ISICA2007. Perspective On Intelligence Science. Zhongzhi Shi. Institute of Computing Technology Chinese Academy of Sciences

Transcription

1 ISICA2007 Perspective On Intelligence Science Zhongzhi Shi Institute of Computing Technology Chinese Academy of Sciences

2 Contents Introduction Neuron models Ion channel model Human Brain Imaging Mind models Conclusions 9/22/2007 Zhongzhi Shi Intelligence Science 2

3 Evolution of Computers 9/22/2007 Zhongzhi Shi Intelligence Science 3

4 Big Challenge Computer: High performance Low intelligence Cite from Scientific American,2005(3) 9/22/2007 Zhongzhi Shi Intelligence Science 4

5 Human-Level AI The long-term goal of Artificial Intelligence is human-level Artificial Intelligence. 9/22/2007 Zhongzhi Shi Intelligence Science 5

6 Intelligence Science Researches on basic theory and technology of intelligence. Intelligence science is an interdisciplinary subject mainly including brain science, cognitive science, artificial intelligence and others. Brain science explores the essence of brain, research on the principle and model of natural intelligence in molecular, cell and behavior level. 9/22/2007 Zhongzhi Shi Intelligence Science 6

7 Intelligence Science Cognitive science studies human mental activity, such as perception, learning, memory, thinking, consciousness etc. In order to implement machine intelligence, Artificial intelligence attempts simulation, extension and expansion of human intelligence using artificial methodology and technology 9/22/2007 Zhongzhi Shi Intelligence Science 7

8 Big Issues in Intelligence Science Signaling in the Nervous System Synaptic Plasticity Perceptual Representation Learning Emergence Coding and Retrieval of Memory Linguistic Cognition Formalizing of Commonsense knowledge and Reasoning Computational Instruction Set of Human Cognition Nature of Consciousness Architecture of Intelligent Systems 9/22/2007 Zhongzhi Shi Intelligence Science 8

9 Contents Introduction Neuron models Ion channel model Human Brain Imaging Mind models Conclusions 9/22/2007 Zhongzhi Shi Intelligence Science 9

10 Signaling in the Nervous System neurons synapses cycle time: 10-3 sec 9/22/2007 Zhongzhi Shi Intelligence Science 10

11 Synaptic Plasticity Hebb s Rule Connection Weights ~ Correlations ``When one cell repeatedly assists in firing another, the axon of the first cell develops synaptic knobs (or enlarges them if they already exist) in contact with the soma of the second cell. (Hebb, 1949) Short-Term Changes in Signaling Long-Term Changes in Signaling 9/22/2007 Zhongzhi Shi Intelligence Science 11

12 Synaptic Plasticity Electrical Synapse 9/22/2007 Zhongzhi Shi Intelligence Science 12

13 Synaptic Plasticity Chemical Synapse 9/22/2007 Zhongzhi Shi Intelligence Science 13

14 McCulloch-Pitts Neuron Model 9/22/2007 Zhongzhi Shi Intelligence Science 14

15 Perceptual Neuron Model x 0 = +1 w 0 v w m = 0 = j = 0 b w jx j Input signal x 1 x 2 M w 1 w 2 M Local Field Summing function Activation function v ϕ( ) Output y x m w m Synaptic weights 9/22/2007 Zhongzhi Shi Intelligence Science 15

16 Hopfield Networks 1982 年,Hoopfield 1984 年,Hoopfield,Tank 9/22/2007 Zhongzhi Shi Intelligence Science 16

17 Neural Computing ART Kohonen: SOM Hinton: Helmboltz Machine Lei Xu: Ying-Yang Machine S.Amari: Information Geometry 9/22/2007 Zhongzhi Shi Intelligence Science 17

18 Neural Computing 9/22/2007 Zhongzhi Shi Intelligence Science 18

19 Contents Introduction Neuron models Ion channel model Human Brain Imaging Mind models Conclusions 9/22/2007 Zhongzhi Shi Intelligence Science 19

