Transcript Document
Acting Humanly: The Turing test (1950) “Computing machinery and intelligence”: Can machine’s think? or Can machines behave intelligently? An operational test for intelligent behavior: the Imitation Game Predicted that by the year 2000, a machine would have a 30% chance of fooling a lay person for 5 minutes Anticipated all major arguments against AI in the following 50 years Suggested major components of AI: knowledge, reasoning, language, understanding, learning. Problem: Turing test is not reproducible, constructive, or amenable to mathematical analysis. Intelligence not determinable by surface behavior alone. The test is not sufficient since the behaviors under adjudication are too limited. As a sufficient condition for intelligence, the test is so difficult as to be uninteresting. Consciousness - The Chinese Room Experiment – Does running the right program generate consciousness? • • • • • • • 1. 2. 3. 4. Human – only understands English Rule book – written in english Stacks of paper – some blank, some with indecipherable symbols on them Small opening to outside world Pieces of paper with symbols on them are passed through the opening The human follows the instructions in the rule book Eventually the human hands a piece of paper with symbols on it through the opening Certain kinds of objects are incapable of conscious understanding The human, paper, and rule book are objects of this kind If each object is incapable, the entire whole is incapable Therefore there is no conscious understanding in the room The Brain Prosthesis Experiment Replace neurons in your brain one at a time with artificial neurons that *exactly* replicate the behavior of the original neurons (then reverse the process). By definition, the subjects external behavior must remain unchanged. What happens? We have two choices, either 1. The causal mechanisms involved in consciousness in the electronic brain are still functioning, and it is therefore conscious. 2. Conscious mental events in the normal brain have no effect on behavior. If neuron replacement is conscious, replacing brain with an entire circuit/lookup table that mapped inputs to outputs *must* also be conscious. Potted history of AI 1943 McCulloch & Pitts: Boolean circuit model of brain 1950 Turing's ``Computing Machinery and Intelligence'' 1952--69 Look, Ma, no hands! 1950s Early AI programs, including Samuel's checkers program, Newell & Simon's Logic Theorist, Gelernter's Geometry Engine 1956 Dartmouth meeting: ``Artificial Intelligence'' adopted 1965 Robinson's complete algorithm for logical reasoning 1966--74 AI discovers computational complexity Neural network research almost disappears 1969--79 Early development of knowledge-based systems 1980--88 Expert systems industry booms 1988--93 Expert systems industry busts: ``AI Winter'' 1985--95 Neural networks return to popularity 1988-Resurgence of probabilistic and decision-theoretic methods Rapid increase in technical depth of mainstream AI ``Nouvelle AI'': ALife, GAs, soft computing Agents Anything which can be viewed as perceiving environment through sensors, etc. and then acting in the environment Current hot buzz-word Looks like the basic computational box Abstractly, an agent is a function from percept histories to actions: f : P* A For any given class of environments and tasks, we seek the agent (or class of agents) with the best (possible) performance (a rational agent). Intelligent Agents Intelligent (rational) agent seeks to maximize its performance measure for any given sequence of percepts Look up table? Text uses intelligent agent approach to bring all aspects of AI into one. What should an intelligent agent have? An intelligent agent should have knowledge, infer, plan, reason with uncertainty, learn, perceive, communicate, etc. What is rational for an agent? It depends on: 1. 2. 3. 4. The performance measure What it has perceived Its current store of knowledge The actions the agent can perform Agent Types Reflex Agent - Actions based only on current percepts (no state memory), condition-action rules Agents with Memory - keep track of internal state, past actions (or their effects), and the dynamically changing environment Goal-Based Agents - Actions driven by overall goal, easy if one step, multi-action sequences (subgoals) often supported by search and planning mechanisms Utility-Based Agents - Best actions - Multiple ways to reach goals - Conflicting Goals - Actions with uncertainty - which approach gives best chance of fulfilling goals PAGE Automated taxi driver: Percepts? Actions? Goals? Environment? PAGE Internet shopping agent Percepts? Actions? Goals? Environment? Environment Issues Accessibility - can agent detect all relevant percepts Determinism - is next state completely determined by current state plus the agent action - if inaccessible, then may appear nondeterministic regardless Episodic - Is environment neatly divided into independent episodes Static vs. Dynamic - Does environment remain static in between agent actions Discrete vs. Continuous - Are there limited distinct percept and action possibilities