CS 371: Introduction to Artificial Intelligence

Introduction

Welcome!

Todd Neller, Glatfelter 209, tneller@gettysburg.edu, 337-6643, etc. Þ Course Information Page

CS 371 – “Introduction to Artificial Intelligence”

Comparing programming tools to handyman tools

Programmers can “machine” their own tools!

The computer is a power tool for the mind.

Outline

Course information

What is AI?

Agents

Agent environments

Course Information

http://cs.gettysburg.edu/~tneller/cs371/

Course information

Handouts, assignments, solutions

Other useful stuff

Our Text

Stuart Russell and Peter Norvig, “Artificial Intelligence: A Modern Approach”

Unified, agent-based approach

Grading

NO MID-TERM, NO FINAL

Approximately 10 short in-class surprise quizzes

Important to keep up with material on a class-by-class basis

You’re responsible to know the lecture and reading material by the beginning of the following class.

Grading (cont.)

General principles:

Grading affects focus of attention

Credit should generally correspond to effort and time required

Study habits translate to work habits

Analogy to physical trainer

So…

~80% - homework assignments

~10% - in-class quizzes (unannounced)

~10% - class participation

Class Participation

Don’t sweat it: be here + willing to engage = full credit

Some project design decisions will be made collaboratively in class

Culture of cooperation

Honor Code

Do not implicitly or explicitly represent the work of others as your own. “Give credit where credit is due.”

Do your own work.*

Assist each other in understanding of general concepts, but leave the valuable problem solving challenges to the individual.*

* except for collaborative group programming projects where details of collaboration will be given.

Homework Programming

All done in Java unless otherwise specified.

All submitted electronically.

Make sure these paths are in your .cshrc file:

/usr/java1.2/bin

/Apps

Introduction to AI

Many different fields, many different challenging problems…

Knowledge Representation & Reasoning/ Expert System lymph-node pathology diagnosis

Robotics & Machine Vision/Autonomous driving outside Pittsburgh

Speech Recognition/PEGASUS airline reservation system

Game-tree search/Big Blue versus Kasparov

… unified by the desire to construct “intelligent” artifacts

Headline: Squirrel hits Cyclist

Tragic squirrel anecdote

How do we define “intelligence”?

What would it mean for someone to make something that is “intelligent”?

What is AI?

So what is AI? What does it mean for an artifact to be intelligent?

What is AI? (cont.)

Dictionary definitions

Deep Blue article

Knowing math: calculator?, Mathematica?, math student?

Are we just anthropomorphizing?

Four Views of AI

Acting humanly – Turing Test (Alan Turing, 1950)

Thinking humanly – cognitive modeling GPS (Newell & Simon, 1960)

Thinking rationally – laws of thought, logic, Aristotle’s syllogisms

Acting rationally – when there’s no provably correct action but one must act, uncertainty, “acting so as to best achieve one’s goals, given one’s beliefs”

Definition of an Agent

Agent Concepts

“Rational”, “Ideal” agent “does the right thing.”

For each possible percept sequence, do whatever action is expected to maximize the performance measure given the agent’s knowledge of its environment (built in or perceived).

Crossing the street without looking rational?

Performance measure; when to evaluate?

A system is “autonomous” to the extent that it thinks and acts on its own.

P.A.G.E. description

Skeleton Agent

Uses memory to store percept sequence

What’s missing?

Performance measure not part of program

Table-Driven Agent

Losing proposition:

table size large (35^100 for chess alone)

table construction difficult

table is fixed and cannot adapt to new environments

learning table would be difficult!

Vacuum Cleaner World

Actions: Turn left 90º, Turn right 90º, Suck dirt, Turn off

Simplified: Go left, Go right, Suck dirt, Turn off

Possible with table driven agent if simplied?

What if many rooms and not simplified?

Simple Reflex Agent

Simple Reflex Agent (cont.)

Condition-Action Rules (e.g. if the car in front of me is braking, then initiate braking)

Assumption: Correct decision based on current percept

Reflex Agent with State

Reflex Agent with State (cont.)

Keeps description of current world state

Updates description from percepts, actions

Example: Sphex wasp behavior

Goal-based Agent

Utility-based Agent

Properties of Environments

(in)accessible, (non)deterministic, (non)episodic, static/dynamic, discrete/continuous

Summary

Agent – something that perceives and acts in an environment

Ideal – “does the optimal thing given experience”

Autonomous – acts on own experience

Reflex (w/&w/o state) – immediately maps perceptsàactions

Goal / utility-based – acts to achieve goals / maximixe utility

PAGE description

Summary (cont.)

