Introduction Artificial Intelligence Santa Clara University 2016
What is AI Definitions of AI Thinking humanly Thinking rationally Acting humanely Acting rationally
Acting Humanly Turing Test (1950) Criterion: Human interrogator cannot decide whether an agent is a computer or a human being Originally communication via typewriter Total Turing Test Computer can see so that the interrogator can test reactions to visual inputs Computer can handle objects given through a hatch
Acting Humanly Little effort on passing the Turing test in the AI community Used in arguments against the possibility of AI Chinese Room: A group of people in an enclosed room. No-one knows Chinese Interact with outside through written communication If they learn how to pass the Turing test in Chinese, where does this knowledge of Chinese reside?
Thinking Humanly Cognitive science / modeling How to get inside the human mind? introspection (Phenomenologists) experiments Fallacy: If a computer performs well on a task that humans can perform well, then it has modeled human reasoning Cross fertilization: Computer vision uses insights from cognitive science
Thinking rationally Laws of thought Around since Aristoteles' syllogisms Made more precisely by logicians in 19th -20th centuries Logicist tradition within AI uses rules and logic engines to create intelligent systems
Acting rationally Agent is one who acts Rational agent acts to achieve the best (expected?) outcome Logicist agent draws interferences Rational agent acts even if it cannot draw an interference on the best possible choice of actions
Foundations of AI Philosophy Can formal rules be used to draw valid conclusions How does the mind arise from the brain Where does knowledge come from How does knowledge lead to action Mathematics What are the formal rules to draw valid conclusions What can be computed How do we reason with uncertain information Economics How should we make decisions to maximize payoff How should we do this when others may not go along How should we do this if payoffs happen at different points in the future Neuroscience How do brains process information Psychology How do humans and animals think and act Computer Engineering How do we build efficient computers Control theory How can artifacts operate under their own control Linguistics How does language relate to behavior
Neuron Axonal arborization Axon from another cell Synapse Dendrite Axon Nucleus Synapses
Short History of AI Gestation of AI (1943-1955) McCulloch & Pitts (1943) model of artificial neurons Hebb (1949): Hebbian learning for artificial neural nets
Short History of AI Birth of AI (1956) McCarthy, Minsky, Shannon, Rochester, More, Samuel, Solomonoff, Selfridge, Newell, Simon 2 month 10 man study of AI
Short History of AI Early enthusiasm, great expectations (1952-1969) First AI programs intended as prototypes General Problem Solver (GPS) - thinking humanly Physical symbol system hypothesis: a physical symbol system has the necessary and sufficient means for general intelligent action Geometric Theorem Prover LISP language Minsky s microwords: SAINT: calculus integration problems ANALOGY: geometric analogy problems as they appear on intelligence tests STUDENT: solved algebra word problems Blocks world: Manipulate a universe of geometric blocks
Blue Red Green Red Green Blue Green Red
State of the art Robotic vehicles Speech recognition Autonomous planning and scheduling Chess playing Spam fighting Logistics planning Robotics Machine Translation
Short History of AI A dose of reality (1966-1973) Early predictions did not come through E.g. Russian translation program turned out to be much more complex: The spirit is willing but the flesh is weak transformed into The vodka is good but the meat is rotten Problems are not scalable Early genetic algorithms could not improve a computer program for the available CPU hours
Short History of AI Knowledge based systems (1969-1979) Weak methods: Applicable to general situations, but do not scale to problem size Alternative: use domain-specific knowledge DENDRAL: Inferring molecular structure from information provided by a mass spectrometer Expert systems MYCIN medical expert system with ~450 rules could outperform junior doctors
Short History of AI Industrial uses of AI (1980 - present) Boom from 1980-1989 AI Winter 1989-1990 as companies could not deliver on extravagant promises
Short History of AI Return of neural networks (1986 - present) Back propagation is a new learning algorithm Replaces symbolic models 2015: Deep neural networks
Short History of AI AI adopts the scientific method Example: Speech recognition Early attempts are ad hoc Hidden Markov Models (HMM) based on a mathematical theory use large corpus of speech data
Short History of AI Emergence of intelligent agents (1995 - present)