10/4/10. An overview using Alan Turing s Forgotten Ideas in Computer Science as well as sources listed on last slide.

Transcription

1 Well known for the machine, test and thesis that bear his name, the British genius also anticipated neural- network computers and hyper- computation. An overview using Alan Turing s Forgotten Ideas in Computer Science as well as sources listed on last slide (23 June): Birth, Paddington, London : Sherborne School 1930: Death of friend Christopher Morcom : Undergraduate at King's College, Cambridge University : Quantum mechanics, probability, logic 1935: Elected fellow of King's College, Cambridge 1936: The Turing machine, computability, universal machine : Princeton University. Ph.D. Logic, algebra, number theory : Return to Cambridge. Introduced to German Enigma cipher machine : The Bombe, machine for Enigma decryption : Breaking of U- boat Enigma, saving battle of the Atlantic : Chief Anglo- American crypto consultant. Electronic work. 1945: National Physical Laboratory, London 1946: Computer and software design leading the world : Programming, neural nets, and artificial intelligence 1948: Manchester University 1949: First serious mathematical use of a computer 1950: The Turing Test for machine intelligence 1951: Elected FRS. Non- linear theory of biological growth 1952: Arrested as a homosexual, loss of security clearance : Unfinished work in biology and physics 1954 (7 June): Death (suicide) by cyanide poisoning, Wilmslow, Cheshire. 2009: Turing is awarded an official public apology by British Prime minister 2 The Bombe Machine for Enigma Decryption Connectionism computing with neural networks of artificial neurons. Modern digital computers Universal Turing Machines Artificial Intelligence and the Turing Test Hyper- computation - Theoretical concept of hyper- computers. Artificial Life - computations of chemistry of biological growth or Morphogenesis. 3 1

2 The Bombe employed what in modern AI are called heuristics to speed up the search. A heuristic is a rule that cuts down the amount of searching required. A heuristic is not necessarily guaranteed to produce the correct solution, but works often enough to be useful. 4 Intelligent Machinery became the first manifesto of AI. Connectionism computing with neural networks of artificial neurons. Frank Rosenblatt 1957 is inaccurately considered the founder of connectionism. 5 Simulating Neurons and interconnections with ordinary computer. A training algorithm that runs and adjusts interconnections of neurons make the program a general purpose dedicated machine. Ex: International currency models Ex: models of aircraft wings and skyscrapers. Ex: International currency models 6 2

3 The first manifesto of AI- Intelligent Machinery (1,6,8) 1948 I propose to investigate the question as to whether it is possible for machinery to show intelligent behavior. It is usually assumed without argument that It is not possible. In so far as a machine can show intelligence this is to be regarded as nothing but a reflection of the intelligence of its creator 7 Intelligent Machinery A- type and B- type machines are Boolean networks made up of extremely simple, randomly interconnected NAND gates (i.e., neurons), each having exactly two inputs (i.e., synapses) from other neurons. The neurons are synchronized with a global clock signal. Unlike A- type networks, Turing's B- type networks have modifiable interconnections (basically a switch) and thus an external agent can "organize" these machines (by enabling and disabling connections) to perform a required job. The idea behind the introduction of B- type networks was to open the possibility of reinforcing successful and useful links and of cutting useless ones. (8) Intelligence Defined(6) 1948 A Type unorganized machine- B- Type unorganized machine- self- modifying system P- Type unorganized machine- pleasure pain system 8 Intelligent Machinery Turing Introduces: Central concepts to Artificial Intelligence and Fundamental concepts of connectionism Belgmont G. Farley and Wesley Clark ran the first computer simulation of a small neural network at MIT. Intelligence Defined(6) 1948 Defines Intelligence as emotional and so the computer is not intelligent and thus carries out instruction in a unintelligent manner. Discusses the mechanical operation, memory and the repetition. 9 3

4 10 Artificial neurons and devices that modify connections between them which enables training. Contain any number of neurons connected in any pattern. Are subject to restriction that each neuron- neuron connection must pass through a modifier type device. 11 All modifiers have two training fibers Pass mode input as 0 or 1 pass through unchanged. Interrupt mode A pulse on the other fiber places the modifier in interrupt mode where the output is always 1. The state modifier always destroys all information attempting to pass along the connection which it is attached. 12 4

5 Contain random inter- neural connections whose modifiers have been set by chance to either pass or fail. Connections are destroyed by switching attached modifiers to interrupt mode. Connections are created by switching a modifier from interrupt to pass. Result: Initially random network becomes organized Direct applications of Turing s models are used in modern day memory and are building blocks of circuit design. 15 5

6 Top down Cognition includes complex sequential processes such as symbolic representation, mathematical calculation and language. Bottom up Cognition is made of simple firings of neurons. Viewed top down approach solves the problem of human cognition because the cortex can be viewed as a neural network acting as a general purpose computer. NAND neurons and interconnecting fibers are simplest possible model of cortex. 16 SYSTEM REQUIREMENTS MODERN DIGITAL MEMORY Stores both program and data and the scanner moves back and forth through the memory, symbol by symbol, reading and writing additional symbols. Complexity comes from chaining large numbers of basic instructions. 17 Turing Machine 1936 Universal Turing Machine 1936 Universal Turing Machine and O machine

7 Terminology Definitions(1,13) Universal Un- computable Hyper- machine Unpredictability O- Machine Limitless memory. Machine capable of any calculation that a mathematician working in accordance with some algorithmic method could perform given unlimited time Processing too difficult for a Turing machine 1936 computes the un- computable. Turing machine cannot predict if the code will run forever. No software can determine every bug. Black box or oracle serving as a mechanism to compute the un- computable. 19 Universal Turing Machine with an O machine. A digitally encoded program is fed in, and the machine produces output using a multi- step procedure of repetitious machine operations. These operations will include passing data to and from the oracle. Turing gave no indication how the oracle might work. B- machine can compute the un- computable provided the networks are desynchronized. 20 INCORRECT Turing demonstrated in the mid s that hyper- machines are impossible. Here he is credited with enunciating a principle to the effect that a universal Turing machine can exactly simulate behavior of any other information processing machine. The implication is that no other machine can carry out information processing beyond the scope of a Universal Turing machine. CORRECTED Church and Turing claimed only that a universal Turing machine can match the behavior of any human mathematician working with paper and pencil in accordance with an algorithmic method. Claim doesn t rule out possibility of hypermachines. 21 7

8 22 In fact, the whole point of Turing's oracle is that it describes a non- mechanical logical element. 23 Non- linear differential equations used in expressing the chemistry of growth. Simulation of a chemical mechanism by which the genes of a fertilized egg cell may determine the anatomical structure of a resulting animal or plant. Considered not unrelated to neural networks because of structure of genetical embryological mechanism. 24 8

9 The theory (which can be called a reaction diffusion theory of morphogenesis), has served as a basic model in theoretical biology, [1] and is seen by some as the very beginning of chaos theory. [2] Reaction diffusion systems have attracted much interest as a prototype model for pattern formation. 25 (1) %27s_Forgotten_Ideas.pdf (2) (3) (4) (5) (6) %3Fid%3D127%26title%3D1 (7) (8) (9) (10) (11) (12) (13) (14)