* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Download Artificial Intelligence and Expert Systems
Survey
Document related concepts
Linear belief function wikipedia , lookup
Intelligence explosion wikipedia , lookup
Ethics of artificial intelligence wikipedia , lookup
Ecological interface design wikipedia , lookup
Existential risk from artificial general intelligence wikipedia , lookup
Human–computer interaction wikipedia , lookup
Computer Go wikipedia , lookup
Philosophy of artificial intelligence wikipedia , lookup
Embodied cognitive science wikipedia , lookup
Expert system wikipedia , lookup
Personal knowledge base wikipedia , lookup
Transcript
CHAPTER 6 Knowledge-Based Decision Support Opening Vignette: Case Study A Knowledge-based DSS in A Chinese Chemical Plant. Dalian Dyestuff plant is one of the largest chemical plants in China. It produces about 100 different kinds of dyes and other chemical products. With the economic reform in China, manufacturing decisions were decentralized. The plant managers were suddenly faced with the problem of determining their own production plans. Because of the size of the plant and the number of products, it became very difficult to make and appropriately change production plans, which depend on market demand. The plant also had to make purchasing decisions and decisions regarding of the disposal of environmentally damaging materials. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-2 6.1 Concepts and Definitions Managerial Decision Makers are Knowledge Workers Use Knowledge in Decision Making Accessibility to Knowledge Issue Knowledge-Based Decision Support: Applied Artificial Intelligence Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-3 6.1 Concepts and Definitions Artificial Intelligence (AI) is a term that (in this textbook) Encompasses many definitions • AI involves studying human thought processes (to understand what intelligence is) • Representing thought processes on machines (such as computer and robots) One well-published definition of AI is : Behavior by a machine that, if performed by a human being, would be considered intelligent “…study of how to make computers do things at which, at the moment, people are better” (Rich and Knight [1991]) Theory of how the human mind works (Mark Fox) • – – – Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-4 6.1 Concepts and Definitions Winston and Prendergast [1984] list three Objectives of AI – Make machines smarter (primary goal) – Understand what intelligence is (Nobel Laureate purpose) – Make machines more useful (entrepreneurial purpose) (Winston and Prendergast [1984], The AI Business. Cambridge, MA: MIT Press) The meaning of the term Intelligent behavior as the follows: Or on other words, signs of Intelligence: Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-5 6.1 Concepts and Definitions – – – – – – – – – Fall, 2007 Learn or understand from experience Make sense out of ambiguous or contradictory messages Respond quickly and successfully to new situations Use reasoning to in solving problems and directing conduct effectively Deal with perplexing situations Understand and infer in ordinary, rational ways Apply knowledge to manipulate the environment Think and reason Recognize the relative importance of different elements in a situation All Right Reserved, Zhong YAO, School of E&M, BHU 6-6 6.1 Concepts and Definitions – However, Although AI’s ultimate goal is to build machines that mimic human intelligence, the capabilities of current commercial AI products are far from exhibiting any significant success in the abilities just listed. Nevertheless, AI programs are continually improving, and they increase productivity and quality by automating several tasks that requires some human intelligence. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-7 6.1 Concepts and Definitions Testing for Intelligence An interesting test designed to determine whether a computer exhibits intelligent behavior was designed by Alan Turing and is called the Turing test. A computer can be considered to be smart only when a human interviewer, “conversing” with both an unseen human being and an unseen computer, could not determine which is which The following definitions and characteristics of AI focus on decision-making and problem solving. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-8 6.1 Concepts and Definitions Symbolic Processing When human experts solve problems, particularly the types that are considered appropriate for AI, they do not do it by solving sets of equations or performing other laborious mathematical computations. Instead, they choose symbols to represent the problem concepts and apply various strategies and rules to manipulate these concept. The AI approach represents knowledge as sets of symbols that stands for problem concepts. In Summary, – Use Symbols to Represent Problem Concepts – Apply Various Strategies and Rules to Manipulate these Concepts Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-9 6.1 Concepts and Definitions A symbol is a string of characters that stands for some real-world concept Examples Product – Defendant – 0.8 – Chocolate These symbols can be combined to express meaningful relationships, which calls for symbol structure. – Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-10 6.1 Concepts and Definitions Symbol Structures (Relationships) – – – – (DEFECTIVE product) (LEASED-BY product defendant) (EQUAL (LIABILITY defendant) 0.8) tastes_good (chocolate). Interpreted to “the product is defective”, “product is leased by the defendant”, “the liability of the defendant is 0.8”, and “chocolate taste good”. However, they may be interpreted differently. This is one of the problems we encounter in building AI systems. