Cognitive Computing 2012 The computer and the mind 3. NEWELL & SIMON PHYSICAL SYMBOL HYPOTHESIS Professor Mark Bishop.

Slides:



Advertisements
Similar presentations
Chapter 5 Development and Evolution of User Interface
Advertisements

Chapter 4: Cognitive science and the integration challenge
Heuristic Search techniques
Intro to Course and What is Learning?. What is learning? Definition of learning: Dictionary definition: To gain knowledge, comprehension, or mastery through.
Presentation on Artificial Intelligence
ARCHITECTURES FOR ARTIFICIAL INTELLIGENCE SYSTEMS
WHAT IS THE NATURE OF SCIENCE?
1 GPS, A Program That Simulates Human Thought 신 동 호신 동 호 Allen Newell & H.A. Simon.
Part2 AI as Representation and Search
Chapter 6: The physical symbol system hypothesis
Use “Search” for Pathfinding FactorySchool Library Hospital Park Newsagent University church Example from Alison Cawsey’s book start finish.
Problem Solving and Search in AI Part I Search and Intelligence Search is one of the most powerful approaches to problem solving in AI Search is a universal.
A Brief History of Artificial Intelligence
CPSC 322 Introduction to Artificial Intelligence October 6, 2004.
Chapter Seven The Network Approach: Mind as a Web.
Overview and History of Cognitive Science. How do minds work? What would an answer to this question look like? What is a mind? What is intelligence? How.
COMP 3009 Introduction to AI Dr Eleni Mangina
Models of Human Performance Dr. Chris Baber. 2 Objectives Introduce theory-based models for predicting human performance Introduce competence-based models.
Applying Multi-Criteria Optimisation to Develop Cognitive Models Peter Lane University of Hertfordshire Fernand Gobet Brunel University.
name___________________________ World of Physical Science
ARTIFICIAL INTELLIGENCE [INTELLIGENT AGENTS PARADIGM] Professor Janis Grundspenkis Riga Technical University Faculty of Computer Science and Information.
Ch1 AI: History and Applications Dr. Bernard Chen Ph.D. University of Central Arkansas Spring 2011.
1 Lyle H. Ungar, University of Pennsylvania What is AI? “Artificial Intelligence is the study of how to make computers do things at which, at the moment,
Chapter 14: Artificial Intelligence Invitation to Computer Science, C++ Version, Third Edition.
19/13/2015CS360 AI & Robotics CS360: AI & Robotics TTh 9:25 am - 10:40 am Shereen Khoja
Artificial Intelligence: Its Roots and Scope
Computer Science as Empirical Inquiry : Symbols and Search Allen Newell and Herbert A.Simon(1976) Interdisciplinary Program in Cognitive Science Lee Jung-Woo.
Cognitive Computing 2012 The computer and the mind INTRODUCTION See: Professor Mark Bishop.
A RTIFICIAL I NTELLIGENCE Introduction 3 October
Empirical Explorations with The Logical Theory Machine: A Case Study in Heuristics by Allen Newell, J. C. Shaw, & H. A. Simon by Allen Newell, J. C. Shaw,
Artificial Intelligence Introductory Lecture Jennifer J. Burg Department of Mathematics and Computer Science.
Artificial Intelligence
Game Playing Chapter 5. Game playing §Search applied to a problem against an adversary l some actions are not under the control of the problem-solver.
Psychology of Thinking: Embedding Artifice in Nature.
