Intelligent Agents

True/False: Agent's choice of action can depend on entire percept sequence

True

Agent ___________ and _____ in an environment

perceives and acts

agent

Agent is anything that perceives its environment through sensors and acts upon that environment through effectors.

Deterministic vs. Stochastic

Deterministic-processing times of the jobs are known Stochastic- processing times of the jobs can vary/are unknown (1/6 properties of environments)

Discrete vs. Continuous

Discrete: a finite number of percepts and actions that can be performed within it else it is called continuous environment. (1/6 properties of environments)

Safety is an example of which PEAS descriptor for a taxi?

*PERFORMANCE MEASURE (ie safe)* Environment (ie road) Actuators (ie steering) Sensors (ie camera) Other Perf. Measures include: fast, obey laws, reach destination, comfortable trip, maximize profits

Properties of Environments...(6)

1. Fully observable vs. partially observable - If sensors give access to complete state of environment 2. Deterministic vs. stochastic - If next state of environment is completely determined by current state and the action executed by the agent (Can't predict environment in stochastic) 3. Episodic vs. sequential - Experience divided into atomic episodes (perceiving and acting) - Next episode does not depend on previous episodes 4. Static vs. dynamic - Environment not change while agent is"thinking" 5. Discrete vs. continuous: Distinct, clearly defined percepts and actions (chess) 6. Single Agent vs. multi-agent - Solving a puzzle is single agent - Chess is competitive multi-agent environment

Basic Types of Agent Programs (5)

1. Simple reflex agents: Condition-action rules on current percept; Environment must be fully observable 2. Model-based reflex agents: Maintain internal state about how world evolves and how actions effect world 3. Goal-based agents: Use goals and planning to help make decision 4. Utility-based agents: What makes the agent "happiest" 5. Learning agents: Makes improvements

Reflex, model-based, goal-based, utility-based, learning

Agent types

Percept

Agent's perceptual inputs at any given instant

Autonomy

Autonomous behavior - Behavior is determined by its own experience Non-autonomous behavior - If no use of percepts (use only built-in knowledge), then system has no autonomy • A clock • But consider a clock that detects and sets to atomic clock, or adjusts to different time zones - All of its assumptions must hold - Certain animal behaviors A rational agent should be autonomous

Percept sequence

Complete history of everything agent has perceived

True/False: Specifying which action to take in response to any given percept sequence is a percept

False: Agent Function = Specifying which action to take in response to any given percept sequence Percept = Agent's perceptual inputs at any given instant

True/False: Agent's choice of action can depend on entire action sequence

False: Agent's choice of action can depend on entire *percept* sequence

True/False: Non-autonomous Behavior is determined by its own experience

False: Autonomous behavior - Behavior is determined by its own experience - Rational Agent should be autonomous Non-autonomous behavior - If no use of percepts (use only built-in knowledge), then system has no autonomy

True/False: Rationality = omniscience

False: Omniscient agent *knows actual outcome* of its actions and can *act accordingly* - Impossible in reality (though available in simulation) Rationality is concerned with *expected success* given what has been perceived - Can "explore" to gather more information

True/False: Rationality just depends on what we know about the environment and the actions the agent can perform.

False: While these are 2 things, there are also 2 more... Rationality depends on... - The performance measure that defines degree of success - Percept Sequence (everything seen so far) - What the agent knows about the environment - The actions that the agent can perform

For each possible percept sequence, do whatever action is expected to maximize its performance measure, using evidence provided by the percept sequence and any built-in knowledge

Ideal Rational Agent - Do actions in correct order

Rationality and Performance Measures lead to...

Ideal Rational Agent - Do actions in correct order

Rational VS Omniscient Agent

Omniscient agent *knows actual outcome* of its actions and can *act accordingly* - Impossible in reality (though available in simulation) Rational Agent is concerned with *expected success* given what has been perceived - Can "explore" to gather more information

Omniscient agent

Omniscient agent knows actual outcome of its actions and can act accordingly - Impossible in reality (though available in simulation)

Agent Percepts

Percept: Agent's perceptual inputs at any given instant Percept "sequence": Complete history of everything agent has perceived Agent's choice of action can depend on entire percept sequence

A way to evaluate the agent's success • Embodies the criterion for success of an agent's behavior

Performance Measure - Specifies numerical value for any environment history toward the goals - When to evaluate is also important (Timespan)

Consider an "automated taxi driver" (Total Recall) PEAS Description?

