Alte ternate te De Definiti tions (Ru Human inte telligence - PowerPoint PPT Presentation

CSE E 3401: Intr tro to to Arti tificial Inte telligence 
 What t is Arti tificial Inte telligence? & Log & Logic P ic Prog rogram rammin ing 
 Intr troducti tion Webster says: a. the capacity to acquire and apply knowledge. ● What is AI AI? b.the faculty of thought and reason. … ● What is inte telligence? ● Required Readings: Russell & Norvig Chapters ● What features/abilities do humans (animals? 1 & 2. animate objects?) have that you think are ● Lecture slides adapted from those of Fahiem indicative or characteristic of intelligence? Bacchus. 1 2 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus Alte ternate te De Definiti tions (Ru Human inte telligence (Russell + Norv ssell + Norvig ig) Like humans Not necessarily like humans ● Is imitating humans the goal? ● Pros? Systems that think Systems that think rationally Think like humans Systems that act like Systems that act rationally humans ● Cons? Act 3 4 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus 1

Human inte telligence Human inte telligence ● The Turing Test: ● Turing provided some very persuasive arguments that a system passing the Turing ■ A human interrogator. Communicates with a hidden test is intelligent. subject that is either a computer system or a human. If the human interrogator cannot reliably ● But too much emphasis on deception. decide whether on not the subject is a computer, ● Moreover, the test does not provide much the computer is said to have passed the Turing test. traction on the question of how to actually build an intelligent system. 5 6 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus Human inte telligence Human inte telligence ● In general there are various reasons why trying to ■ But more importantly, we know very little about how the mimic humans might not be the best approach to human brain performs its higher level processes. Hence, this AI. point of view provides very little information from which a scientific understanding of these processes can be built. ■ Computers and Humans have a very different architecture with quite different abilities. ■ However, Neuroscience has been very influential in some ● Numerical computations areas of AI. For example, in robotic sensing, vision ● Visual and sensory processing processing, etc. ● Massive and slow parallel vs. fast serial 7 8 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus 2

Rati tionality ty Rati tionality ty ● Mathematical characterizations of rationality ● The alternative approach relies on the notion of have come from diverse areas like logic (laws rati tionality ty. of thought) and economics (utility theory how ● Typically this is a precise mathematical notion best to act under uncertainty, game theory how of what it means to do the right thing in any self-interested agents interact). particular circumstance. Provides ● There is no universal agreement about which ■ A precise mechanism for analyzing and notion of rationality is best, but since these understanding the properties of this ideal behavior notions are precise we can study them and give we are trying to achieve. exact characterizations of their properties, ■ A precise benchmark against which we can measure good and bad. the behavior the systems we build. ● We ’ ll focus on acting rationally ■ this has implications for thinking/reasoning 9 10 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus Computa tati tional Inte telligence Agency Agency ● It is also useful to think of intelligent systems ● AI tries to understand and model intelligence as being agents ts, either: as a computational process. ■ with their own goals ● Thus we try to construct systems whose ■ or that act on behalf of someone (a “ user ” ) ● An agent is an entity that exists in an computation achieves or approximates the environment and that acts on that environment desired notion of rationality. based on its perceptions of the environment ● Hence AI is part of Computer Science. ● An intelligent agent acts to further its own ■ There are other areas interested in the study of interests (or those of a user). intelligence, e.g., cognitive science which focuses on ● An autonomous agent can make decisions human intelligence. Such areas are very related, but without the user’s intervention, possibly based their central focus tends to be different. on what it has learned 11 12 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus 3

Agent t Schemati tic (I) Types of Agents ts ● Simple reflex agents ts: apply simple condition- Agent action rules to decide next action based on current percepts acts perceives ● Model-based reflex agents: maintain a model of the world, apply rules to decide next action Environment based on current world model ● Goal-based agents ts: decide next action based on current model of world state and current ● This diagram oversimplifies the internal go goal(s) s); may do planning; more flexible! structure of the agent. 13 14 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus Agent t Schemati tic (II) Types of Agents ts user prior knowledge ● Uti tility ty-based agents ts: choose actions to maximize their expected uti tility ty in uncertain Agent Goals Knowledge worlds ● Learning agents: explore space of possible acts actions, evaluate performance, and modify perceives agent to improve Environment ● Supports more flexible interaction with the environment, ability to modify one ’ s goals, knowledge that is applied flexibly to different situations. 15 16 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus 4

De Degrees of Inte telligence AI AI Successes Successes ● In 1997 IBM’s Deep Blue beat chess world ● Building an intelligent system as capable as humans champion remains an elusive goal. ● In 1999, NASA Remote Agent used AI planning ● However, systems have been built which exhibit various specialized degrees of intelligence. to control a spacecraft ● Formalisms and algorithmic ideas have been identified ● In 2005 Stanford team won DARPA Grand as being useful in the construction of these Challenge 132mi race in desert “ intelligent ” systems. ● In 2011, IBM’s Watson beat the top Jeopardy ● Together these formalisms and algorithms form the winners foundation of our attempt to understand intelligence as a computational process. ● In 2016, Google DeepMind AlphaGo beat ● In this course we will study some of these formalisms decade’s top player and see how they can be used to achieve various ● Many successes in speech recognition, machine degrees of intelligence. translation, robotics, scheduling, spam fighting 17 18 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus Reasons for Recent t Progress Sub Subareas s of f AI AI ● Better hardware ● Perception: vision, speech understanding, etc. ● AI techniques are improving, especially: ● Robotics ■ Search methods and heuristics ● Natural language understanding ■ Improved representations ● Machine learning ■ Machine learning, large corpuses ● Reasoning and decision making (our focus) ■ Knowledge representa tati tion ■ Reas Reason onin ing (logical, probabilistic) ■ De Decision making (search, planning, decision theory) 19 20 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus 5

Some Inte teresti ting & En Ente terta taining Prospects ts for AI Vid Video eos ● James May’s Big Idea Man-Machine episode where ● Recent progress has been rapid he meets Honda’s Asimo robot programmed so it ● Concerns about the risks of developing AI can learn to recognize objects http:// ● Are current learning-based AI systems really www.youtube.com/watch?v=QfPkHU_36Cs intelligent? ● Google's self-driving car https:// www.youtube.com/watch?v=TsaES--OTzM ■ Winograd Schema Challenge, e.g. ● Google self-driving care Waymo (recent) https:// The city councilmen refused the demonstrators a permit www.youtube.com/watch?v=uHbMt6WDhQ8 because they fea feared ed violence. Who feared violence? vs ● Google Deep Mind AlphaGo win http:// The city councilmen refused the demonstrators a permit www.theguardian.com/technology/video/2016/ because they advocate ted violence. Who advocated violence? mar/09/alphago-computer-beats-go-champion- video 21 22 CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus CSE 3401 Fall 2017 Yves Lesperance & Fahiem Bacchus 6