Artificial Intelligence: Definition

Transcription

1 Lecture Notes Artificial Intelligence: Definition Dae-Won Kim School of Computer Science & Engineering Chung-Ang University

2 What are AI Systems?

3 Deep Blue defeated the world chess champion Garry Kasparov in 1997

4 During the 1991 Gulf War, US forces deployed an AI logistics planning and scheduling program that involved up to 50,000 vehicles, cargo, and people

5 Proverb solves crossword puzzles better than most humans

6 Sony s AIBO and Honda s ASIMO

7 Web Agents & Search engines: Google, Yahoo

8 Recognition Systems: Speech, Character, Face, Iris, Fingerprint

9 Virtual Reality and Computer Vision

10 Some researchers consider AI as one of the four concepts:

11 1. Systems that think like humans

12 2. Systems that think rationally

13 3. Systems that act like humans

14 4. Systems that act rationally

15 AI: Acting humanly

16 Turing (1950): The Turing Test

17 Can machines think?

18 Can machines behave intelligently?

19

20 Turing test is The Imitation Game

21 Predicted that by 2000, a machine might have 30% chance of fooling a lay person for 5 min. In 2014, something has happened.

22 Problem: Turing test is NOT

23 Turing test is NOT reproducible and amendable to mathematical analysis

24 AI: Thinking humanly

25 It requires scientific theories of internal activities of the brain

26 What level of abstraction? Knowledge or circuits.

27 How to validate? Requires something

28 Requires: Cognitive Science Predicting and testing behavior of human subjects (top-down)

29 Requires: Cognitive Neuroscience Direct identification from neurological data (bottom up)

30 Problem: Thinking humanly is NOT

31 Both are distinct from AI in CS The available theories do not explain anything resembling human-level general intelligence.

32 AI: Thinking rationally

33 Laws of Thought: What are correct arguments/thought processes? by Aristotle

34 Several Greek schools developed various forms of logic:

35 Logic: notation and rules of derivation of thoughts

36 Problem: Thinking rationally is NOT

37 Not all intelligent behavior is mediated by logical deliberation

38 AI: Acting rationally

39 Rational behavior: doing the right thing

40 The right thing: that which is expected to maximize goal achievement, given the available information

41 An agent is an entity that perceives and acts.

42 Agents include humans, robots, thermostats, etc.

43 This course is about designing rational agents.

44 Abstractly, an agent is a function from percept histories to actions f : P A

45 The agent program runs on the physical architecture to produce f

46 For any given class of tasks and environments, we seek the agent with the best performance.

47 Problem: Acting rationally is NOT

48 Computational limitations make perfect rationality unachievable

49 Design best program for given machine resources

50 Which of the following can be done at present? Play a decent game of table tennis Drive safely along a curving mountain road Drive safely along Telegraph Avenue Buy a week s worth of groceries on the web Discover and prove a new mathematical theorem Design and execute a research program in biology Write an intentionally funny story Give legal advice in a specialized area of law Translate spoken English into Swedish in real time Perform a complex surgical operation Converse successfully with another person for an hour

51 Artificial Intelligence Intelligent Agents Dae-Won Kim School of Computer Science & Engineering Chung-Ang University

52 The agent function maps from percept histories to actions: f : P A

53 A Vacuum-cleaner Agent

54 Perception:? Actions:?

55 Perception: location and contents [A, Dirty]. Actions: Left, Right, Suck, NoOp

56

57 Problem: A Vacuum-cleaner Agent

58 What is the right function?

59 Let s talk about Rationality

60 A rational agent chooses whichever action maximizes the expected value of the performance measure given the percept sequence to date

61 What is performance measure?

62 1 point per square cleaned up in time T?

63 Minus 1 point per move?

64 Penalize for > k dirty squares?

65 Therefore, we can say

66 Rational omniscient

67 Perception may not supply all information

68 Rational clairvoyant

69 Action outcomes may not be as expected

70 Hence, rational perfect

71 To design a rational agent, we must specify the task environment (PEAS)

72 Performance measure Environment Actuators Sensors

73 Consider the task of designing the Google driverless car

74 P: safety, comfort, profits, legality E: streets, freeways, traffic, weather A: streering, accelerator, break S: velocity, GPS, engine sensors

75 Consider the task of designing an automated internet shopping agent: e.g., Recommender system

76 P: price, quality, efficiency E: WWW sites, vendors A: display to user, follow URL S: HTML, XML pages

77 Agent Types: four basic types in order of increasing generality

78 Simple reflex agents Reflex agents with state Goal-based agents Utility-based agents

79 Simple Reflex Agents 1. If a student sleeping, then assign a penalty. 2. When applied to Vehicle driving?

80 Reflex Agents with State 1. Check the student s academic history, habits.. 2. Vehicle driving?

81 Goal-based Agents 1. Consider goals: be a good professor in AI class 2. Vehicle driving?

82 Utility-based Agents 1. Utility: performance measure 2. How good grade I will assign / Prof. I could be.

83 Learning Agents All agents will be turning into learning agents.