INTELLIGENT AGENTS

What is an intelligent agent?

	Analogue to human agents -- travel agent, real-estate 
	agent, negotiation agent

	View: being busy people, we want computers to do more 
	for us.

	Problem: we want our agents to use common sense and to 
	make good decisions without our supervision... we 
	want intelligent agents.

How difficult is this?

	Consider difficulties for human agents:
	examples from McGuckin's hardware

	Who own's the problem?
	real-world problem/solution co-evolution

	How indirect is the decision?
	sales representatives can inform solution
	intermediary may not be able to abstract problem

	How much knowledge about goals/world is needed for 
	success?



			MECHANISTIC VIEW OF AGENTS

Agents use
	SENSORS (e.g. cameras, microphones, etc.)
	to attain
	PERCEPTS (information about environment)
	and a
	REASONING ALGORITHM (e.g. lookup, planning)
	to determine a course of 
	ACTION (to change the world)
	which is implemented through
	EFFECTORS (things that implement change.)

	At some level an agent can be seen as a mapping 
	from perceived states to actions.

	Percepts, Actions, Goals, Environment
	(examples...)

Autonomy:

	book's definition -- ability to learn from experience (do 
	more than was originally entered by designer)

	another view -- ability to take action without consulting 
	person being represented



			TYPES OF AGENTS

Simple reflex agents
	percepts used to select action

Agents that represent world
	percepts and representation of environment used to 
	select action

Goal-based agents
	percepts, representation of environment, and 
	representation of goals used to select action

Utility-based agents
	percepts, representation of environment, and utility 
	function used to select action

A question of what is represented and how it is used...

How many representations for a chess game can we come 
up with?



			PROPERTIES OF ENVIRONMENTS

accessible vs. inaccessible
	percepts include all relevant information
	chess vs. poker

deterministic vs. nondeterministic
	environment changes determined by prior state and agent action
	chess vs. backgammon

episodic vs. nonepisodic
	effects of actions independent within episodes
	bowling vs. chess

static vs. dynamic
	environment changes during agent deliberation
	chess vs. multi-player Doom

discrete vs. continuous
	limited number of distinct, clearly defined percepts and actions
	chess vs. billiards



			WHAT IS AN AGENT?

Which of the following should be considered agents?

	file cabinet

	coffee maker

	biff (electronic mail notification)

	tax preparation software with advice

	spelling checker that points out unknown words

	spelling checker that suggests replacements for 
	unknown words

	spelling checker that automatically replaces unknown 
	word

	"do what I mean" (DWIM) operating system shell

	Microsoft Office helpers (animated paper clip and 
	friends)

	E-Cash Haggler

	Mar's Rover



How do you make the decision?

	What about Anthropomorphism & Personification?

	Are delegation, responsibility, or autonomy a key?