1 Lyle H. Ungar, University of Pennsylvania What is AI? “Artificial Intelligence is the study of how to make computers do things at which, at the moment, people are better.” E. Rich, Artificial Intelligence “Artificial Intelligence is the part of computer science concerned with designing intelligent computer systems, that is, systems that exhibit characteristics we associate with intelligence in human behavior” A. Barr and E. Feigenbaum, The A.I. Handbook “Artificial Intelligence is that branch of computer science dealing with symbolic, non-algorithmic methods of problem solving” B.G. Buchanan and E. H. Shortliffe, Rule-Based Expert Systems.
2 Lyle H. Ungar, University of Pennsylvania Period Activities: Attempts at Machine Translation Algebraic manipulations Game Playing; Chess, Checkers Pattern Recognition Computational Logic General Problem Solver
3 Lyle H. Ungar, University of Pennsylvania Lessons Learned AI is very hard Much harder than expected! Search is a key tool Limits combinatorial explosion Difficult to handle a broad domain, e.g. common sense
4 Lyle H. Ungar, University of Pennsylvania AI as Search
5 Lyle H. Ungar, University of Pennsylvania AI as Search
6 Lyle H. Ungar, University of Pennsylvania Period Feasible approaches demonstrated Natural Language Processing Computer Vision Expert Systems Speech Understanding New knowledge representation techniques appear Search techniques begin to mature Interaction with other fields Examples: AM (Amateur Mathematician), Eurisko, R1, Hearsay, Shrdlu, Mycin, Prospector. Lessons learned: Knowledge central to intelligence Complex systems are feasible
7 Lyle H. Ungar, University of Pennsylvania Basic Components of Applied A.I. Knowledge representation LogicFrames Semantic NetsScripts ProceduresRules Problem-solving Methodologies Logic-based inference Search Machine learning RegressionNeural Nets Decision trees
8 Lyle H. Ungar, University of Pennsylvania AI is Based on Knowledge representation Representing knowledge correctly is crucial to being able to solve problems Arabic vs. Roman numerals Working from a flow sheet vs. from a list of parts and what each is connected to
9 Lyle H. Ungar, University of Pennsylvania AI is Based on Inference Using knowledge requires inference or search Inference: deducing consequences of facts or hypotheses; inferring what could cause observed facts Problem solving can be viewed as a search through the space of possible solutions Check each possible fault to see if it causes observed symptoms Search needs to be done efficiently because there are typically many possible solutions
10 Lyle H. Ungar, University of Pennsylvania Types of Knowledge ObjectsReactors, people,… RelationshipsInteractions between objects,… ConceptsFundamental principles ConstraintsCost, performance, capacity TaxonomyHierarchical organizations HeuristicsIf reactor level is low then check valve 26 EventsInteresting time points, birthdays, operator interventions,… Plans, tasksDiagnosis as a planned activity, start-ups
11 Lyle H. Ungar, University of Pennsylvania Inference Methods
12 Lyle H. Ungar, University of Pennsylvania Generate and Test Slow, but will work if all potential solutions are generated
13 Lyle H. Ungar, University of Pennsylvania Means-End Analysis DifferenceOperator Delta1Opr-1 Delta2Opr-2 Delta3Opr-3 Delta4Opr-4 Delta5Opr-5 Given a starting situation and a desired situation Compute the difference Look up the difference (delta) in the table Apply the corresponding operator Repeat until final goal is reached
14 Lyle H. Ungar, University of Pennsylvania Heuristic Search
15 Lyle H. Ungar, University of Pennsylvania Heuristic Search
16 Lyle H. Ungar, University of Pennsylvania Current Trends in AI ES tools available in conventional software Enterprise integration Information collect from many sources Intelligent agents Revival of probability and statistics Bayesian belief networks Statistical natural language processing Machine Learning Improved perceptual ability Vision, speech