, one is selected (see conflict resolution, priorities of rules) as next rule for execution (“firing”) – consequence of rule can add new facts to Working Memory, modify facts, retract facts, and more
© C. Kemke15Reasoning - Introduction Architecture of Rule-Based XPS 3 Inference Engine + additional components might be necessary for other functions, like calculation of certainty values, determination of priorities of rules and conflict resolution mechanisms, a truth maintenance system (TMS) if reasoning with defaults and beliefs is requested
© C. Kemke16Reasoning - Introduction Rule-Based Systems - Example ‘Grades’ - Rules to determine ‘grade’ 1. study good_grade 2. not_study bad_grade 3. sun_shines go_out 4. go_out not_study 5. stay_home study 6. awful_weather stay_home
© C. Kemke17Reasoning - Introduction Example ‘Grades’ 1. study good_grade 2. not_study bad_grade 3. sun_shines go_out 4. go_out not_study 5. stay_home study 6. awful_weather stay_home Q1: If the weather is awful, do you get a good or bad grade? Q2: When do you get a good grade? Rule-Base to determine the ‘grade’:
© C. Kemke18Reasoning - Introduction Forward and Backward Reasoning forward reasoning Facts are given. What is the conclusion? A set of known facts is given (in WM); apply rules to derive new facts as conclusions (forward chaining of rules) until you come up with a requested final goal fact. backward reasoning Hypothesis (goal) is given. Is it supported by facts? A hypothesis (goal fact) is given; try to derive it based on a set of given initial facts using sub-goals (backward chaining of rules) until goal is grounded in initial facts.
© C. Kemke19Reasoning - Introduction 1.study good_grade 2.not_study bad_grade 3.sun_shines go_out 4.go_out not_study 5.stay_home study 6.awful_weather stay_home Example ‘Grades’ forward reasoningrule chain given fact: awful_weather 6,5,1 backward reasoning hypothesis/goal: good_grade 1,5,6
© C. Kemke20Reasoning - Introduction Example – Grades Working MemoryAgenda awful weatherRule 6 Select and apply Rule 6 awful weather stay home Rule 5 Select and apply Rule 5
© C. Kemke21Reasoning - Introduction Example – Grades Working MemoryAgenda Select and apply Rule 1 awful weather stay home study Rule 1 awful weather stay home study good grade empty DONE!
© C. Kemke22Reasoning - Introduction forward reasoning: Shield AND Pistol Police backward reasoning: Police Badge AND gun Police BadgeGun Shield PistolRevolver AND OR Bad Boy Example ‘Police’ – Reasoning Tree Q: What if only ‘Gun’ is known?
© C. Kemke23Reasoning - Introduction good grade Example ‘Grades’ – Reasoning Tree bad grade not studystudy go outstay home sun shinesawful weather
© C. Kemke24Reasoning - Introduction Police BadgeGun Shield PistolRevolver AND OR Bad Boy Example ‘Police’ – Reasoning Tree Q: What if only ‘Pistol’ is known as ground fact?
© C. Kemke25Reasoning - Introduction Police BadgeGun Shield PistolRevolver AND OR Bad Boy Example ‘Police’ – Reasoning Tree Task: Write down the Rule-Base for this example!
© C. Kemke26Reasoning - Introduction Forward vs. Backward Chaining Forward ChainingBackward Chaining planning, controldiagnosis data-drivengoal-driven (hypothesis) bottom-up reasoningtop-down reasoning find possible conclusions supported by given facts find facts that support a given hypothesis similar to breadth-first search similar to depth-first search antecedents (LHS) control evaluation consequents (RHS) control evaluation
© C. Kemke27Reasoning - Introduction Alternative Reasoning Methods Theorem Proving emphasis on mathematical proofs and correctness, not so much on performance and ease of use Probabilistic Reasoning integrates probabilities into the reasoning process Certainty Factors Express subjective assessment of truth of fact or rule Fuzzy Reasoning allows the use of vaguely defined predicates and rules
© C. Kemke28Reasoning - Introduction Metaknowledge deals with “knowledge about knowledge” e.g. reasoning about properties of knowledge representation schemes, or inference mechanisms usually relies on higher order logic in (first order) predicate logic, quantifiers are applied to variables second-order predicate logic allows the use of quantifiers for function and predicate symbols may result in substantial performance problems CLIPS uses meta-knowledge to define itself, i.e. CLIPS constructs, classes, etc. - in a bootstrapping form
© C. Kemke29Reasoning - Introduction Expert Systems Task Areas
© C. Kemke30Reasoning - Introduction Task Areas of Expert Systems System-based Problem or Task Description Analysis Tasks (Interpretation) Diagnosis Classification Synthesis Tasks (Construction) Construction Configuration Design Planning
© C. Kemke31Reasoning - Introduction Analysis Tasks Analysis Tasks (Diagnosis, Classification) determine specific solution element (diagnosis) based on a description of the system (symptoms or other descriptive facts) rules formulate connections between symptoms etc. and diagnostic class e.g. the medical expert system MYCIN for diagnosing bacterial infections e.g. tutoring systems like GUIDEON for diagnosing student’s mistakes
© C. Kemke32Reasoning - Introduction Synthesis Tasks Synthesis Tasks (Construction, Configuration, Design, Planning) combine elements from a component (solution) space and check consistency of complete solution rules formulate constraints and extensions for partial solution, similar to planning e.g. the technical expert system R1/XCON to configure computer systems
© C. Kemke33Reasoning - Introduction Expert Systems – Tasks Interpretation Prediction Diagnosis Design Planning Monitoring Debugging Instruction Control (after Hayes-Roth et al. (1983), Waterman (1986), cited from Luger and Stubblefield ‘Artificial Intelligence’, 1998, see Jackson, p.208)
© C. Kemke34Reasoning - Introduction Expert Systems – Tasks 1 Interpretation forming high-level conclusions from raw data Prediction projecting probable consequences of given situations Diagnosis determining the cause of malfunctions in complex situations based on observable symptoms Design finding a configuration of system components that meets performance goals while satisfying a set of design constraints
© C. Kemke35Reasoning - Introduction Expert Systems – Tasks 2 Planning devising a sequence of actions that will achieve a set of goals given certain starting conditions and run-time constraints Monitoring comparing a system’s observed behavior to its expected behavior Debugging and Repair prescribing and implementing remedies for malfunctions Instruction detecting and correcting deficiencies in students’ understanding of a subject domain Control governing the behavior of a complex environment