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-11 6.1 Concepts and Definitions – – – – Fall, 2007 To solve a problem, an AI Programs Manipulate Symbols to Solve Problems Symbols and Symbol Structures Form Knowledge Representation Symbolic processing (definition) is an essential characteristics of AI. As reflected by a definition of AI in a branch of computer science: Artificial Intelligence Dealings Primarily with Symbolic, Non-algorithmic Problem- Solving Methods These definition focuses on two characteristics: All Right Reserved, Zhong YAO, School of E&M, BHU 6-12 6.1 Concepts and Definitions – Fall, 2007 Numeric versus Symbolic Computers were originally designed specifically to process numbers (numeric processing). However, people tend to think symbolically; our intelligence seems to be based, in part, on our mental ability to manipulate symbols rather than just numbers. Although symbol processing is a core in AI, this does not mean that AI does not involve math; rather, the emphasis in AI is on the manipulation of symbols. All Right Reserved, Zhong YAO, School of E&M, BHU 6-13 6.1 Concepts and Definitions – Fall, 2007 Algorithmic versus Non-algorithmic An algorithm is a step-by-step procedure that has welldefined starting and ending points and is guaranteed to reach a solution to a specific problem. Most computer architectures readily lend themselves to this step-bystep approach. Many human reasoning processes tend to be non-algorithmic; in other words, our mental activities consist of more than just following logical, step-by-step procedures. All Right Reserved, Zhong YAO, School of E&M, BHU 6-14 6.1 Concepts and Definitions Heuristics Heuristics are included as a key element of AI in the following definition: “AI is the branch of computer science that deals with ways of representing knowledge using symbols rather than numbers and with rules-of-thumb, or heuristics, methods for processing information.(Encyclopedia Britannica) – Inferencing AI involves an attempt by machines to exhibit reasoning capabilities. The reasoning consists of inferencing from facts and rules using heuristics or other search approaches. AI is unique in that it makes inferencing by using a pattern Matching approach. – Pattern Matching Attempt to describe objects, events, or processes in terms of their qualitative features and logical and computational Fall, 2007 relationships. All Right Reserved, Zhong YAO, School of E&M, BHU 6-15 6.2 Artificial Intelligence versus Natural Intelligence The potential value of AI can be better understood by contrasting it with natural, or human intelligence. More permanent. Natural intelligence is perishable from a commercial standpoint in that workers can change their place of employment or forget information. However, AI is permanent as long as the computer systems and programs remain unchanged. Ease of duplication and dissemination. Transferring a body of knowledge from one person to another usually requires a lengthy process of apprenticeship; even so, expertise can never be duplicated completely. However, when knowledge is embodies in a computer system, it can be copies from what computer and easily moved to another computer, sometimes across the globe. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-16 6.2 Artificial Intelligence versus Natural Intelligence Less expensive than the natural intelligence. There are many circumstances in which buying computer service costs less than having corresponding human power carry out the same tasks. AI, being a computer technology, is consistent and thorough: Natural intelligence is erratic because people are erratic; they do not always perform consistently. Can be documented. Decisions made by a computer can be easily documented by tracing the activities of the system. Natural intelligence is difficult to reproduce. For example, a person may may reach a conclusion but at some later date may be unable to recreate the reasoning process that led to that conclusion or to even recall the assumptions that were a part of the decision. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-17 6.2 Artificial Intelligence versus Natural Intelligence Can execute certain tasks much faster than a human Can perform certain tasks better than many or even most people Natural language does have several advantages over AI. Some are: Natural intelligence is creative, whereas AI is rather uninspired. The ability to acquire knowledge is inherent in human beings, but with AI, tailored knowledge must be built into a carefully constructed system. Natural intelligence enables people use sensory experience directly, whereas most AI system must work with symbolic input and representations. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-18 6.2 Artificial Intelligence versus Natural Intelligence Perhaps most importantly, human beings reasoning is able to use at all times a wide context of experience bring that to bear on individual problems. In contrast, AI systems typically gain their power by having a Very Narrow Focus Information Processing – – – Fall, 2007 Computers can collect and process information efficiently (such as a large amount of information) People instinctively: • Recognize relationships between things • Sense qualities • Spot patterns indicating relationships BUT, AI technologies can provide significant improvement in productivity and quality! All Right Reserved, Zhong YAO, School of E&M, BHU 6-19 6.3 Knowledge in Artificial Intelligence What is knowledge? (based on Sowa[1985]) – – Fall, 2007 Knowledge encompasses the implicit and explicit restrictions placed upon objects (entities), operations and relationships along with general and specific heuristics and inference procedures involved in the situation being modeled. Major characteristics that distinguishes AI from other CBIS is that AI’s major emphasis is knowledge processing (rather than data or information processing). Knowledge is now recognized as a major organization resource. • Data, information and knowledge can be classified by their degree of abstraction and by their quantity. Knowledge is the most abstraction and exists in the smallest quantity. All Right Reserved, Zhong YAO, School of E&M, BHU 6-20 6.3 Knowledge in Artificial Intelligence High Degree of Abstraction Knowledge Information Low Data Quantity Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-21 Processed INFORMATION Relevant and actionable DATA KNOWLEDGE 6.3 Knowledge in Artificial Intelligence Relevant and actionable data Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-22 6.3 Knowledge in Artificial Intelligence Uses of knowledge – – Fall, 2007 Although the computers can not have a diversity of experiences, or study and learn as the human