20 Signaling in the Nervous System Ion Channels and Signaling 9/22/2007 Zhongzhi Shi Intelligence Science 20

21 Signaling in the Nervous System Transport across Cell Membranes 9/22/2007 Zhongzhi Shi Intelligence Science 21

22 Signaling in the Nervous System Ionic Basis of the Action Potential 9/22/2007 Zhongzhi Shi Intelligence Science 22

23 Signaling in the Nervous System How: - AP comes down the axon. - At the synapses, VG Ca++ channels let in calcium. - This triggers the release (exocytosis!) of the contents of vesicles in the axonal boton. - The contents are: NEUROTRANSMITTERS. 9/22/2007 Zhongzhi Shi Intelligence Science 23

24 Signaling in the Nervous System Transmitter Release How are neurotransmitters packaged and released (signal transduction CaM pathway, second messenger, transporters, receptors)? Ca 2+ VGCC Ca 2+ -CaM K +ATP Ca 2+ -CaMK P P P P.. 9/22/2007 Zhongzhi Shi Intelligence Science 24 P P P P -P -P

25 Ion channel based model justify neurons signals computational model molecules ion channels 9/22/2007 Zhongzhi Shi Intelligence Science 25

26 Hodgkin-Huxley Model C C du dt 100 = mv g K 0 I I Na g Na g l I K Ion channels Ion pump Ileak 3 4 gna m h( u ENa) + gkn ( u EK ) + gl ( u El ) + inside Ka Na outside stimulus I( t) dh dm dn hn m h nm0( u) u) = dt dt ( u () u ) τ h n τ m 9/22/2007 Zhongzhi Shi Intelligence Science 26

27 Hodgkin-Huxley Model pulse input I(t) inside Ka C du dt = 3 4 g Na m h( u ENa) + gkn ( u EK ) + gl ( u El ) + dh dm dn hn m h nm0( u) u) = dt dt ( u () u ) τ h n τ m I Na stimulus I( t) 9/22/2007 Zhongzhi Shi Intelligence Science 27 I K Ion channels u Ion pump Ileak m 0 (u) τ h (u) h 0 (u) τ m (u) Na outside u

28 Hodgkin-Huxley Model refractoriness 100 Action potential mv 0 ϑ 0 ms 20 Strong stimulus strong stimuli 9/22/2007 Zhongzhi Shi Intelligence Science 28

29 Model of fast spiking interneuron Spike I(t) I C g Na g Kv1 g Kv3 g l Subthreshold Detailed model, based on ion channels 9/22/2007 Zhongzhi Shi Intelligence Science 29

30 Model of fast spiking interneuron Current pulse Detailed model, based on ion channels constant current I C g Na g Kv1 g Kv3 g l 9/22/2007 Zhongzhi Shi Intelligence Science 30

31 Contents Introduction Neuron models Ion channel model Human Brain Imaging Mind models Conclusions 9/22/2007 Zhongzhi Shi Intelligence Science 31

32 Human Brain Imaging 9/22/2007 Zhongzhi Shi Intelligence Science 32

33 9/22/2007 Zhongzhi Shi Intelligence Science 33

34 9/22/2007 Zhongzhi Shi Intelligence Science 34

35 Virtual Sensor COGNITIVE TASK Cognitive state sequence 1. Does fmri contain enough information? Virtual sensors of cognitive state 2. Can we devise learning algorithms to construct such virtual sensors? 9/22/2007 Zhongzhi Shi Intelligence Science 35

36 9/22/2007 Zhongzhi Shi Intelligence Science 36

37 Cognitive Processing 9/22/2007 Zhongzhi Shi Intelligence Science 37

38 Training Classifiers of Cognitive State 9/22/2007 Zhongzhi Shi Intelligence Science 38

39 Training Classifier for Word Categories 9/22/2007 Zhongzhi Shi Intelligence Science 39

40 Six Categories Study 9/22/2007 Zhongzhi Shi Intelligence Science 40

41 Mental States Recognition 9/22/2007 Zhongzhi Shi Intelligence Science 41

42 ACT-R Experimentally grounded Performs experiment similar to human subjects Detailed and precise accounting of the data Predictions about every aspect of experiment E.g. errors and latency Learnable through experience Complex Phenomena Principled parameters Neurologically plausible 9/22/2007 Zhongzhi Shi Intelligence Science 42