Performance Measure? - Safe, fast, obey laws, reach destination, comfortable trip, maximize profits Environment? - Roads, other traffic, pedestrians, weather, customers Actuators? - Steering, accelerator, brake, signal, horn, speak, display Sensors? - Cameras, microphone, sonar, speedometer, GPS, odometer, accelerometer, engine sensors, keyboard

PEAS Example Agent type: Satellite image analysis system

Performance Measures: Correct image classification Environment:Downlink from orbiting satellite Actuators: Display classification of scene Sensors: Color pixel arrays (cameras)

PEAS Example Agent type: Medical diagnosis system

Performance Measures: healthy patient, minimize lawsuits/costs: Environment: patient, staff, hospital Actuators: Display questions, tests, diagnoses, treatments, referrals Sensors: Keyboard entry of symptoms, findings, patient's answers

Specifies numerical value for any environment history toward the goals

Performance measure - A way to evaluate the agent's success • Embodies the criterion for success of an agent's behavior • When to evaluate is also important (Timespan)

The road is an example of which PEAS descriptor for a taxi?

Performance measure (ie safe) *ENVIRONMENT (ie road)* Actuators (ie steering) Sensors (ie camera) Other Environment factors include: other traffic, pedestrians, weather, customers

Steering is an example of which PEAS descriptor for a taxi?

Performance measure (ie safe) Environment (ie road) *ACTUATORS (ie steering)* Sensors (ie camera) Other Actuators include: accelerator, brake, signal, horn, speak, display

A camera is an example of which PEAS descriptor for a taxi?

Performance measure (ie safe) Environment (ie road) Actuators (ie steering) *SENSORS (ie camera)* Other sensors include: microphone, sonar, speedometer, GPS, odometer, accelerometer, engine sensors, keyboard

Structure of Intelligent Agents (2 parts)

Structure of Intelligent Agent = Architecture + Program 1. Architecture is the computing device • Makes sensor percepts available to the program • Runs the program • Feeds action choices to effectors 2. Program • Implements agent function mapping of percepts to actions The job of AI is to design Agent Programs - Though much current emphasis on embodiment

Program

Structure of Intelligent Agent = Architecture + Program Program • Implements agent function mapping of percepts to actions

True/False: The most complex environment is one that is inaccessible, non-deterministic, non-episodic, dynamic and continuous.

TRUE : google slides : Environments that are partially observable, stochastic, sequential, dynamic, continuous, and multi-agent are hardest (real world)

Ideal Mapping of Percepts to Actions

Table of actions in response to each possible percept sequence BUT - Simple table representation can be huge • For chess, the table would have 35^100 entries! - Takes too long to build the table => define mapping

The "problems" to which rational agents are the"solutions"

Task environment

Generic Agent Diagram

The agent uses its sensors (eye) to perceive its environment. It uses its effectors (X) to act upon the the environment. eye = SENSOR _____________________ [ | environment | ---> percepts --->[ eye _____________________ <--- actions <---X[ AGENT [ X X X=effectors

Partially Observable Environment

The agent's sensors do not have complete access to the state of the task environment

True/False: Rationality ≠ omniscience

True! Omniscient agent *knows actual outcome* of its actions and can *act accordingly* - Impossible in reality (though available in simulation) Rationality is concerned with *expected success* given what has been perceived - Can "explore" to gather more information

True/False: A rational action is measurable

True: Need a way to measure success => performance measure embodies the criterion for success of an agent's behavior

True/False: Must specify the setting for intelligent agent design

True: Task environments: The "problems" to which rational agents are the"solutions" • Multiple flavors of task environments -> Directly affects the design of the agent • PEAS description (P)erformance Measure (E)nvironment (A)cutators (S)ensors

True/False: A rational agent is one that does the right thing (to be most successful)

True: ie) every entry in the function table is filled out correctly

effectors

anything an agent uses to act upon its environment

sensors

anything that agent uses to perceives its environment through

Deterministic Environment

processing times of the jobs are known - opposite stochastic where process times vary or are unknown - property of environment

Agent = __________ + Program

Architecture

What does PEAS description stand for

(P)erformance Measure (E)nvironment (A)cutators (S)ensors

Agent design: PEAS

(P)erformance measure (E)nvironment (A)ctuators (S)ensors

Sensors and Effectors for Robot Agents

- Sensors: cameras, infrared range finders - Effectors: various motors

Sensors and Effectors for humans

- Sensors: eyes, ears, etc. - Effectors: hands, legs, mouth, etc.

