Expert Systems An Overview of Expert Systems Expert

Expert Systems An Overview of Expert Systems

Expert Systems TOPICS q The nature of expertise • Who is an Expert, and Why? q The Characteristics of an Expert Systems • What Makes it different and Why ? q Additional Issues in Expert Systems • Knowledge acquisition (Building knowledge bases) • Knowledge assessment • Explanation facilities

Expert Systems The Nature of Expertise q Assumes a highly specialized set of Skills • NOT just general knowledge q Assumes a very specialized problem domain • Analogous to our previous ‘Forest vs. Tree’ Idea q Assumes logic, problem solving and experience • NOT simple intuition or indefinable behaviors

Expert Systems The Nature of Expertise Performance q Who is an Expert? ? • That is NOT an easy Question • There are many practitioner but very few experts Expertise • Notice that just because you have experience, that does NOT mean that you are an expert q Characteristics of Experts • Fast, ACCURATE, problem Solving • Pattern Recognition • Use of Heuristics – Based on past experience • Scarcity

Expert Systems The Nature of Expertise q Necessary Expert Traits • Be Recognized as an Expert • Know how they perform the task • Can NOT just act intuitively without being able to explain their behaviors • Have the time and ability to explain how they perform • Be Motivated to Cooperate

Expert Systems The Nature of Expertise q How do you know who is an expert? ? • Also NOT an easy Question, although some are obvious • There are references, However (a few off the Internet): • Expert. Pages. com: A directory for legal professionals in search of experts, expert witnesses, or consultants. Search by state, country, or subject area. http: //www. expertpages. com/ • Experts Directory A searchable directory of experts from the legal, medical, journalism and other professions. http: //www. experts. com Are they really Experts ? ? ? Don’t Mortgage the House!

Expert Systems Expert System Characteristics “An expert system is a computer program that represents and reasons with knowledge of some specialist subject with a view to solving problems or giving advice. ” Jackson (1999) q Turing Test • A computer program demonstrates artificial intelligence if it can “pass’ as a human (c. 1950) 1912 -54 • In 1990, the Cambridge Center for Behavioral Studies began offering the $100, 000 Loebner Prize to the first program whose responses were indistinguishable from a human’s (No one has ever won)

Expert Systems Expert System Characteristics • Gary Kasparov vs. IBM’s Deep Blue • May 11, 1997 • Garry Kasparov resigned 19 moves into Game 6 • Deep Blue wins the Best of Six game series 3. 5 to 2. 5 • IBM Development Team wins $700, 000 • Kasparov wins $400, 000 • The first win by a computer program over an International Grand Master since man/computer games were first began in 1970

Expert Systems Expert System Characteristics q Basic Requirements • simulates human reasoning • Rule/Heuristic Based: Rule: If there is a potato in the tailpipe, the car will not start. Finding: There is a potato in the tailpipe. Conclusion: The car will not start. (Truth preserving inference) Rule: If there is a potato in the tailpipe, the car will not start. Finding: My car will not start. Conclusion: Therefore, there is a potato in the tailpipe. (Non-Truth preserving inference)

Expert Systems Expert System Characteristics q Basic Requirements • simulates human reasoning • Inference Engines • The ‘Driving’ Force in an Expert System • Reasons with any rule constructed via rule set manager • Searches for applicable rules • Evaluates the predicates of those rules to determine their “truth” • Executes the actions specified in “fired” (activated) rules

Expert Systems Expert System Characteristics q Basic Requirements • simulates human reasoning • Inference Engines • Forward Chaining • Corresponds to the idea of Deductive reasoning Theory Birds can Fly Hypothesis Ostriches Can Fly Observation OK – I was wrong ! Rejection (I Fly to Australia) Confirmation

Expert Systems Expert System Characteristics q Basic Requirements • simulates human reasoning • Inference Engines • Forward Chaining • Corresponds to the idea of Deductive reasoning • Consists of a condition part and an action part • Conditions (rules) are matched against the database • If true, the action is fired • The forward chaining engine cycles repeatedly until it runs out of rules or a rule instructs it to stop.