Introduction to search Chapter 3. Why study search? §Search is a basis for all AI l search proposed as the basis of intelligence l all learning algorithms,
Artificial Intelligence Prepared By: AIMCORP-Automated Information Management Corporation.
Science Fair How To Get Started… (
ARTIFICIAL INTELLIGENCE [INTELLIGENT AGENTS PARADIGM] Professor Janis Grundspenkis Riga Technical University Faculty of Computer Science and Information.
© 2008 The McGraw-Hill Companies, Inc. Chapter 8: Cognition and Language.
WHAT IS THE NATURE OF SCIENCE?. SCIENTIFIC WORLD VIEW 1.The Universe Is Understandable. 2.The Universe Is a Vast Single System In Which the Basic Rules.
1 Introduction to Artificial Intelligence (Lecture 1)
1 What is OO Design? OO Design is a process of invention, where developers create the abstractions necessary to meet the system’s requirements OO Design.
1 CS 385 Fall 2006 Chapter 1 AI: Early History and Applications.
1 The main topics in AI Artificial intelligence can be considered under a number of headings: –Search (includes Game Playing). –Representing Knowledge.
Introduction to Earth Science Section 2 Section 2: Science as a Process Preview Key Ideas Behavior of Natural Systems Scientific Methods Scientific Measurements.
Artificial intelligence
Automated Reasoning Early AI explored how to automated several reasoning tasks – these were solved by what we might call weak problem solving methods as.
SIMULATIONS, REALIZATIONS, AND THEORIES OF LIFE H. H. PATTEE (1989) By Hyojung Seo Dept. of Psychology.
Introduction to Artificial Intelligence CS 438 Spring 2008 Today –AIMA, Chapter 1 –Defining AI Next Tuesday –Intelligent Agents –AIMA, Chapter 2 –HW: Problem.
Course Overview  What is AI?  What are the Major Challenges?  What are the Main Techniques?  Where are we failing, and why?  Step back and look at.
Introduction to Artificial Intelligence CS 438 Spring 2008.
Lecture №1 Role of science in modern society. Role of science in modern society.
College of Computer Science, SCU Computer English Lecture 1 Computer Science Yang Ning 1/46.
Artificial Intelligence: Research and Collaborative Possibilities a presentation by: Dr. Ernest L. McDuffie, Assistant Professor Department of Computer.
Research Word has a broad spectrum of meanings –“Research this topic on ….” –“Years of research has produced a new ….”
CMPB454 ARTIFICIAL INTELLIGENCE (AI) CHAPTER 1.1 Background Information CHAPTER 1.1 Background Information Instructor: Alicia Tang Y. C.
Some Thoughts to Consider 5 Take a look at some of the sophisticated toys being offered in stores, in catalogs, or in Sunday newspaper ads. Which ones.
Artificial Intelligence Hossaini Winter Outline book : Artificial intelligence a modern Approach by Stuart Russell, Peter Norvig. A Practical Guide.
1 Solving Problems with Methods Questions. 2 Problem solving is a process similar to working your way through a maze. But what are these “steps” and what.
WHAT IS THE NATURE OF SCIENCE?
What is cognitive psychology?
Done Done Course Overview What is AI? What are the Major Challenges?
Done Done Course Overview What is AI? What are the Major Challenges?
Artificial Intelligence
Introduction Artificial Intelligent.
Artificial Intelligence introduction(2)
Psychology of Thinking: Embedding Artifice in Nature
Search.
Search.
The Network Approach: Mind as a Web
Presentation transcript:

