What is AI?
AI Sources Philosophy Mathematics Psychology Economics Linguistics Neuroscience Control theory
Approaches: What does in mean to be AI? THINK like HUMANS THINK rationally ACT like HUMANS ACT rationally
History of the field 1870’s - Ada wrote that machines might compose music and do other creative things – but not think. 1943 McCulloch & Pitts – a neural model 1948 Cybernetics (Norbert Wiener) 1950 Turing’s “Computing Machinery and Intelligence” The Test 1950’s Early AI Samuel’s checkers Newell and Simon’s Logic Theorist Gelernter’s Geometry Engine 1956 Dartmouth meeting
History (cont.) 1962 Perceptrons (Rosenblatt) 1966-74 AI encounters computational complexity 1969-1979 Early knowledge based approaches 1980-88 Expert Systems and the Japanese 5’th Generation – boom years 1988-93 Expert System bust – “AI winter” 1985-95 Return of neural systems Parallel Distributed Processing – (Rummelhart and McCelland 1986)
History (cont.) 1988- 1995 - Agents Resurgence of probability “Nouvelle AI” Alife Genetic Algorithms 1995 - Agents
Stan Franklin’s Three Questions Is AI possible? Can a machine ‘think”? Is AI necessarily symbolic? Neural Systems Is representation necessary? (“Nouvelle AI”) Reference: “Artificial Minds” by Stan Franklin Publisher: MIT Press ISBN: 0262561093
Fundamental Problems! “TOY” problems Solutions that don’t scale Solving the wrong problem Over hype!
Critics Mind–Body problem Herbert Dreyfus Roger Penrose John Searle The Moralists From the field itself!
Herbert Dryfus “If background understanding is indeed a skill and if skills are based on whole patterns and not on rules, we would expect symbolic representations to fail to capture our commonsense understanding”
Roger Penrose Views mind as having Quantum Theory Effects Does not think a consciousness will emerge from a computer program.
Searle’s Chinese Room Questions Story Script Searle Chinese Room (English) Searle (Does not know Chinese) Chinese Room Answer (Chinese)
The Moralists Weizenbaum – Jaron Lanier Bill Joy Eliza “Computer Power and Human Reason” Jaron Lanier Bill Joy
David Marr’s Comment Two problem types in AI as a study of information processing problems Type 1 – separates into two parts Theory that explains the problem solution method Implementation (algorithm) Type 2 – lacks Type 1 theory May require many interacting parts with no simpler description Within this view – the field has no progressed much! Mimicry versus Exploration
Agents
Agents Agents imply an environment + entity Robots are agents, so are web crawlers Agents provide a excellent tool for study Environments: Artificial Real World A taxonomy of agents
Environment Observable Deterministic vs. Stochastic Episodic vs. Sequential Static vs. Dynamic Discrete vs. Continuous Single vs. Multiple agents
Simple Reflex Agents
An Agent System Rossum’s Playhouse Projects Simulation Java based abstraction of the world of a Robot Java based Client – Server System World is a Room Projects to build agents using AI techniques
What do we put inside the agent? Logic Representational systems Frames Semantic nets Neural Nets
Readings “The Cambridge Quintet” by John L. Casti “Arguing A.I. The Battle for Twenty-first Century Science” by Sam Williams