What is AI? Big Questions • There are no crisp definitions • Here’s one from John McCarthy, (He coined the phrase AI in Introduction to • Can machines think? 1956) - see http://www.formal.Stanford.EDU/jmc/whatisai/) • And if so, how? Q. What is ? Artificial A. It is the science and engineering of making intelligent • And if not, why not? machines, especially intelligent computer programs. It is related to the similar task of using computers to understand Intelligence • And what does this say about human intelligence, but AI does not have to confine itself to methods that are biologically observable. human beings? Q. Yes, but what is intelligence? Chapter 1 A. Intelligence is the computational part of the ability to achieve • And what does this say about the goals in the world. Varying kinds and degrees of intelligence mind? occur in people, many animals and some machines.

Other possible AI definitions What’s easy and what’s hard? History • AI is a collection of hard problems which can be • It’s been easier to mechanize many of the high level tasks solved by humans and other living things, but for we usually associate with “intelligence” in people which we don’t have good algorithms for solving. – e.g., Symbolic integration, proving theorems, playing chess, medical diagnosis, etc. – E.g., understanding spoken natural language, medical • It’s been very hard to mechanize tasks that lots of animals diagnosis, circuit design, etc. can do • AI Problem + Sound theory = Engineering problem – walking around without running into things • Some problems used to be thought of as AI but are – catching prey and avoiding predators now considered not – interpreting complex sensory information (e.g., visual, aural, …) – modeling the internal states of other animals from their behavior – e.g., compiling Fortran in 1955, symbolic mathematics in – working as a team (e.g. with pack animals) 1965 • Is there a fundamental difference between the two • AI is thus, by nature pre-scientific in Kuhn’s terms categories?

Foundations of AI Why AI? Possible Approaches & • Engineering: To get machines to do a wider variety Like Engineering of useful things humans Well Mathematics Philosophy – e.g., understand spoken natural language, recognize individual people in visual scenes, find the best travel plan Rational for your vacation, etc. GPS Think agents • Cognitive Science: As a way to understand how AI tends to work mostly natural minds and mental phenomena work in this area Psychology – e.g., visual perception, memory, learning, language, etc. Heuristic Linguistics Act Eliza • Philosophy: As a way to explore some basic and systems Cognitive interesting (and important) philosophical questions Science – e.g., the mind body problem, what is consciousness, etc. Like Like Like humans Well humans Well humans Well

Rational GPS Rational Think well Think GPS Rational Act well Think GPS Think like humans Think agents agents agents • For a given set of inputs generate an Heuristic Eliza Heuristic Eliza Heuristic • Cognitive science approach Act Eliza Act systems appropriate output that is not necessarily Act systems systems • Develop formal models of knowledge correct but gets the job done. • Focus not just on behavior and I/O representation, reasoning, learning, •Aheuristic (heuristic rule, heuristic method)isaruleof but also look at reasoning process. memory, problem solving, that can be thumb, strategy, trick, simplification, or any other kind of • Computational model should reflect "how" results were rendered in algorithms. device which drastically limits search for solutions in large obtained. • There is often an emphasis on a systems that are problem spaces. Heuristics do not guarantee optimal • Provide a new language for expressing cognitive theories provably correct, and guarantee finding an optimal solutions; in fact, they do not guarantee any solution at all: and new mechanisms for evaluating them all that can be said for a useful heuristic is that it offers • GPS (General Problem Solver): Goal not just to produce solution. solutions which are good enough most of the time. humanlike behavior (like ELIZA), but to produce a -- Feigenbaum and Feldman, 1963, p. 6 sequence of steps of the reasoning process that was similar to the steps followed by a person in solving the same task.

Like humans Well Eliza Rational Act like humans Think GPS agents • ELIZA: A program that simulated a psychotherapist interacting with a patient and successfully passed the Turing Eliza Heuristic • 3 rooms contain a person, a computer, and an interrogator Act systems Test. The interrogator can communicate with the other 2 by • Coded at MIT during 1964-1966 by Joel Weizenbaum. teleprinter. • Behaviorist approach. • First script was DOCTOR. • The interrogator tries to determine which is the person and – The script was a simple collection of syntactic patterns not unlike • Not interested in how you get results, just the which is the machine. regular expressions similarity to what human results are. • The machine tries to fool the interrogator into believing that – each pattern had an associated reply which might include bits of the • Exemplified by the Turing Test it is the person. input (after simple transformations (my -> your) (, 1950). • If the machine succeeds than we conclude that the machine • Weizenbaum was shocked at reactions: can think. – thought it had potential. – People unequivocally anthropomorphised. – Many thought it solved the NL problem.