43 ACT-R History Predecessor HAM (Anderson & Bower 1973) Theory versions ACT-E (Anderson, 1976) ACT* (Anderson, 1978) ACT-R (Anderson, 1993) ACT-R 4.0 (Anderson & Lebiere, 1998) ACT-R 5.0 (Anderson & Lebiere, 2001) Implementations GRAPES (Sauers & Farrell, 1982) PUPS (Anderson & Thompson, 1989) ACT-R 2.0 (Lebiere & Kushmerick, 1993) ACT-R 3.0 (Lebiere, 1995) ACT-R 4.0 (Lebiere, 1998) ACT-R/PM (Byrne, 1998) ACT-R 5.0 (Lebiere, 2001) Windows Environment (Bothell, 2001) Macintosh Environment (Fincham, 2001) ACT-R 6.0 (Bothell, 2004??) 9/22/2007 Zhongzhi Shi Intelligence Science 43

44 ACT-R 6.0 Goal Imaginal imaginal-action goal Declarative imaginal visual-location visual Vision manual Motor Productions Match Select Execute vocal Speech retrieval aural-location aural Aural 9/22/2007 Zhongzhi Shi Intelligence Science 44 ENVIRONMENT

45 Mental Algebra Task (Anderson et. 2002) 9/22/2007 Zhongzhi Shi Intelligence Science 45

46 Activity Predicted by ACT-R 9/22/2007 Zhongzhi Shi Intelligence Science 46 (Anderson et. 2002)

47 9/22/2007 Zhongzhi Shi Intelligence Science 47

48 Learned Bayes Model 9/22/2007 Zhongzhi Shi Intelligence Science 48 P(BrainActivity WorldCategory=People)

49 Scientific Revolution 9/22/2007 Zhongzhi Shi Intelligence Science 49

50 Contents Introduction Neuron models Ion channel model Human Brain Imaging Mind models Conclusions 9/22/2007 Zhongzhi Shi Intelligence Science 50

51 Architecture of Intelligent Systems Mind is here defined as our thinking faculties as well as the arena where we process, store, and manage mental activity. In humans, mind is a faculty and/or location where imagination and intellect, image and idea, percept and concept, body and soul, matter and spirit contact, interact, and intermingle. 9/22/2007 Zhongzhi Shi Intelligence Science 51

52 Mind Development Mind Development Simple condition reflection Condition reflection Tool employment Linguistic symbol 9/22/2007 Zhongzhi Shi Intelligence Science 52

53 Alan Turing On computable numbers with an application to the Entscheidungsproblem (1936) Church, Kleene, Post Alan Turing ( ) 9/22/2007 Zhongzhi Shi Intelligence Science 53

54 Turing Machine T A C A G C T C G 1 - A C G T 0 HALT HALT HALT HALT HALT 1 -,<=,0 A,=>,1 C,=>,1 G,=>,2 T,=>,1 2 -,<=,0 A,=>,1 C,<=,3 G,=>,2 T,=>,1 3 T,=>,4 4 A,=>,1 9/22/2007 Zhongzhi Shi Intelligence Science 54 Replaces GC with TA

55 Turing Test Turing s 1950 paper in Mind, Computing Machinery and Intelligence: 'imitation game', in which a human being and a computer would be interrogated under conditions where the interrogator would not know which was which, the communication being entirely by textual messages. Turing argued that if the interrogator could not distinguish them by questioning, then it would be unreasonable not to call the computer intelligent. 9/22/2007 Zhongzhi Shi Intelligence Science 55 Replaces GC with TA

56 Physical Symbol System Hypothesis Newell and Simon's paper "Computer Science as Empirical Inquiry: Symbols and Search" which define the essential ideas of the Physical Symbol System Hypothesis: "A physical symbol system has the necessary and sufficient means for intelligent action." 5) 9/22/2007 Pinker, Steven, 1997, How the Mind Works, W.W. Zhongzhi Norton, Shi New Intelligence York, NY, Science pp

