Transcript document

CMPUT 366 Intelligent
Systems:
Introduction to Artificial Intelligence
Instruction Team
Prof: Dekang Lin


Office hours: Tue, Thur: 3:30-4:30, or by appointment
Phone: 492-9920
TAs: Yaling Pei, Mark Schmidt, Gang Wu
E-mail: [email protected]
Home Page: http://www.cs.ualberta.ca/~lindek/366



Announcements
Slides
Assignments
Textbooks
Required

S Russell and P Norvig, Artificial Intelligence: A
Modern Approach, Prentice Hall, 1995.
Recommended


D Poole, A Mackworth and R Goebel,
Computational Intelligence: A Logical Approach ,
Oxford, 1998.
Nilsson, Artificial Intelligence: A New Synthesis,
Morgan Kaufmann, 1998.
Evaluation
4 Assignments


16% each. Solo! (see code of conducts)
Paper/Pencil
 Submit hard copy on due date before class, write ligibly

Implementations (C++/Java)
 Submit using ‘try’. The deadline is 11:59pm on the due
date.
 The implementations must run on the lab machines (in
CSC 219)
Final Exam

36%
Other Issues
Prerequisites


Programming skills (C++, Java)
Elementary probability theory
AI Seminar



http://www.cs.ualberta.ca/~ai/seminars
Friday noons, CSC333
Neat topics, great speakers, FREE PIZZA!
Course Overview
Introduction: intelligent agent
Search and constraint satisfaction
Logical agent and planning
Probabilistic reasoning
Natural language and speech
Perception (if there is time)
What is Artificial Intelligence
(AI)?
Discipline that systematizes and automates
intellectual tasks to create machines that:
Act like humans
Act rationally
Think like humans
Think rationally
Act Like Humans
AI is the art of creating machines that
perform functions that require
intelligence when performed by humans
Methodology: Take an intellectual task
at which people are better and make a
computer do it •Prove a theorem
•Play chess
•Plan a surgical operation
•Diagnose a disease
•Navigate in a building
Turing Test
Alan Turing, a mathematician who not only
cracked the German code making machine,
Enigma during the Second World War, but
invented the concept of computers as we
know them.
Turing asserted that if you can fool a human
into believing that he/she is receiving
answers from another human when in fact it
is a computer, this proves that computers are
doing essentially what human brains do.
“Can machines think” -> “Can machines
behave intelligently?”
Operational test of intelligence: Imitation
Game:
Problem:

Turing Test is not reproducible, constructive, or
amenable to mathematical analysis.
Think Like Humans
How the computer performs functions
does matter
Comparison of the traces of the
reasoning steps
Cognitive science  testable theories of
the workings of the human mind
Examples
Garden-Path Sentence:

The horse raced past the barn fell.
Center-embedding:


The cat that the dog that the mouse that
the elephant admired bit chased died.
The elephant admired the mouse that bit
the dog that chased the cat that died.
But, do we want to duplicate human imperfections?
Think Rationally: Laws of Thought
Normative (or prescriptive) rather than
descriptive
Aristotle: what are correct arguments/thought
processes?

Several Greek schools developed forms of logic:
notation and rules of derivation for thoughts.
Problems:


Not all intelligent behavior is mediated by logical
deliberation
What is the purpose of thinking? What thoughts
should I have?
Act Rationally
Rational behavior: doing the right thing
“The right thing”:



that which is expected to maximize goal
achievement, given the available information
Limited resource, imperfect knowledge
Rationality ≠ Omniscience, Rationality ≠ Clairvoyance,
Rationality ≠ Successes
Doesn't necessarily (but often) involve thinking
Ignores the role of consciousness, emotions, fear
of dying, …
Doesn’t necessarily have anything to do with
how humans solve the same problem.
Example: Semantic
Orientation
In many tasks, it is necessary to
determine the semantic orientation of
words


Mining movie reviews
Routing custermer e-mail
Turney 2002

Determine the semantic orientation of
words using internet search engines.
AI History
Trends Since 90’s
Relying less on logic and more on probability
theory and statistics.
More emphasis on objective performance
evaluation.
Intelligent Agents
Accomplishments in