Cognitive Computing 2012 The computer and the mind 3. NEWELL & SIMON PHYSICAL SYMBOL HYPOTHESIS Professor Mark Bishop

01/04/2014(c) Bishop: The computer and the mind2 Background The computing machine is not just hardware or software but specifies the execution of software by hardware. Computer Science is an empirical (derived from experience) discipline. Each machine is an experiment - development of knowledge through empirical enquiry. Newell and Simons paper focuses on two important examples: the symbolic system; heuristic search.

01/04/2014(c) Bishop: The computer and the mind3 Symbols and physical symbol systems All sciences are concerned with the essential nature of the systems they study. Such characterisations are usually qualitative. E.g. In biology the building block of all living things is the cell; In geology the surface of earth a collection of huge plates etc. Laws of qualitative structure are everywhere in science, usually defining the terms on which a science operates. On a fundamental level computer science can be considered an attempt to (empirically) explain what symbols really are.

22/10/2012(c) Bishop: The computer and the mind4 Ernst Cassirer: on signals and symbols Man has, as it were, discovered a new method of adapting himself to his environment. Between the receptor system and the effector system, which are to be found in all animal species, we find in man a third link which we may describe as the symbolic system. This new acquisition transforms the whole of human life. As compared with the other animals man lives not merely in a broader reality; he lives, so to speak, in a new dimension of reality. Cassirer, An Essay on Man, (1944)

01/04/2014(c) Bishop: The computer and the mind5 Cassirer: the philosophy of symbolic forms A symbol has no actual existence as part of the physical world; it has a meaning. Symbols - in the proper sense of this term - cannot be reduced to mere signals. Symbols and signals belong to two different universes of discourse: Signals are operators - a signal is part of the physical world of being and has a physical or substantial being. Symbols are designators - a symbol is part of the human world of meaning.

01/04/2014(c) Bishop: The computer and the mind6 Physical Symbol Systems, (PSS) In a physical symbol system the term physical denotes that the system is realizable in [machine] hardware and not restricted to human symbols systems. A PSS defines symbol as a set of physical patterns Such patterns are components of a symbol structure, (or expression). A symbol structure is composed of tokens related in a physical way, (e.g. by adjacency), plus processes capable of operating on such expressions to produce further expressions. Thus a physical symbol system is a set of symbol structures mechanically evolving over time.

01/04/2014(c) Bishop: The computer and the mind7 The physical symbol systems hypothesis A PSS has the necessary and sufficient means forgeneral intelligent action. It is an empirical and qualitative law. A PSS is an instance of a universal machine, hence intelligent action will be realizable by universal computer. The physical symbol system hypothesis asserts that the intelligent machine IS a PSS.

01/04/2014(c) Bishop: The computer and the mind8 Steps towards the physical symbol system Formal logic (cf. Russell & Whitehead) An attempt to purge meaning from symbols. Turing Machines and Digital Computers Universal computing machines. Stored Program as data: The UTM: Turing Machine data can equally specify a program. NB. Albeit Newell and Simon observe that in the early days no attempt was made to investigate machines with self modifying code.

01/04/2014(c) Bishop: The computer and the mind9 The designating symbol List processing (LISP): Data can now designate (indirection) Indirection gives us pointers and dynamic memory Conclusion: We needed emergence of all of the above to facilitate the emergence of the concepts of: (i) symbol manipulation; (ii) the designating symbol; and (iii) the physical symbol system.

01/04/2014(c) Bishop: The computer and the mind10 Artificial intelligence (A.I.) and psychology A.I. is concerned with constructing intelligent computer systems. Over its history lots of programs have been developed to investigate specific small [toy] aspects of intelligent action. A common principle is heuristic (short-cut; non-exhaustive) search. In Psychology, investigation of human problem solving entails the observation and modelling of human symbolic behaviour. AI programs such as GPS (the General Problem Solver) were developed to simulate human problem solving behaviour. GPS: an attempt at generality in problem solving. The core concepts of GPS derived observation of human problem solving in action.

01/04/2014(c) Bishop: The computer and the mind11 Principle evidence for PSS At the time of Newell & Simons paper there appeared to be NO serious competing hypothesis. Contrast currently with the emergence of: connectionism; embodied cognitive science; Brooks: A.I. and the subsumption architecture. Radical embodied cognitive science & dynamical system theory; enactivism etc. Newell and Simon claim that gestalt theory (on wholeness) and behaviourism are inadequate as they lack the explanatory power of the PSS. However, although the PSS hypothesis demonstrates that computer science can be construed as an empirical/experimental scientific enterprise, it does not specify how to use a PSS for intelligent action…

01/04/2014(c) Bishop: The computer and the mind12 Heuristic search hypothesis Problem solutions represented as symbol structures. A PSS exercises intelligence by search. i.e. A PSS progressively modifies the symbol structure until it produces a possible problem solution. Time is a constraint in any real PSS. Newell & Simon do not consider fully parallel search etc. Limited resource PSS and the serial PSS are synonymous.

01/04/2014(c) Bishop: The computer and the mind13 Problem solving: Plato and Meno How do you enquire Socrates into that which you know not? What will you put forth as the subject of enquiry? And if you find what you want, how will you ever know that this is what you did not know? Theory of Recollection (Plato) Knowledge is the recollection of timeless forms from before our immortal souls were imprisoned inside our bodies.