mind can, it can use knowledge given to it by human experts. Such knowledge consists of facts, concepts, theories, heuristics methods, procedures and relationships. Knowledge is also information that has been organized and analyzed to make it understandable and applicable to problem solving or decision making. The collection of knowledge related to a problem (or an opportunity) used in an AI system is organized together and it is called a knowledge base. Most knowledge bases are limited in that they typically focus on some specific, usually narrow subject area or domain. All Right Reserved, Zhong YAO, School of E&M, BHU 6-23 6.3 Knowledge in Artificial Intelligence In fact, the narrow domain of knowledge and the fact that an AI system must involve some qualitative aspects of decision making are viewed as critical for AI application success. – Once a Knowledge base is built, AI techniques are used to give the computer inference capabilities based on the facts and relationships contained in the knowledge base. Knowledge Bases – With a knowledge base and the ability to draw inferences from it, the computer can be put to practical use as a problem solver and decision maker. Figure shows a application of KB. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-24 6.3 Knowledge in Artificial Intelligence Computer Inputs (questions, problem, etc. Knowledge Inferencing Base capability Outputs (answers, alternative, solution, etc. By searching the knowledge base for relevant facts and relationships, the computer can reach one or more alternative solutions to the given problems. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-25 6.3 Knowledge in Artificial Intelligence Knowledge Engineering (definition) (Feigenbaum and McCorduck [1983]) – The art of bringing the principles and tools of AI research to bear on difficult applications problems requiring expert’s knowledge for their solutions. – The technical issues of acquiring this knowledge, representing it and using it appropriately to construct and explain lines of reasoning are important problems in the design of knowledge-based systems. – The art of constructing intelligent agents is both part of and an extension of the programming art. It is the art of building complex computer programs that represent and reason with knowledge of the world. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-26 6.3 Knowledge in Artificial Intelligence Knowledge engineering process: (Narrow scope) – Knowledge acquisition: acquisition of knowledge from human experts, books, documents, sensors, or computer files. Knowledge may be specific to the problem domain or to the problem-solving procedures, or it may be general knowledge, or it may be metaknowledge (knowledge about knowledge --- information about how experts use their knowledge to solve problems and problem-solving procudures) – Knowledge representation: Acquired knowledge is organized in an activity called knowledge representation. For example, preparation of knowledge map and encoding the knowledge in the knowledge base. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-27 6.3 Knowledge in Artificial Intelligence – Knowledge validation: knowledge is validated and verified until its quality is acceptable. Test case results are usually shown to the experts to verify the accuracy of the ES. – Inferencing: Design of software to enable the computer to make inference based on the knowledge and specifics of a problem. Then the system can provide advice to a nonexpert user. – Explanation and justification: Design and programming of an explanation capability; e.g., programming the ability to answer question like why a specific piece of information is needed by the computer or how a certain conclusion was derived by a computer. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-28 6.3 Knowledge in Artificial Intelligence Knowledge engineering process: A overview Knowledge validation (test cases) Sources of knowledge (experts, others) Knowledge Acquisition Knowledge base Encoding Knowledge Representation Explanation justification Inferencing Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-29 6.3 Knowledge in Artificial Intelligence – Wide scope • Entire process of developing and maintaining AI Knowledge Types Advantaged knowledge Base knowledge Trivial knowledge (琐细的\一般性的) Explicit knowledge Objective, rational, technical Easily documented Easily transferred / taught / learned Tacit knowledge Subjective, cognitive, experiential learning Hard to document Hard to transfer / teach / learn Involves a lot of human interpretation Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-30 6.3 Knowledge in Artificial Intelligence Knowledge Management A process of elicitation, transformation, and diffusion of knowledge throughout an enterprise so that it can be shared and thus reused Helps organizations find, select, organize, disseminate, and transfer important information and expertise Transforms data / information into actionable knowledge to be used effectively anywhere in the organization by anyone Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-31 6.3 Knowledge in Artificial Intelligence Process of explication may ge ner ate ne w tacit knowle dge Explicit Knowledge Tacit Tacit Knowledge Knowledge Conver t tacit knowledge into artic ulated and measurable explicit knowledge Policies, patents, decisions, stra tegies, IS, white papers, etc. Expertise, know-how, ideas, orga nization culture, values, etc. Core Competencies of the Organization Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-32 6.3 Knowledge in Artificial Intelligence KM Objectives Create knowledge repositories Improve knowledge access Enhance the knowledge environment Manage knowledge as an asset Chief Knowledge Officer (CKO) Fall, 2007 Maximize firm’s knowledge assets Design and implement KM strategies Effectively exchange knowledge assets Promote system use All Right Reserved, Zhong YAO, School of E&M, BHU 6-33 6.4 Knowledge Acquisition Knowledge Acquisition Difficulties – Problems in Transferring Knowledge • Expressing Knowledge • Transfer to a Machine • Number of Participants • Structuring Knowledge – Experts may lack time or not cooperate – Testing and refining knowledge is complicated – Poorly defined methods for knowledge elicitation – System builders may collect knowledge from one source, but the relevant knowledge may be scattered across several sources Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-34 6.4 Knowledge Acquisition Knowledge Acquisition Difficulties – – – – – Fall, 2007 May collect documented knowledge rather than use experts The knowledge collected may be incomplete Difficult to recognize specific knowledge when mixed with irrelevant data Experts may change their behavior when observed and/or interviewed Problematic interpersonal communication between the knowledge engineer and the expert All Right Reserved, Zhong YAO, School of E&M, BHU 6-35 6.4 Knowledge Acquisition Knowledge Acquisition Methods – Manual – Semiautomatic – Automatic (Computer Aided) Manual Methods - Structured Around Interviews – Process (See Figure) – Interviewing – Tracking the Reasoning Process – Observing – Manual methods: slow, expensive and sometimes inaccurate Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-36 6.4 Knowledge Acquisition Experts Knowledge engineer Coding Knowledge base Documented knowledge Semiautomatic Methods – – Fall, 2007 Support Experts Directly (see Figure) Help Knowledge Engineers All Right Reserved, Zhong YAO, School of E&M, BHU 6-37 6.4 Knowledge Acquisition Expert Computer-aided (interactive) interviewing Coding Knowledge base Knowledge engineer Automatic (Computer Aided) – – Fall, 2007 Expert’s and/or the knowledge engineer’s roles are minimized (or eliminated) Induction Method (see Figure ) All Right Reserved, Zhong YAO, School of E&M, BHU 6-38 6.4 Knowledge Acquisition Case histories and examples Induction system Knowledge base Knowledge Modeling – – Fall, 2007 The knowledge model views knowledge acquisition as the construction of a model of problem-solving behavior-- a model in terms of knowledge instead of representations Can reuse models across applications All Right Reserved, Zhong YAO, School of E&M, BHU 6-39 6.4 Knowledge Acquisition -----Manual methods Interviews – – Most Common Knowledge Acquisition: Face-to-face interviews Interview Types • Unstructured (informal) • Semi-structured • Structured Tracking Methods – – – Fall, 2007 Techniques that attempt to track the reasoning process of an expert From cognitive psychology Most common formal method: Protocol Analysis All Right Reserved, Zhong YAO, School of E&M, BHU 6-40 6.4 Knowledge Acquisition – Protocol Analysis: • Protocol: a record or documentation of the expert's step-by-step information processing and decisionmaking behavior • The expert performs a real task and verbalizes his/her thought process (think aloud) Observations and Other Manual Methods • • • • Fall, 2007 Case analysis Critical incident analysis Discussions with the users Commentaries All Right Reserved, Zhong YAO, School of E&M, BHU 6-41 6.4 Knowledge Acquisition • Conceptual graphs and models • Brainstorming • Prototyping • Multidimensional scaling • Johnson's hierarchical clustering • Performance review Expert-driven Methods – Fall, 2007 Knowledge Engineers Typically • Lack Knowledge About the Domain • Are Expensive • May Have Problems Communicating With Experts All Right Reserved, Zhong YAO, School of E&M, BHU 6-42 6.4 Knowledge Acquisition – – Knowledge Acquisition May be Slow, Expensive and Unreliable Can Experts Be Their Own Knowledge Engineers? Expert-driven Methods May Use – – Fall, 2007 Manual---Expert's Self-reports Computer-Aided (Semiautomatic) – REFINER+ - case-based system – Visual modeling techniques – New machine learning methods to induce decision trees and rules – Tools based on repertory grid analysis All Right Reserved, Zhong YAO, School of E&M, BHU 6-43 6.4 Knowledge Acquisition Automated Knowledge Acquisition (Machine Learning) – – Fall, 2007 Rule Induction • Induction: Process of Reasoning from Specific to General • In ES: Rules Generated by a Computer Program from Cases • Interactive Induction Case-based Reasoning • For Building ES by Accessing Problem-solving Experiences for Inferring Solutions for Solving Future Problems • Cases and Resolutions Constitute a Knowledge Base All Right Reserved, Zhong YAO, School of E&M, BHU 6-44 6.4 Knowledge Acquisition Automated Knowledge Acquisition (Machine Learning) – – Neural Computing • Fairly Narrow Domains with Pattern Recognition • Requires a Large Volume of Historical Cases Intelligent Agents • KQML (Knowledge Query and Manipulation Language) for Knowledge Sharing • KIF, Knowledge Interchange Format (Among Disparate Programs) Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-45 6.5 Knowledge Representation Knowledge representation Once acquired, knowledge must be organized for use. – A good knowledge representation naturally represents the problem domain – An unintelligible (难解的) knowledge representation is wrong – Most artificial intelligence systems consist of: • Knowledge Base – Forms the system's intelligence source – Inference mechanism uses to reason and draw conclusions • Inference Mechanism (Engine) Examines the knowledge base to answer questions, solve problems or make decisions domain Fall, 2007 All Rightwithin Reserved,the Zhong YAO, School of E&M, BHU 6-46 6.5 Knowledge Representation Knowledge representation – Many knowledge representation schemes • Can be programmed and stored in memory • Are designed for use in reasoning – Major knowledge representation schemas: • Production rules • Frames Representation in Logic and other Schemas – General form of any logical process – Inputs (Premises) – Premises used by the logical process to create the output, consisting of conclusions (inferences) Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-47 6.5 Knowledge Representation – Facts known true can be used to derive new facts that are true Symbolic logic: System of rules and procedures that permits the drawing of inferences from various premises Basic Forms of Computational Logic – Propositional logic (or propositional calculus) – Predicate logic (or predicate calculus) Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-48 6.5 Knowledge Representation Propositional Logic – – – – A proposition is a statement that is either true or false Once known, it becomes a premise that can be used to derive