Transcript 智能科学与逻辑理论
Stage New Problems Approach Proposal -- to The Celebration of The 50th Anniversary of AI Y. X . Zhong Chinese Association for AI (CAAI) University of Posts & Telecom, Beijing [email protected] 1 List of Contents 1, Introduction 2, New Problems 3, New Approach 4, New Proposal 2 1, Introduction 3 The 50th Anniversary of The Birth of AI A good time for AI researchers worldwide to review what happened in the past 50 years, to analyze what will happen in the future, and to discuss what and how we should do next. 4 2, New Problems -- Inconsistent approaches and New demand 5 The Structuralism Approach 1943, McCulloch-Pitts: Logic Model of Nerve Cell. 1981, Hopfield: New Model and Learning Algorithm of Neural Networks 1990, Computational Intelligence -- Neural Networks -- Fuzzy Logic -- Evolution Computing -- Chaotic Theory -- Rough Set Theory 6 The Functionalism Approach 1956, Newell, Simon et al Logic Theorist, GPS, Means-Ends Analysis 1970, Feigenbaum et al Expert Systems (Knowledge Engineering) 1990 Hybrid Systems Data Mining and Knowledge Discovery Fuzzy Set Theory Machine Learning Multi-Agent System 7 The Behaviorism Approach 1990, Brooks: Intelligence without Representation; 1991, Brooks: Intelligence without Reasoning Pattern Recognition Action Response Effectiveness 8 Brief Comments 1, All have made great progresses while facing critical difficulties. 2, All are independent to each other and lack of coordination. 3, It leaves questions: What is the relationship among the three? Are there any better approaches to AI ? 9 New Demand from Intelligence Research N Natural Intelligence e w Emotion and Artificial Emotion Consciousness and Artificial Consciousness Cognitive Informatics Artificial Life Intelligent Robot Intelligent Agent, multi-Agent and Distribute AI Complex Systems and Intelligent Information Network 10 3, New Approach -- A better approach to AI 11 Mechanism Approach to AI Behavior Mechanism Structure, function and behavior of intelligent systems can provide some meaningful information on intelligence though Function a deeper insight approach to AI research should be concerned with the mechanism of intelligence formation. Structure 12 Mechanism Model on Human Intelligence Processed Information Processing Acquired Information Cognition Transferring Acquisition Knowledge Decision Transferring Real World Original Information Intelligent Strategy Execution Intelligent Action 13 Core Mechanism of Intelligence Formation Decision Making Cognition Information Knowledge Intelligence Transformations from information to knowledge and further to intelligence are the keys. 14 New Concept & Theory Needed (1) Comprehensive Information Theory (Y. X. Zhong, 1988-1996-2002) Formal Description Syntactic Object Meaning Symbol Semantic States Certainty Truth Utility x1 c1 t1 u1 Utility Subject Pragmatic xn cn tn un xN cN tN uN 15 n* log n* n log n (C P Q) n n * I ( , ; R) 1 [ * log * (1 * ) log(1 * ) n n n n CF N n n log n (1 n ) log(1 n )] Where n (cn)·(tn)·(un) 16 New Concept & Theory Needed (2) Knowledge Theory (Y. X. Zhong, 2000) -- Definition: Description about a class of events on their states at which the events may stay and the law by which the states may vary. -- Categorization: formal, content, value -- Representation: p (possibility), r (rationality), v (value) States x1 Possibility p1 Rationality r1 Value v1 xn pn rn vn xN pN rN vN 17 -- Measures: K(P, P*;U), K(R, R*;U), K(V,V*;U) n* log n* n log n (C P Q) n n * K ( , ; R) 1 [ * log * (1 * ) log(1 * ) n n n n CF N n n log n (1 n ) log(1 n )] Where n (pn)·(rn)·(vn) 18 -- Ecology of Knowledge Growth Empirical Knowledge Regular Knowledge Commonsense Knowledge Inherent Knowledge 19 Algorithms: Information Knowledge Information CSK-1 Base Common Sense Knowledge-1 Induction Learning Empirical Knowledge Validation/Deduction Regular Knowledge Information Popularization Common Sense Knowledge-2 20 Algorithms: Knowledge Intelligence Common Sense Knowledge Empirical Knowledge Regular Knowledge Sensor-Motor Intelligent Strategy Neural Network Intelligent Strategy Expert System Intelligent Strategy 21 All Algorithms Are Feasible and Open Algorithms for Information Knowledge Transformation -- information experience: Induction Algorithms (Data-Mining, Knowledge Discovery, …) -- old knowledge new one: Deduction Algorithms (Logic Reasoning, Rough Set Theory…) -- RK common knowledge: popularization Algorithms for Knowledge Intelligence Transformation -- Experience-Based: Neural Networks and the like -- Regular K-Based: Expert Systems -- Common K-Based: Senor-Motor Algorithms for Interfaces: All are interoperable. 22 A Unified Model of AI C.K Sensor-Motor C.K Execution Acquisition P-C-G Neural Network E.K Validation R.K I-Action Expert System R.K Popularization C.K-2 Information Knowledge Intelligence 23 Consciousness-Emotion-Intelligence CSK-Cons. Syntactic Info Retrieval Emotion Reflection CSK-Base G Conversion CI K Cognition DM I-Strategy Information K-Base 24 Implications & Open Problems 1, Instead of being contradictory among the three, AI Theory is now becoming a big and harmonious family, a unified and systematic discipline, thus gaining greater momentum. 2, As results, AI should now mean the trinity of traditional AI, neural network (Computational Intelligence) and the senormotor systems. 3, The unified theory of AI does not close the door but rather, it opens up more future works: -- Specific algorithms in all possible applications -- More Challenges: Implicit Intelligence – finding and defining problems 25 4, New Proposal -- International Studies on Advanced Intelligence 26 Should We Need To Prepare A New Platform? International Platform on Advanced Intelligence (IPAI): A New Platform for Conference on Advanced Intelligence. Advanced Intelligence: Natural Intelligence - Machine Intelligence (NN+ES+SM) Intelligence-Emotion-Consciousness-Cognition Complex System-Distributed Intelligence-Intelligent Web Basic Principles for IPAI: Freedom: For freely exchanging ideas and sharing progress. Free in and free out. Equity: All individuals are equal in IPAI. Democracy: Representatives of regions as operational body Host: in turn via Application 27 Comments Are Welcome. Thank You ! 28