57 SOAR Match Rule Memory Procedural Knowledge Long-term Knowledge Act Conflict Resolution Working Memory Declarative Knowledge Short-term Knowledge 9/22/2007 Zhongzhi Shi Intelligence Science 57 This slide borrowed from John Laird

58 SOAR Computational Structure Encode knowledge Access encodings Produce actions Achieve goals 9/22/2007 Zhongzhi Shi Intelligence Science 58

59 Processing Input Propose Operator Compare Operators Select Operator Apply Operator Output If cell in direction <d> is not a wall, --> propose operator move <d> If operator <o1> will move to a empty bonus food cell and operator <o2> --> will move to a normal food, operator --> <o1> < operator <o1> > <o2> If an operator is selected to move <d> --> create output move-direction <d> Production Memory East North South movedirection North North > East South < > East North = South Working Memory 9/22/2007 Zhongzhi Shi Intelligence Science 59 This slide borrowed from John Laird

60 Processing Input Propose Operator Compare Operators Select Operator Apply Operator Output Tie Impasse East North South North > East South > East North = South Chunking creates rules that create preferences based on what was tested Evaluate-operator (North) = 10 Evaluate-operator = 10 Evaluate-operator = 5 (South) (East) North = 10 Chunking creates rule that applies evaluate-operator 9/22/2007 Zhongzhi Shi Intelligence Science 60 = 10 This slide borrowed from John Laird

61 Society of Mind M. Minsky 9/22/2007 Zhongzhi Shi Intelligence Science 61

62 Society of Mind World A+B World A+B Client Client Mind M A+B Mind M A+B World A World A World B World B Mind M A Mind M A Mind M B Mind M B Mind Mind Mind Mind Mind Mind Mind Mind Mind Mind Mind Mind 9/22/2007 Zhongzhi Shi Intelligence Science 62

63 Constructing complex minds, online Ciarán O Leary Dublin Institute of Technology 22nd May 2003 Technology for automated assessment: The World-Wide-Mind 9/22/2007 Zhongzhi Shi Intelligence Science 63

64 Constructing complex minds, online Mind Mind Server Server Mind Mind M Server Server World World Server Server Client Client Mind Mind Server Server 9/22/2007 Zhongzhi Shi Intelligence Science 64

65 IDA s Architecture(Stan Franklin ) Metacognition Database Perception Linear Functional Deliberation Negotiation Write Orders Behavior Net Perception Consciousness Conceptual & Behavioral Learning Associative Memory Episodic Memory Emotions 9/22/2007 Zhongzhi Shi Intelligence Science 65

66 IDA: an Intelligent Distribution Agent (Stan Franklin ) Modules and Mechanisms Perception Copycat Architecture Hofstadter Action Selection Behavior Net Maes Associative Memory Sparse Distributed Memory Kanerva Episodic Memory Case-based Memory Emotions Pandemonium Theory Jackson Metacognition Fuzzy Classifier Systems Holland, Zadeh Learning Copycat Architecture, Case-based Reasoning Constraint Satisfaction Linear Functional Language Generation Pandemonium Theory Deliberation Pandemonium Theory Consciousness Pandemonium Theory 9/22/2007 Zhongzhi Shi Intelligence Science 66

67 Consciousness Model Cognitive System Perception Learning Memory Thought Goal 感知 Attention Working Memory Awareness Consciousness Consciousness System 9/22/2007 Zhongzhi Shi Intelligence Science 67

68 Consciousness Machine 9/22/2007 Zhongzhi Shi Intelligence Science 68

69 Perspective Intelligence Science Researches on basic theory and technology of intelligence New concept New theory New Method New Technology Will get very important achievement in the 21 century. 9/22/2007 Zhongzhi Shi Intelligence Science 69

70 Thank You Intelligence Science 9/22/2007 Zhongzhi Shi Intelligence Science 70