new propositions or inferences Rules are used to determine the truth (T) or falsity (F) of the new proposition Symbols represent propositions, premises or conclusions Statement: A = The mail carrier comes Monday through Friday. Statement: B = Today is Sunday. Conclusion: C = The mail carrier will not come today. – Fall, 2007 Propositional logic: limited in representing real-world knowledge All Right Reserved, Zhong YAO, School of E&M, BHU 6-49 6.5 Knowledge Representation Predicate Calculus – – – – Predicate logic breaks a statement down into component parts, an object, object characteristic or some object assertion Predicate calculus uses variables and functions of variables in a symbolic logic statement Predicate calculus is the basis for Prolog (PROgramming in LOGic) Prolog Statement Examples • comes_on(mail_carrier, monday). • likes(jay, chocolate). – (Note - the period “.” is part of the statement) Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-50 6.5 Knowledge Representation Scripts Knowledge Representation Scheme Describing a Sequence of Events Elements include – – – – – Fall, 2007 Entry Conditions Props Roles Tracks Scenes All Right Reserved, Zhong YAO, School of E&M, BHU 6-51 6.5 Knowledge Representation Lists: Written Series of Related Items – Normally used to represent hierarchical knowledge where objects are grouped, categorized or graded according to • Rank or • Relationship Decision Tables (Induction Table): Knowledge Organized in a Spreadsheet Format – Attribute List – Conclusion List – Different attribute configurations are matched against the conclusion Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-52 6.5 Knowledge Representation Decision Trees – Related to tables – Similar to decision trees in decision theory – Can simplify the knowledge acquisition process – Knowledge diagramming - very natural O-A-V Triplet – Objects, Attributes and Values – O-A-V Triplet – Objects may be physical or conceptual – Attributes are the characteristics of the objects – Values are specific measures of the attributes Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-53 6.5 Knowledge Representation Representative O-A-V Items Object Attributes Values House Bedrooms 2, 3, 4, etc. House Color Green, white, brown, etc. Admission to a university Grade-point average 3.0, 3.5, 3.7, etc. Inventory control Level of inventory 14, 20, 30, etc. Bedroom Size 9 X 10, 10 X 12, etc. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-54 6.5 Knowledge Representation • Default Logic Knowledge Maps Deals with uncertainties Incomplete information Visual representation Cognitive maps Semantic Networks Fall, 2007 Graphic Depiction of Knowledge Nodes and Links Showing Hierarchical Relationships Between Objects Nodes: Objects Arcs: Relationships All Right Reserved, Zhong YAO, School of E&M, BHU 6-55 6.5 Knowledge Representation – • is-a • has-a Semantic networks can show inheritance Semantic Nets - visual representation of relationships Can be combined with other representation methods – Semantic Network Example – – Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-56 6.5 Knowledge Representation Human Being Boy Needs Goes to School Woman Joe Food Has a child Kay Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-57 6.5 Knowledge Representation Production Rules – Condition-Action Pairs • IF this condition (or premise or antecedent) occurs, • THEN some action (or result, or conclusion, or consequence) will (or should) occur • IF the stop light is red AND you have stopped, THEN a right turn is OK – – Fall, 2007 Each production rule in a knowledge base represents an autonomous chunk of expertise When combined and fed to the inference engine, the set of rules behaves synergistically All Right Reserved, Zhong YAO, School of E&M, BHU 6-58 6.5 Knowledge Representation Rules can be viewed as a simulation of the cognitive behavior of human experts – Rules represent a model of actual human behavior Forms of Rules – IF premise, THEN conclusion • IF your income is high, THEN your chance of being audited by the IRS (美国国税局)is high – Conclusion, IF premise • Your chance of being audited is high, IF your income is high Inclusion of ELSE – IF your income is high, OR your deductions are unusual, THEN your chance of being audited by the IRS is high, OR ELSE your chance of being audited is low – Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-59 6.5 Knowledge Representation More Complex Rules – IF credit rating is high AND salary is more than $30,000, OR assets are more than $75,000, AND pay history is not "poor," THEN approve a loan up to $10,000, and list the loan in category "B.” – Action part may have more information: THEN "approve the loan" and "refer to an agent" Knowledge and Inference Rules – Common Types of Rules: • Knowledge rules, or declarative rules, state all the facts and relationships about a problem • Inference rules, or procedural rules, advise on how to solve a problem, given that certain facts are known • Inference rules contain rules about rules (metarules) • Knowledge rules are stored in the knowledge base • Inference rules become part of the inference engine Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-60 6.5 Knowledge Representation Advantages of Rules – – – – – Easy to understand (natural form of knowledge) Easy to derive inference and explanations Easy to modify and maintain Easy to combine with uncertainty Rules are frequently independent Limitations of Rules – – – Fall, 2007 Complex knowledge requires many rules Builders like rules Search limitations in systems with many rules All Right Reserved, Zhong YAO, School of E&M, BHU 6-61 Characteristics of Rule Representation First Part Second Part Names Premise Antecedent Situation IF Conclusion Consequence Action THEN Nature Conditions, similar to declarative knowledge Resolutions, similar to procedural knowledge Size Can have many IFs Usually one conclusion AND statements All conditions must be true for a conclusion to be true OR statements If any of the OR statement is true, the conclusion is true Statements Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-62 6.5 Knowledge Representation Frames: Definitions and Overview – Frame: Data structure that includes all the knowledge about