1. 1 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart CSC384: Intro to Artificial Intelligence  Fahiem Bacchus fbacchus@cs.toronto.edu Office Hours: TBA (watch Web page).  Website www.cs.toronto.edu/~fbacchus/Teaching/csc384.html  Text: Computational Intelligence, A Logical Approach; Poole, Mackworth, Goebel I will place copies on reserve in the library. The book store also has a some copies. Important to keep up to date with the readings.

2. 2 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart CSC384: Intro to Artificial Intelligence  The course (and the text) will use PROLOG. We assume that you have knowledge of the language.

3. 3 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart CSC384: Today’s Announcements  Tutorials will commence next week Jan 16 th. Tutorial room assignments will be announced/posted to the web site.  The web site will be the primary source of more detailed information, announcements, etc. So check the site often (at least twice a week.)  You will be responsible for material covered in class, tutorials, as well as the material covered in the assigned readings.

4. 4 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart CSC384: Today’s Announcements  Outline of today’s lecture. First we will look briefly at the question of what Artificial intelligence studies and what we will be doing in this class. Then we will start in on describing a particular representation and reasoning system.  Readings for today’s lecture Chapter 1 and Sections 2.1-2.4  You may find it useful to do the readings prior to class (for future lectures the readings will be posted prior to the class).

5. 5 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Artificial Intelligence (AI)  What is AI?  What is intelligence?  What features/abilities do humans (animals? animate objects?) have that you think are indicative or characteristic of intelligence?  abstract concepts, mathematics, language, problem solving, memory, logical reasoning, emotions, morality, ability to learn/adapt, etc… Webster says: a. the capacity to acquire and apply knowledge. b.the faculty of thought and reason. …

6. 6 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Some Definitions (Russell + Norvig, 1995) The exciting new effort to make computers that think… machines with minds in the full and literal sense [Haugeland 85] [The automation of] activities that we associate with human thinking, such as decision making, problem solving, learning [Bellman 78] The study of mental faculties through the use of computational models [Charniak & McDermott 85] The study of computations that make it possible to perceive, reason and act [Winston 92] The art of creating machines that perform functions that require intelligence when performed by a human [Kurzweil 90] The study of how to make computers do things at which, at the moment, people are better [Rich&Knight 91] A field of study that seeks to explain and emulate intelligent behavior in terms of computational processes [Schalkoff 90] The branch of computer science that is concerned with the automation of intelligent behavior [Luger&Stubblefield93]

7. 7 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Some Definitions (Russell + Norvig, 1995) Systems that think like humans Systems that think rationally Systems that act like humans Systems that act rationally

8. 8 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Computational Intelligence  Notice the term systems in each box most AI researchers build systems to do these things computational theories or models of intelligence (and systems that implement them) are the goal computational, not “artificial”, intelligence  But it still comes down to intelligence: what is it?  Our defns break down as follows: Two in terms of thinking, two acting Two in terms of what humans do, two rationality  Consider relative advantages of each

9. 9 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Aside on Rationality  What is rationality? A precise mathematical notion of what it means to do the right thing in any particular circumstance (see Ch.10.4 of our text). If you find yourself in a situation where you have several courses of action, choose one that’s best for you: one that has the best chance of achieving your goals (or furthering your interests)… balancing short and long term objectives.

10. 10 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Humans and Rationality  Questions: Any differences between acting like humans and acting rationally? In other words: Do humans act rationally? Short answer: not always  Humans limited by their reasoning capabilities  Human behaviour only partially understood  Humans vs computers Different architecture Different reasoning capabilities  Computers crunch numbers faster than humans  Humans recognize objects in images faster than computers

11. 11 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Mathematically Precise Characterizations of Rationality The advantage of precise mathematical formalisms for capturing rationality is that we can precisely characterize various consequences of “rational behavior”. This provides us with guarantees for the systems we build that realize these formal forms of rationality. Humans vs computers For this reason Artificial Intelligence as studied as a sub-field of Computer Science concentrates on studying mechanisms for achieving formally characterized versions of rationality.

12. 12 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Agency  We’ll focus on acting rationally which has implications for thinking/reasoning  Our aim is to build agents, either: with their own goals or that act on behalf of someone (a “user”)  An agent is an entity that exists in an environment and that acts on that environment based on its perceptions of the environment  An intelligent agent acts to further its own interests (or those of a user)

13. 13 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Agent Schematic (I)  By consider a thermostat or ROSI we see that this diagram is not complete. AgentEnvironment perceives acts

14. 14 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Agent Schematic (II)  Require more flexible interaction with the environment, the ability to modify one’s goals, knowledge that be applied flexibly to different situations. AgentEnvironment perceives acts Knowledge Goals prior knowledge user