Game playing: Deep blue, Chinook, …
Space Probe
Biological sequence analysis
OCR
Consumer electronics ……
Notion of an Agent
sensors
?
environment
agent
actuators
laser range
finder
sonars
Source: robotics.stanford.edu/~latombe/cs121/2003/home.htm
touch sensors
Notion of an Agent
sensors
?
environment
agent
actuators
•Locality of sensors/actuators
•Imperfect modeling
•Time/resource constraints
•Sequential interaction
•Multi-agent worlds
Source: robotics.stanford.edu/~latombe/cs121/2003/home.htm
Example: Tracking a Target
• The robot must keep
the target in view
• The target’s trajectory
is not known in advance
• The robot may not know
all the obstacles in
advance
• Fast decision is required
Source: robotics.stanford.edu/~latombe/cs121/2003/home.htm
robot
target
What is Artificial
Intelligence? (revised)
Study of design of rational agents
agent = thing that acts in environment
Rational agent = agent that acts
rationally:


actions are appropriate for goals and
circumstances to changing environments
and goals
learns from experience
Goals of Artificial Intelligence
Scientific goal:

understand principles that make rational
(intelligent) behavior possible, in natural or
artificial systems.
Engineering goal:

specify methods for design of useful,
intelligent artifacts.
Psychological goal:


understanding/modeling people
cognitive science (not this course)
Goals of This Course
Introduce key methods & techniques
from AI





searching,
reasoning and decision making (logical and
probabilistic)
learning (covered in detail in CMPUT466)
language understanding,
...
Understand applicability and limitations
of these methods
Goals of This Course
Our approach:


Characterize Environments
Identify agent that is most effective for
each environment
Study increasingly complicated agent
architectures requiring


increasingly sophisticated representations,
increasingly powerful reasoning strategies
Intelligent Agents
Definition: An Intelligent Agent perceives its
environment via sensors and acts rationally
upon that environment with its acutators.
Hence, an agent gets percepts one at a time,
and maps this percept sequence to actions.
Properties




Autonomous
Interacts with other agents
plus the environment
Adaptive to the environment
Pro-active (goal-directed)
Applications of Agents
Autonomous delivery/cleaning robot

roams around home/office environment, delivering
coffee, parcels,. . . vacuuming, dusting,. . .
Diagnostic assistant helps a human
troubleshoot problems and suggest repairs or
treatments.

E.g., electrical problems, medical diagnosis.
Infobot searches for information on computer
system or network.
Autonomous Space Probes
...
Task Environments: PEAS
Performance Measure

Criterion of success
Environment
Actuators

Mechanisms for the agent to affect the
environment
Sensors

Channels for the agent to perceive the
environment
Example: Taxi Driving
Performance Measure

Safe, fast, legal, comfortable trip, maximize profit
Environment

Roads, other traffic, pedestrians, customers
Actuators

Steering, accelerator, break, signal, horn, …
Sensors

Cameras, sonar, speedometer, GPS, …
Types of Environments
Fully observable (accessible) or not
Deterministic vs. stochastic
Episodic vs. sequential
Static vs. dynamic
Discrete vs. continuous
Single agent vs. multiagent

competitive vs. cooperative
Example: Cleaning Agent
Performance Measure

??
Environment

??
Actuators

??
Sensors

??
SurfBot
Automated web surfing
A SurfBot operates in the environment
of the web.



takes in high-level, perhaps informal,
queries
finds relevant information
presents information in meaningful way
Performance Measure

??
Environment

??
Actuators

??
Sensors

??
Agent Function and Program
Agent specified by agent function


mapping percept sequences to actions
Aim: Concisely implement “rational agent
function”
Agent program



input: a single percept-vector
(keeps/updates internal state)
returns action
Skeleton Agent Program
function SkeletonAgent(percept) returns action
static: memory, [agent's memory of the world]
memory  UpdateMemory(memory,percept)
action  ChooseBestAction(memory)
memory  UpdateMemory(memory, action)
return action
Types of Agents
Simple reflex agents

Actions are determined by sensory input only
Model-based reflex agents

Has internal states
Goal-based agents

Action may be driven by a goal
Utility-based agents

Maximizes a utility function
Simple Reflex Agent
Example
A LEGO MindStormTM program:
if (isDark(leftLightSensor))
turnLeft()
else if (isDark(rightLightSensor))
turnRight()
else goStraight()
What’s the agent function?
Model-Based Agent
Goal-based Agent
Utility-based Agent
Summary
What is AI?
Rationality
A bit of History
Intelligent Agent
PEAS
Types of Agents