Course Info Instructor: Pascal Poupart Email: - PDF document

Course Info •Instructor: Pascal Poupart – Email: cs486@students.cs.uwaterloo.ca CS 486/686 – Office Hours: TBA (watch Web page), by appt. Artificial Intelligence •Lectures: Tue & Thu – Sect. 1: 08:30-09:50 (RCH306) May 3rd, 2005 – Sect. 2: 11:30-12:50 (MC2054) •Textbook: Artificial Intelligence: A Modern University of Waterloo Approach (2 nd Edition) , by Russell & Norvig •Website – http://www.students.cs.uwaterloo.edu/~cs486 1 2 cs486/686 Lecture Slides (c) 2005 K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart Outline Artificial Intelligence (AI) Webst er says: a. t he capacit y •What is AI ? t o acquire and apply knowledge. b. t he f acult y of •What is intelligence ? t hought and reason. … • What is AI? (Chapter 1) •What features/abilities do humans (animals? • Rational agents (Chapter 2) animate objects?) have that you think are • Some applications indicative or characteristic of intelligence? • Course administration • abstract concepts, mathematics, language, problem solving, memory, logical reasoning, emotions, morality, ability to learn/adapt, etc… 3 4 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart Some Definitions (Russell & Norvig) Some Definitions (Russell & Norvig) The exciting new effort to make The study of mental faculties through the computers that think… machines with use of computational models minds in the full and literal sense [Charniak & McDermott 85] [Haugeland 85] The study of computations that make it [The automation of] activities that we possible to perceive, reason and act associate with human thinking, such as Syst ems t hat Syst ems t hat [Winston 92] decision making, problem solving, t hink like humans t hink rat ionally learning [Bellman 78] The art of creating machines that perform A field of study that seeks to explain and functions that require intelligence when emulate intelligent behavior in terms of Syst ems t hat act Syst ems t hat act performed by a human [Kurzweil 90] computational processes [Schalkoff 90] like humans rat ionally The study of how to make computers do The branch of computer science that is things at which, at the moment, people concerned with the automation of are better [Rich&Knight 91] intelligent behavior [Luger&Stubblefield93] 5 6 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart 1

What is AI? What is AI? • Systems that behave like humans – Turing (1950) “Computing machinery and intelligence” • Systems that think like humans – Cognitive science – Fascinating area, but we will not be covering it in this course • Systems that think rationally – Predicted that by 2000 a computer would have a 30% – Aristotle: What are the correct thought chance of fooling a lay person for 5 minutes processes – Anticipated all major arguments against AI in the – Systems that reason in a logical manner following 50 years – Systems doing inference correctly – Suggested major components of AI: knowledge, reasoning, language understanding, learning 7 8 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart Intelligent Assistive Technology What is AI? • Let’s facilitate aging in place • Systems that act rationally – Rational behavior: “doing the right thing” • Intelligent assistive technology – Rational agent approach – Non-obtrusive, yet pervasive • Agent: entity that perceives and acts – Adaptable • Rational agent: acts so to achieve best outcome • Benefits: – This is the approach we will take in this course – Greater autonomy • General principles of rational agents • Components for constructing rational agents – Feeling of independence 9 10 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart COACH project COACH project System Overview • Automated prompting system to help elderly persons wash their hands planning • Collaborators: Szymon Wartak, Geoff Fernie, Alex sensors Mihailidis, Jennifer Boger, Jesse Hoey and Craig Boutilier hand verbal washing cues 11 12 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart 2

Video Clip #1 Video Clip #2 Video Clip #1 Video Clip #2 13 14 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart Topics covered A brief history of AI • Search – Uninformed and heuristic search • 1943-1955: Initial work in AI – CSP’s and optimization – McCulloch and Pitts produce boolean model of the brain – Game playing • Reasoning under uncertainty – Turing’s “Computing machinery and intelligence” – Probability theory, utility theory and decision theory • Early 1950’s: Early AI programs – Bayesian networks and decision networks – Samuel’s checker program, Newell and Simon’s Logic – Multi-agent systems Theorist, Gerlenter’s Geometry Engine • Learning • 1956: Happy birthday AI! – Decision trees, neural networks, ensemble learning, reinforcement learning – Dartmouth workshop attended by McCarthy, Minsky, • Specialized areas Shannon, Rochester, Samuel, Solomonoff, Selfridge, Simon and Newell – Natural language processing, computational vision and robotics 15 16 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart A brief history of AI A brief history of AI • 1950’s-1969: Enthusiasm and expectations • 1969-1979: Knowledge-based systems – Many successes (in a limited way) – LISP, time sharing, Resolution method, neural • 1980-1988: Expert system industry booms networks, vision, planning, learning theory, Shakey, • 1988-1993: Expert system busts, AI Winter machine translation,… • 1986-present: The return of neural networks • 1966-1973: Reality hits • 1988-present: – Early programs had little knowledge of their subject matter – Resurgence of probabilistic and decision-theoretic methods • Machine translation – Computational complexity – Increase in technical depth of mainstream AI – Negative result about perceptrons - a simple form of – Intelligent agents neural network 17 18 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart 3

Rational Agents Agents and Environments • Recall: A rational agent “does the right thing” • Performance measure – success criteria sensors – Evaluates a sequence of environment states percept s environment ? • A rational agent chooses whichever action agent maximizes the expected value of its performance act ions measure given the percept sequence to date act uat ors – Need to know performance measure, environment, possible actions, percept sequence Agent s include humans, robot s, sof t bot s, t hermost at s… The agent f unct ion maps percept s t o act ions f :P* � A • Rationality ≠ Omniscience, Perfection, Success The agent program runs on t he physical archit ect ure t o produce f • Rationality � exploration, learning, autonomy 19 20 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart PEAS Properties of task environments • Specify the task environment: – Performance measure, Environment, Actuators, Sensors • Fully observable vs. partially observable Example: COACH syst em • Deterministic vs. stochastic Perf M: t ask complet ion, t ime t aken, amount of int ervent ion • Episodic vs. sequential Envir: Bat hroom st at us, user st at us Act u: Verbal prompt s, CallCaregiver, DoNot hing • Static vs. dynamic Sens: Video cameras, microphones, t ap sensor • Discrete vs. continuous Example: Aut onomous Taxi • Single agent vs. multiagent Perf M: Saf et y, dest inat ion, legalit y… Envir: St reet s, t raf f ic, pedest rians, weat her… Hardest case: Part ially observable, st ochast ic, Act u: St eering, brakes, accelarat or, horn… sequent ial, dynamic, cont inuous and mult iagent . Sens: GP S, engine sensors, video… (Real world) 21 22 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart Examples Many Applications Solitaire Backgammon Internet Taxi •credit card fraud detection Shopping •printer diagnostics, help in Windows, spam filters Fully Fully Partially Partially •medical diagnosis, teleoperated/micro surgery Observable Observable Observable Observable •information retrieval, Google Deterministic Stochastic Stochastic Stochastic •TAC (Trading Agent Competition) Sequential Sequential Sequential Episodic •scheduling, logistics, etc. Static Static Dynamic Dynamic •aircraft, pipeline inspection •speech understanding, generation, translation Discrete Discrete Discrete Continuous •Mars rovers Single agent Multiagent Multiagent Multiagent •and, of course, cool robots 23 24 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart 4

Mobile Robotics Next Class • Uninformed search • Chapter 3 (Russell & Norvig) 25 26 cs486/686 Lecture Slides (c) K. Larson and P. Poupart cs486/686 Lecture Slides (c) K. Larson and P. Poupart 5