Environment characteristics:

Accessible vs. Inaccesible

Deterministic vs. Nondeterministic

Episodic vs. Nonepisodic

Static vs. Dynamic

Discrete vs. Continuous

As you approach each new problem, think PAGE description and associate it to others.


	Todd Neller, Glatfelter 209, tneller@gettysburg.edu, 337-6643, etc. Þ Course Information Page
	CS 371 – “Introduction to Artificial Intelligence”
		Comparing programming tools to handyman tools
		Programmers can “machine” their own tools!
	The computer is a power tool for the mind.


	http://cs.gettysburg.edu/~tneller/cs371/
		Course information
		Handouts, assignments, solutions
		Other useful stuff


	Stuart Russell and Peter Norvig, “Artificial Intelligence: A Modern Approach”
	Unified, agent-based approach


	NO MID-TERM, NO FINAL
	Approximately 10 short in-class surprise quizzes
	Important to keep up with material on a class-by-class basis
	You’re responsible to know the lecture and reading material by the beginning of the following class.


	General principles:
		Grading affects focus of attention
		Credit should generally correspond to effort and time required
		Study habits translate to work habits
	Analogy to physical trainer
	So…
		~80% - homework assignments
		~10% - in-class quizzes (unannounced)
		~10% - class participation


	Don’t sweat it: be here + willing to engage = full credit
	Some project design decisions will be made collaboratively in class
	Culture of cooperation


	Do not implicitly or explicitly represent the work of others as your own. “Give credit where credit is due.”
	Do your own work.*
	Assist each other in understanding of general concepts, but leave the valuable problem solving challenges to the individual.*
	* except for collaborative group programming projects where details of collaboration will be given.


	All done in Java unless otherwise specified.
	All submitted electronically.
	Make sure these paths are in your .cshrc file:
		/usr/java1.2/bin
		/Apps


	Many different fields, many different challenging problems…
		Knowledge Representation & Reasoning/ Expert System lymph-node pathology diagnosis
		Robotics & Machine Vision/Autonomous driving outside Pittsburgh
		Speech Recognition/PEGASUS airline reservation system
		Game-tree search/Big Blue versus Kasparov
	… unified by the desire to construct “intelligent” artifacts


	Tragic squirrel anecdote
	How do we define “intelligence”?
	What would it mean for someone to make something that is “intelligent”?


	So what is AI? What does it mean for an artifact to be intelligent?


	Dictionary definitions
	Deep Blue article
	Knowing math: calculator?, Mathematica?, math student?
	Are we just anthropomorphizing?


	Acting humanly – Turing Test (Alan Turing, 1950)
	Thinking humanly – cognitive modeling GPS (Newell & Simon, 1960)
	Thinking rationally – laws of thought, logic, Aristotle’s syllogisms
	Acting rationally – when there’s no provably correct action but one must act, uncertainty, “acting so as to best achieve one’s goals, given one’s beliefs”


	“Rational”, “Ideal” agent “does the right thing.”
		For each possible percept sequence, do whatever action is expected to maximize the performance measure given the agent’s knowledge of its environment (built in or perceived).
		Crossing the street without looking rational?
	Performance measure; when to evaluate?
	A system is “autonomous” to the extent that it thinks and acts on its own.


	Uses memory to store percept sequence
	What’s missing?
		Performance measure not part of program


	Losing proposition:
	table size large (35^100 for chess alone)
	table construction difficult
	table is fixed and cannot adapt to new environments
	learning table would be difficult!


	Actions: Turn left 90º, Turn right 90º, Suck dirt, Turn off
	Simplified: Go left, Go right, Suck dirt, Turn off
	Possible with table driven agent if simplied?
	What if many rooms and not simplified?


	Condition-Action Rules (e.g. if the car in front of me is braking, then initiate braking)
	Assumption: Correct decision based on current percept


	Keeps description of current world state
	Updates description from percepts, actions
	Example: Sphex wasp behavior


	(in)accessible, (non)deterministic, (non)episodic, static/dynamic, discrete/continuous


	Agent – something that perceives and acts in an environment
		Ideal – “does the optimal thing given experience”
		Autonomous – acts on own experience
		Reflex (w/&w/o state) – immediately maps perceptsàactions
		Goal / utility-based – acts to achieve goals / maximixe utility
	PAGE description


	Environment characteristics:
		Accessible vs. Inaccesible
		Deterministic vs. Nondeterministic
		Episodic vs. Nonepisodic
		Static vs. Dynamic
		Discrete vs. Continuous
	As you approach each new problem, think PAGE description and associate it to others.