01/04/2014(c) Bishop: The computer and the mind14 Problem solving: heuristic search Establish a test for a class of symbol structures and a generator of possible solution PSSs. Repeatedly generate and test until the PSS satisfies the test. NB. This process is not like dreams; we can't immediately just generate a solution as we dont know how good a solution is until we test it. Consider chess: We can't immediately generate best positions (as if in a dream). We must painstakingly search thru all available solutions and somehow evaluate them to find 'best' (cf. Meno). Even Deep Blue can't solve perfectly for chess (and so must approximate). A PSS must use limited processing to search the space of possible solutions; an intelligent PSS makes wise choices of what to evaluate next (e.g. alpha-beta pruning)

01/04/2014(c) Bishop: The computer and the mind15 Extracting information from the problem space If solutions are randomly distributed we can do no better than random search. But in tasks which require intelligence: The expectation is that solution distributions are not random; the pattern of solutions is detectable; the generator is capable of adaptive behaviour as it discovers this pattern.

01/04/2014(c) Bishop: The computer and the mind16 Intelligence by incremental application of knowledge Consider the algebraic expression: AX + B = CX + D. We could generate solutions for X by repeated trial and error. This would be extremely unintelligent. The problem is best solved by incremental application of knowledge. Thus rearranging the above algebraic expression demonstrates intelligent problem solving. I.e. X = (D – B) / (A – C) This methodology demonstrates intelligence to find a solution without the need for either reincarnation or access to Platos perfect forms !

01/04/2014(c) Bishop: The computer and the mind17 Search trees Combinatorial explosion in search If the generator forms B branches then we get a tree ofB D tests where D is depth of search. Chess involves very broad trees which rapidly form very large search spaces. Although we note that human players only evaluate a few moves by being very selective in what branches to evaluate.

01/04/2014(c) Bishop: The computer and the mind18 Measuring intelligence The raw amount of search involved in solving a problem is not a measure of intelligence.. A problems difficulty is more accurately approximated by the amount of search that would be required without application of intelligence. The hope is periodically ignited that a chess playing program will be developed to win by brute force. As it happened IBMs Deep Blue [effectively] did just this! The need to avoid combinatorial explosion Only rarely can breadth be reduced to unity - as in the algebraic manipulation example - so typically we must deploy heuristics to guide the search and narrow the search tree.

01/04/2014(c) Bishop: The computer and the mind19 Simple heuristic search Attempts to answer the question, What is to be done next in the search process? E.g. From what node in tree do we (re) start next search? Best-fit search: examine the relative distance of each node from final solution. Means-end analysis: the search process over the problem space combines aspects of both forward and backward reasoning in that both the condition and action portions of rules are looked at when considering which rule to apply. Differences between the current and goal states are used to propose operations which reduce the differences.

01/04/2014(c) Bishop: The computer and the mind20 Weak and strong search Best Fit search evolved from Logic Theorist and Means-End Analysis evolved with GPS. For reasons related to the size of their memory, early chess programs usedDepth First search as they couldn't store large search trees in memory. It is suggested that the poor performance of A.I. systems against humans is due to humans using more SEMANTIC knowledge on very parallel hardware; Deep Blue?? Newell and Simon observe that there has been little mathematical investigation into search algorithms. This is not so true now… The above weak search methods seek to control combinatorial explosion when it cant be avoided (e.g. TSP). Strong search methods aim to eliminate it by utilizing structure in the solution domain.

01/04/2014(c) Bishop: The computer and the mind21 Intelligence without much search The use of non-local information When playing chess usually only use knowledge locally - should try to combine it to produce a global picture. eg. HEARSAY speech understanding project uses various types of search on phonemes; lexical; syntatic & semantic and combines it globally onto a blackboard. Semantic Recognition Systems Store lots of problem specific knowledge. eg. For chess might store upwards of 50K interesting patterns [Deep Blue]. Appropriate Representations Consider chess board with two diagonal corners removed. Can it be covered by exactly 31 tiles? We could evaluate (massive) search tree or observe that opposite ends of a diagonal are same colour, so cant do it.

01/04/2014(c) Bishop: The computer and the mind22 Conclusion With the development of physical symbol systems and heuristic search Newell and Simon significantly moved the process of intelligent problem solving away from Plato, at a time when intelligent thought was deemed intangible.