Expert Systems Expert System Characteristics q Basic Requirements • simulates human reasoning • Inference Engines • Forward Chaining • Backward Chaining • Corresponds to the idea of Inductive reasoning Theory Ostriches Can’t Fly (what a Moron I was!) Not all Birds can Fly Tentative Hypothesis Pattern Observation Birds Flying, but no Ostriches I’m back in The Australian Outback – Bird watching

Expert Systems Expert System Characteristics q Basic Requirements • simulates human reasoning • Inference Engines • Forward Chaining • Backward Chaining • Corresponds to the idea of Inductive reasoning • Involves trying to prove a given goal by using rules to generate sub-goals and recursively trying to satisfy them. • The engine looks at conclusions and determines all rules that could reach that conclusion • Each rule is then examined for its premises • If true, the rule is fired and a value is established • The process continues until all possible solutions are generated

Expert Systems Expert System Characteristics q Basic Requirements • simulates human reasoning • Knowledge Representation • Knowledge Bases • A repository (Database) of data and metadata • Contains all the Rules established by the manager • The data are stored as objects, which can be fired as needed • Includes Symbolic data • Includes Relationships between data • May be used in conjunction with a standard database

Expert Systems Expert System Characteristics q Basic Requirements • • simulates human reasoning Knowledge Representation Deal with realistically complex Problems Reach Multiple Conclusions • Especially as a result of backward chaining • Explain the conclusions reached • The logic used must be demonstratable • Deal with Missing Information • “Fuzzy Logic” • Non-numerical Analysis • Demonstrate High Performance • Should approximate the performance of the expert

Expert Systems Expert System Characteristics q Basic Requirements q ES Components User Interface Inference Engine Database ES Shell A rule engine and scripting Environment Knowledge Base

Expert Systems Expert System Characteristics q Basic Requirements q ES Components q Differences Between ES and DSS Expert Systems • Based On Expert • Based on Logical Reasoning Decision Support Systems • No Experts Available • System Questions User • Used Frequently • Based on Numerical Analysis • User Questions System • Used for Ad-hoc Problems • • Final Solution(s) Provided Very Accurate Multiple Solutions Learning Possible Outputs provided based Analysis Unknown Accuracy Always the same output

Expert Systems Additional Topics q Knowledge Acquisition “The transfer and transformation of potential problem-solving expertise from some knowledge source to a program” - Buchanan et al. (1983) • Transfer of the Expert’s Knowledge as a set of rules into the Knowledge Base • Since the Expert is not expected to code the rules, a Knowledge Engineer is required • lengthy & intense interviews Required • slow (2 to 5 units of knowledge /day) ? ? ? Why ? ? ? • Imprecise, illogical, jargon or colloquialisms, experience, contextual detail, reliability of sources, . . .

Expert Systems Additional Topics q Knowledge Acquisition • Example: How to find a forgotten Password: Expert (Computer Center Guru): Well, if it’s a YP password, I first log on as root on the YP master KE: (Knowledge Engineer): Er, what’s the YP master? Expert: It’s the diskful machine that contains a database of network information KE: ‘Diskful’ meaning - ? Expert: -it has the OS installed on local disk KE: Ah. (scribbles furiously) So you log on… Expert: As root. Then I edit the password datafile, remove the encrypted entry, and make the new password map. . . This is the weakest link in the process !!

Expert Systems Additional Topics q Knowledge Acquisition • Potential Solutions/Problems • automated knowledge elicitation • interactive programs/automated conversation • Problem: There are no Good Programs available (yet) • textual scanning • Parsing of conversations to extract the important components • Problem: NLP is still in its infancy • machine learning • deriving decision rules from examples • evaluating / weighting rules • performance optimization of rules • Problem: Only Limited Success to date I don’t get it ! Me Neither

Expert Systems Additional Topics q Knowledge Acquisition q Knowledge Assessment • logical adequacy • sound & complete inferencing • heuristic Power • efficiency Vs. optimality (Effectiveness) • notational Convenience • How accurately do the rules reflect the logic?

Expert Systems Additional Topics q Knowledge Acquisition q Knowledge Assessment q Explanation Facility • • Necessary to check validity of Solutions The Chain of reasoning must be logged Solution Accountability must be determined Deficiencies must be corrected

Expert Systems Additional Topics q q Knowledge Acquisition Knowledge Assessment Explanation Facility Available Packages/Tools • Symbolic Manipulation Languages • LISP (LISt Processor) • Prolog • Expert Shells • CLIPS (Free Download: http: //www. ghg. net/clips/CLIPS. html) • Jess (Free Download: http: //herzberg. ca. sandia. gov/jess/ ) • Others: A good list can be found at http: //www-2. cs. cmu. edu/afs/cs/project/ai-repository/ai/areas/expert/systems/0. html

Expert Systems