a particular object – Knowledge organized in a hierarchy for diagnosis of knowledge independence – Form of object-oriented programming for AI and ES. – Each Frame Describes One Object – Special Terminology Default Instantiation Demon Master frame Facet Object Hierarchy of frames Range If added Slot If needed Value (entry) Instance of Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-63 6.5 Knowledge Representation Concise, natural, structural representation of knowledge Encompasses complex objects, entire situations or a management problem as a single entity Frame knowledge is partitioned into slots Slot can describe declarative knowledge or procedural knowledge Major Capabilities of Frames (next slide) Typical frame describing an automobile Hierarchy of Frames: Inheritance Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-64 6.5 Knowledge Representation Frame Capabilities Ability to clearly document information about a domain model; for example, a plant's machines and their associated attributes Related ability to constrain the allowable values that an attribute can take on Modularity of information, permitting ease of system expansion and maintenance More readable and consistent syntax for referencing domain objects in the rules Platform for building a graphic interface with object graphics Mechanism that will allow us to restrict the scope of facts considered during forward or backward chaining Access to a mechanism that supports the inheritance of information down a class hierarchy Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-65 6.5 Knowledge Representation Multiple Knowledge representations – Rules + Frames – Others Knowledge Representation Must Support – – – Fall, 2007 Acquiring knowledge Retrieving knowledge Reasoning All Right Reserved, Zhong YAO, School of E&M, BHU 6-66 6.5 Knowledge Representation Considerations for Evaluating a Knowledge Representation – Naturalness, uniformity and understandability – Degree to which knowledge is explicit (declarative) or embedded in procedural code – Modularity and flexibility of the knowledge base – Efficiency of knowledge retrieval and the heuristic power of the inference procedure Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-67 6.5 Knowledge Representation No single knowledge representation method is ideally suited by itself for all tasks Multiple knowledge representations: each tailored to a different subtask Production Rules and Frames works well in practice Object-Oriented Knowledge Representations – Hypermedia Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-68 6.5 Knowledge Representation Experimental Knowledge Representations – – – Cyc NKRL Spec-Charts Language The Cyc System – – – – – Fall, 2007 Attempt to represent a substantial amount of common sense knowledge Bold assumptions: intelligence needs a large amount of knowledge Need a large knowledge base Cyc over time is developing as a repository of a consensus reality - the background knowledge possessed by a typical U.S. resident There are some commercial applications based on portions of Cyc All Right Reserved, Zhong YAO, School of E&M, BHU 6-69 6.5 Knowledge Representation NKRL – Narrative Knowledge Representational Language (NKRL) – Standard, language-independent description of the content of narrative textual documents – Can translate natural language expressions directly into a meaningful set of templates that represent the knowledge – Fall, 2007 Knowledge Interchange Format (KIF): To Share Knowledge and Interact All Right Reserved, Zhong YAO, School of E&M, BHU 6-70 6.5 Knowledge Representation The Spec-Charts Language – – – Based on Conceptual Graphs: to Define Objects and Relationships Restricted Form of Semantic Networks Evolved into the Commercial Product – STATEMATE Knowledge Representation and the Internet – – – Fall, 2007 Hypermedia documents to encode knowledge directly Hyperlinks Represent Relationships MIKE (Model-based and Incremental Knowledge Engineering All Right Reserved, Zhong YAO, School of E&M, BHU 6-71 6.5 Knowledge Representation Formal model of expertise: KARL Specification Language Semantic networks: Ideally suited for hypermedia representation Web-based Distributed Expert System (Ex-W-Pert System) for sharing knowledge-based systems and groupware development Representing Uncertainty: An Overview – Dealing with Degrees of Truth, Degrees of Falseness in ES – Uncertainty • When a user cannot provide a definite answer • Imprecise knowledge • Incomplete information Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-72 6.5 Knowledge Representation Uncertainty: Several Approaches Related to Mathematical and Statistical Theories – Bayesian Statistics – Dempster and Shafer's Belief Functions(规则的有效性 和限制 ) – Fuzzy Sets Uncertainty in AI: Approximate Reasoning, Inexact Reasoning Relevant Information is Deficient in One or More – Information is partial – Information is not fully reliable Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-73 6.5 Knowledge Representation – Representation language is inherently imprecise – Information comes from multiple sources and it is conflicting – Information is approximate – Non-absolute cause-effect relationships exist – Can include probability in the rules – IF the interest rate is increasing, THEN the price of stocks will decline (80% probability) Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-74 6.6 How AI differs from conventional computing Conventional Computing. Conventional computer programs are based on an algorithm, which is a clearly defined, step-by-step procedure for solving a problem. It may be a mathematical formula or a sequential procedure that will let to a solution. The algorithm is converted into a computer program that tells the computer exactly what operations to perform. The algorithm then uses data, symbol, or words to solve the problem. – Conventional computer process data: Calculate, Perform Logic, Store, Retrieve, Sort, Edit, Make structured decision, Monitor, Control, Networking AI computing: AI software is based on symbolic representation and manipulation. A symbol is a letter, word, or number that is used to represent objects, process, – Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-75 6.6 How AI differs from