E:E: I I am am the the psychotherapist. psychotherapist. Please, Please, describe describe your your problems. problems. Each time you are finished talking, type RET twice. H:H: Good Good afternoon. afternoon. My My name name is is Eliza Each time you are finished talking, type RET twice. Parry P:P: I I don't don't understand understand your your motives. motives. P:AllmenarealikeP:Allmenarealike Colby’s PARRY DoctorDoctor Jones. Jones. E:E: Why Why don't don't you you understand understand my my motives? motives? E:E: Why Why do do you you say say all all men men are are alike? alike? P:P: Pleased Pleased to to meet meet you. you. meets P: They're always bugging us about something or other. P: Shouldn't I be? •Thisisa P: They're always bugging us about something or other. • Kenneth Colby modeled H:H: How How are are you? you? P: Shouldn't I be? transcript of a E:E: Is Is it it because because of of your your plans plans that that you you say say they're they're always always P: All right I guess. Eliza E: Why do you say that just now? bugging us about something or other? a paranoid using the P: All right I guess. E: Why do you say that just now? conversation bugging us about something or other? H: Could you tell me your name? P: My boyfriend made me come here same techniques circa H: Could you tell me your name? P:P: That's That's just just the the way way it it is. is. with an Eliza P: My boyfriend made me come here P: Pat Smith. E:E: Maybe Maybe your your life life have have something something to to do do with with this. this. 1968. P: Pat Smith. E: Suppose that's the way it were not. like program H: I though you might have some E: Suppose that's the way it were not. P:P: He He thinks thinks I I am am depressed depressed • PARRY has basic H: I though you might have some found in Emacs problems. P:P: So? So? E:E: Are Are you you depressed depressed often? often? emotions. If it gets problems. (Gnu 20.2.2) P: You are asking me a lot of questions P: I've been upset lately. E:E: I I am am not not sure sure I I understand understand you you fully. fully. P: You are asking me a lot of questions angry, it’s replies P: I've been upset lately. E:E: What What makes makes you you believe believe I I am am asking asking you you a a lot lot of of H:H: By By what? what? P:P: You You harp harp on on the the same same questions questions too too much. much. • Try typing questions? become more hostile. questions? P:P: People People get get on on my my nerves nerves E: What does that similarity suggest to you? P: Because you seem very nosey • In a Turing test, expert E: What does that similarity suggest to you? $x doctor P: Because you seem very nosey sometimes.sometimes. E:E: Is Is it it because because I I seem seem very very nosey nosey that that you you came came to to me? me? psychiatrists were unable P:P: I I don't don't get get you. you. H:H: Tell Tell me me more. more. in your favorite P:P: I I am am leaving. leaving. to distinguish PARRY’s E: Do you wish that you did get I? P:P: Do Do you you know know anything anything about about E: Do you wish that you did get I? Emacs. E:E: Earlier Earlier you you said said you you felt felt depressed? depressed? ramblings from those of P: Good bye. bookies?bookies? P: Good bye. real paranoids. E:E: My My secretary secretary will will send send you you a a bill. bill...... The Loebner Contest What can AI systems do What can’t AI systems do yet? • A modern version of the turning test with $100,000 cash prize. Here are some example applications • Hugh Loebner was once director of UMBC’s Academic • Computer vision: face recognition from a large set • Robust natural language understanding (e.g. read and Computing Services (nee UCS) • Robotics: autonomous car understand articles in a newspaper) • http://www.loebner.net/Prizef/loebner-prize.html • Natural Language Processing: simple machine translation • Arbitrary visual scene understanding • Restricted topic (removed in 1995) and limited time. • Expert Systems: medical diagnosis in a narrow domain • Learning a natural language • Participants include a set of humans and a set of computers • Spoken language systems: ~1000 word continuous speech • PlayGowell and a set of judges. • Planning and scheduling: scheduling Hubbell experiments •Scoring • Time series predications: – Rank from least human to most human. • Learning: text categorization into ~1000 topics – Highest median rank wins $2000. • User modeling: Baysean reasoning in windows help – If better than a human, win $100,000. • Games: grandmaster level in chess, checkers, etc.