15. 15 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Monitoring Intensive-Care Patients The “alarm” network 37 variables, 509 parameters (instead of ~2 52 ) PCWP CO HRBP HREKG HRSAT ERRCAUTER HR HISTORY CATECHOL SAO2 EXPCO2 ARTCO2 VENTALV VENTLUNG VENITUBE DISCONNECT MINVOLSET VENTMACH KINKEDTUBE INTUBATIONPULMEMBOLUS PAPSHUNT ANAPHYLAXIS MINOVL PVSAT FIO2 PRESS INSUFFANESTHTPR LVFAILURE ERRBLOWOUTPUT STROEVOLUMELVEDVOLUME HYPOVOLEMIA CVP BP

16. 16 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Degrees on Intelligence  A rational agent’s degree of intelligence might be characterized by 1. ability to act rationally in a variety of circumstances 2. ability to adopt complex goals and balance their achievement 3. ability to adapt to new circumstances

17. 17 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Agent Properties (Wooldridge, Jennings)  Autonomy: needs no direct intervention to perform duties  Reactivity: perceives its environment and reacts appropriately to it  Proactivity: exhibits goal-directed behavior  Sociability: interacts with other agents

18. 18 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Areas of AI  Perception: vision, speech understanding, etc.  Robotics  Natural language understanding  Reasoning and decision making (our focus) Knowledge representation Reasoning (logical, probabilistic) Decision making (search, planning, decision theory) Machine Learning

19. 19 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Topics We’ll Cover  What we’ll cover in this class logical knowledge representation and reasoning problem solving; graph-based search (AI-style) game tree search planning probabilistic reasoning Bayesian networks utility theory decision making under uncertainty  Lots of other advanced AI courses in other areas

20. 20 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Representation and Reasoning Systems  Perhaps the most important technique developed in AI research is that of a Representation and Reasoning system. A RRS supports the flexible application of knowledge. AgentEnvironment perceives acts Knowledge Goals prior knowledge user

21. 21 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Representation and Reasoning Systems  The idea is that if one uses a specific algorithm operating on specific data structures one is locked into solving a specific problem.  Instead the idea is to represent information or knowledge of the environment declaratively. That is, we represent the information without any commitment to a particular way of using it.  This is the Representation component of the RRS system.

22. 22 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Representation and Reasoning Systems  Then one designs a general purpose way of Reasoning with that information that can be applied to solve many different types of problems.  This is the Reasoning component of the RRS.  A well studied and very powerful way of realizing the two goals of a declarative representation along with a powerful and flexible general purpose reasoner is through the use of logical RR systems.

23. 23 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Logical Representation and Reasoning Systems  In a logical RRS one uses logic as the representation system and a sound (and most likely complete) proof system as the reasoner.  An essential feature logical RRS is that the representation now has a clear formally defined semantics inherited from the logic, and the reasoning system’s effects can be precisely characterized by the concept of Soundness and Completeness with respect to the representation’s semantics.

24. 24 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Logical Representation and Reasoning Systems  As with formally defined notions of rationality this provides us with guarantees on the behavior of the system.

25. 25 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Logical KR and Reasoning  We start with a specific way of doing logical KR and logical reasoning since first order logic (FOL) and definite clauses (Prolog) familiar: we’ll go fast  Representation and Reasoning Systems (RRS) a formalism for representing knowledge about the world in a computer/agent a mechanism for reasoning with that knowledge to draw new conclusions. initially we focus on static environments (no uncertainty)

26. 26 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Three considerations in RRSs  Specification: formal foundations for the knowledge expressed in the system the meaning of this knowledge (what is the information asserting about the environment). how we draw conclusions (derive new facts) from the initial set of facts (e.g., answer questions)  Implementation: how we implement the spec. on a computer (later…)  Representational Methodology: how we use the system to represent specific types of knowledge different RRSs more natural, compact, efficient for different kinds of knowledge.

27. 27 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Logical RRSs  Specification syntax used to express sentences (knowledge) Semantics: method for determining the meaning of sentences Proof Procedures: how to answer questions, derive new facts  The RRS we study: Definite Clause Language (DCL) subset of FOL (no disjunction, negation) forms the basis of logic programming (e.g., Prolog) restrictive, but very powerful

28. 28 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Assumptions  Restrictions we’ll start with (these justify our use of DCL)  IR: Agent’s world usefully described in terms of individuals and relations (properties) of them  Example: Computer travel agent domain Individuals: clients, destinations, hotels, airflight segments, airlines, prices, dates, … Properties: cost (hotel/segment), reliability (airline), rating (hotel), satisfied (client), carrier (segment)… Relations: desirable (dest’n for client on date); available (hotel, date); location (hotel, city), etc.

29. 29 CSC 384 Lecture Slides (c) 2002-2003, C. Boutilier and P. Poupart Assumptions (2) DK: Agent’s knowledge is positive and definite no imprecise or negative knowledge OK: home(pascal, toronto) not OK: home(pascal,toronto) OR home(pascal,york) not OK: NOT home(pascal, vancouver) definiteness not the same as completeness (3) SE: Agent’s environment is static this is only temporary (once we get to problem solving, planning, decision making, we’ll relax this)