conventional computing – Fall, 2007 and their relationships. Objects can be people, things, ideas, concepts, events, or statements of fact. By using symbol , it is possible to create a knowledge base that states facts, concepts, and the relationships among them. Then, various processes are used to manipulate the symbols to generate advice or a recommendation for solving problems. Once Knowledge Base has been built, it can be used to solve the problem. How does the AI software reason or infer with KB? Basic technique is search and pattern matching. Given some initial information, the AI software searches the knowledge base, looking for specific conditions or patterns. It looks for match ups All Right Reserved, Zhong YAO, School of E&M, BHU 6-76 6.6 How AI differs from conventional computing – – – that satisfy the criteria set up to solve the problem. The computer literally hunts around until it finds the best answer, based on the knowledge it has. Even though AI problem solving does not take place directly by algorithmic processes, algorithms are used to perform the search process. AI is not magic. An AI system is a computer-based information system, although it has some distinct characteristics. AI vs. conventional programming Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-77 6.6 How AI differs from conventional computing AI vs. Conventional programming Processing Nature of input Search Explanation Major interest structure Nature of output Maintenance and update Hardware Reasoning capability Fall, 2007 Primarily symbolic Can be incomplete Heuristics(mostly) Provided Knowledge Separation of control from knowledge Can be incomplete Easy because of modularity Mainly workstations and personal computers Limited, but improving Primarily algorithm Must be complete Algorithm Usually not provided data, information Control integrated with information (data) Must be complete usually difficult All types None All Right Reserved, Zhong YAO, School of E&M, BHU 6-78 6.7 Does A Computer really Thinking? Knowledge base and search technique certainly make computers more useful, but can they really make computer more intelligent? AI specialists, computer scientists, and others regularly debate this question. The fact that most AI programs are implemented by search and pattern-matching techniques leads to the conclusion that computers are not really intelligent. Although AI is making computers act smarter and more powerful, the dream of building a machine that can fully duplicate the human brain will not be realized in our lifetime. AI methods are valuable. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-79 6.8 Artificial Intelligence Field Expert Systems are computerized advisory programs that attempt to imitate the reasoning processes and knowledge of experts in solving specific problems. They are used more than any other applied AI technology. Expert systems are of great interest to organization because of their ability to enhance productivity and to augment workforces in many specialty areas where human experts are becoming increasing difficult to find and retain. – Software: • EXSYS (EXSYS Inc. http://www.exsysinfo.com) • K-version (Ginesys Corp. http://www.ginesys.com) • KnowledgePro (Knowledge Garden Inc. http://www.kgarden.com) Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-80 6.8 Artificial Intelligence Field Natural Language Technology Natural language technology gives computer users the ability to communicate with the computer in their native language. It allows for a conversational type of interface, in contrast to using a programming language of computer jargon, syntax, and commands. However, limited success in this area is typified by current systems. – Natural Language Processing (NLP) consists of two sub-fields: • Natural Language Understanding investigates methods of enabling the computer to comprehend instructions given in ordinary English so that computer can understand people more easily. • Natural Language Generation strives to have computers produce ordinary English language so that people can understand computer are easily. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-81 6.8 Artificial Intelligence Field Speech (voice) Understanding is the recognition and understanding by a computer of spoken language. (Next chapter will be talked in detail) Robotics and Sensory Systems – Sensory Systems, such as vision systems, tactile systems, and signal processing systems, when combined with AI, define a broad category of systems generally called Robotics. – A robot is an electromechanical device that can be programmed to perform manual tasks. – Not all of robotics is considered to be part of AI. Computer Vision and Scene Recognition – Visual recognition has been defined as the addition of some of form of computer intelligence and decision Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-82 6.8 Artificial Intelligence Field making to digitized visual information received from a machine sensor such as camera. The combined information is then used to perform, or control, such operations as robotic movement, convey speeds, and producing line quality. – The basic objective of CV is to interpret scenarios rather than generate pictures. Intelligent Computer-Aided Instruction (ICAI) – Refers to machines that can tutor humans. To a certain extent, such a machine can be viewed as an expert system. However, the major objective of an ES is to render advice, whereas the purpose of ICAI is to teach. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-83 6.8 Artificial Intelligence Field Neural Computing – A neural network is a mathematical model of the way a brain functions. Details are not provided in this course. Other Applications – Automatic Programming-CASE – Summarizing News – Language Translation – Fuzzy Logic – Genetic Algorithm – Intelligent Agents Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-84 6.8 Artificial Intelligence Field Big Picture ( see figure 6.3) Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-85 6.9 Types of Knowledge-based DSS The activities supported by the expert systems in this case were different from the activities supported by the data and model components of the DSS. Thus, the knowledge component enabled a wide range of decisions: it extends the capabilities of computer well beyond data-based and model-based DSS. Other possible supports are as follows: – Support the decisions un-addressed by mathematics. – Support the building, storing, and managing of models in a multiple-model DSS. Enhance the MBMS, making it intelligence. – Support for the uncertainty, where expertise in applying tools ranging from fuzzy logic to neural computing. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-86 6.9 Types of Knowledge-based DSS – – Fall, 2007 Support for user interfaces. The user interface plays a major role in DSS implementation. For example, natural language processing and voice techniques can make the interfaces very easy and natural. Others such as Intelligent Agents. All Right Reserved, Zhong YAO, School of E&M, BHU 6-87 6.10 Intelligent Decision Support Systems Representative IDSS: Active (symbiotic) DSS: Regular DSS play a passive role in human-machine interaction. The DSS executes computations, presents data, and responds to standard commands. But it can not play the role of an intelligent assistant to the decision makers. This restricts the use of some DSS to well-defined and unambiguous tasks. – However, certain tasks in problem solving are ambiguous and complex. Then, we need an active DSS. – According to Mili [1990], active DSS is applicable to the following tasks: • Understanding the domain (terminology, parameters, interactions) , for example, provide a explanation. • Formulating problems. Help in determining assumptions, abstracting reality, deciding what is relevant, and so on. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-88 6.11 Intelligence DSS Relating a problem to a solver. The active DSS can assist with proper problem-solver interaction, advise what procedures to use and what solution techniques to follow, and so on. – Interpreting results. – Explaining results and decisions. Self-evolving DSS. A self-evolving DSS is an approach to DSS design, whose basic premise is that a DSS should be “aware” of how it is being used, and then it should automatically adapt to the evolution of its users. This capability is achieved by adding an extra component: an intelligent self-evolving mechanism. – A dynamic menu provides different hierarchies to fulfill different user requirements – A dynamic user interface that provides different output representation for different users – Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-89 6.11 Intelligence DSS – An intelligent model base management system can select appropriate models to satisfy different preferences. Structure of Self-Evolving DSS Major components are: – Data management, model management, and a user interface, which are basic components of any DSS. – A usage record that contains system usage data pertinent to the evolution of the system and its management (EMS, or evolution management systems) – User interface elements that are needed for creating a very user-friendly interface. – Central control mechanism that coordinates all the operations of the DSS. This is an intelligent control that contains a knowledge base. Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-90 6.11 Intelligence DSS User Data User Model Management Interface management Control Mechanism Knowledge base Usage User Record EMS Interface elements Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-91 6.11 Intelligence DSS – Fall, 2007 The control mechanism collects the user’s usage data and stores them in the usage record base. Then the control mechanism analyzes these data and a new version of the DSS is created. This process is repetitive and continues as long as necessary. The control can work on each component of the DSS independently. In other words, it provides an intelligent interfaces, intelligent model management, and intelligent data management. Finally, the system is an online, real-time system. All Right Reserved, Zhong YAO, School of E&M, BHU 6-92 6.11 Intelligence DSS Problem Management – – – – – Fall, 2007 Most DSS revolve around the design and choice phases of decision making. The intelligence phase, which includes problem finding, problem representation, and information surveillance, is neglected by most DSS. Moreover, several activities in the design and choice phases, such as model management, are executed manually To make DSS more effective, it is necessary to automate as many tasks as possible. Weber and Konsynski [1987] suggested dividing the decision-making process into five steps and proposed architecture support to the functional requirements of these steps. They called their approach problem management. Table 6.3 shows it most contents. All Right Reserved, Zhong YAO, School of E&M, BHU 6-93 6.12 Problem Management table 6.3 Problem Management: functional requirements, and architechutural support Problem management stage Functional Requirements Architectural Support Flexible knowledge Perceptual filters, knowledge management, intelligent Problem finding management filters Flexible dialogue and knowledge management, reason maintenance Problem Model and pattern management, system, pattern search representation suspension of judgement strategies Demons, intelligent Information lenses, scanners, serveillance Knowledge and model management evaluators interpreters Idea and solution model Knowledge management, idea management, heuristic Solution generation generation and analytical drives Flexible knowledge Meta-level dialog and knowledge management, analytic Solution evalution management and symbolic processors Fall, 2007 All Right Reserved, Zhong YAO, School of E&M, BHU 6-94 Assignments (individual) 1. 2. 3. 4. 5. 6. 7. 8. 9. 10. 11. Fall, 2007 Distinguish among data, information and knowledge. Define a Knowledge base, as one of component of DSS, what can it support in DSS. Define an AI. What is symbolic processing. Define an Expert System. What is a natural language processing? What is an intelligent DSS (IDSS)? What is Active DSS and Self-Evolving DSS? Explain what is problem management. The knowledge acquisition/representation has several methods, please list these ways. What is the Production Rule in knowledge representation? All Right Reserved, Zhong YAO, School of E&M, BHU 6-95