episodic vs sequential

- Sequential environments require memory of past actions to determine the next best action. - Episodic environments are a series of one-shot actions, and only the current (or recent) percept is relevant. (1/6 properties of environments)

Rationality Depends on...(4)

1. The performance measure that defines degree of success 2. Percept Sequence (everything seen so far) 3. What the agent knows about the environment 4. The actions that the agent can perform

rational agent

A rational agent is one that does the right thing (to be most successful) ie) every entry in the function table is filled out correctly

Performance measure

A way to evaluate the agent's success • Embodies the criterion for success of an agent's behavior Specifies numerical value for any environment history toward the goals

What type of Agent can "explore" to gather more information? A. Omniscent B. Rational C. Vacuum D. Biological

A. Omniscient agent knows actual outcome of its actions and can act accordingly - Impossible in reality (though available in simulation)

Implements the agent function for an agent

Agent Program

Environment Examples: Taxi Driver 6 Properties of Environment

Automated TaxiDriver 1. Fully observable vs. *partially observable* 2. Deterministic vs. *stochastic* 3. Episodic vs. *sequential* 4. Static vs. *dynamic* 5. Discrete vs. *continuous* 6. Single Agent vs. *multi-agent*

Simple reflex agents

Condition-action rules on current percept; Environment must be fully observable

Environment Examples: CrossWord Puzzle 6 Properties of Environment

Crossword puzzle 1. *Fully observable* vs. partially observable 2. *Deterministic* vs. stochastic 3. Episodic vs. *sequential* 4. *Static* vs. dynamic 5. *Discrete* vs. continuous: Distinct, clearly defined percepts and actions (chess) 6. *Single Agent* vs. multi-agent

Agent environments

Environments that are partially observable, stochastic, sequential, dynamic, continuous, and multi-agent are hardest (real world)

Ideal Rational Agent

For each possible percept sequence, do whatever action is expected to maximize its performance measure, using evidence provided by the percept sequence and any built-in knowledge - Do actions in correct order

Fully observable vs. partially observable

Fully Observable: An agent's sensors give it access to the complete state of the environment at each point in time. Partially Observable: The agent's sensors do not have complete access to the state of the task environment (1/6 properties of environments)

Agent program

Implements the agent function for an agent - Runs on the agent architecture

Model-based reflex agents

Maintain internal state about how world evolves and how actions effect world

Agent's perceptual inputs at any given instant

Percept

Complete history of everything agent has perceived

Percept sequence

Agent = Architecture + _________

Program

The job of AI is to design Intelligent Agent __________

Programs

Agent types

Reflex, model-based, goal-based, utility-based, learning

- Sensors: eyes, ears, etc. - Effectors: hands, legs, mouth, etc.

Sensors and Effectors for Human Agents

- Sensors: cameras, infrared range finders - Effectors: various motors

Sensors and Effectors for Robot Agents

Agent function

Specifying which action to take in response to any given percept sequence - abstract mathematical description - Maps any given percept sequence to an action

Static vs. Dynamic

Static: the actions of an agent modify it. Dynamic: other processes are operating on it; environment changes while 'agent' is thinking (1/6 properties of environments)

Architecture

Structure of Intelligent Agent = Architecture + Program Architecture is the computing device • Makes sensor percepts available to the program • Runs the program • Feeds action choices to effectors

Task environments

The "problems" to which rational agents are the"solutions" • Multiple flavors of task environments -> Directly affects the design of the agent • PEAS description

Goal-based agent

Use goals and planning to help make decision

Utility-based agent

What makes the agent "happiest"

Single vs. Multi-Agent

When there is only one agent in a defined environment, it is named the Single-Agent System (SAS). This agent acts and interacts only with its environment. - Solving a puzzle is single agent If there is more than one agent and they interact with each other and their environment, the system is called the Multi-Agent System - Chess is competitive multi-agent environment (1/6 properties of environments)

What is the rational action for a particular circumstance?

Whichever action that will cause the agent to be most successful - given what has been seen/know Need a way to measure success: performance measure - "Whichever action maximizes the expected value of the performance measure given the percept sequence to date"

anything that perceives its environment through sensors and acts upon that environment through effectors.

agent

Maps any given percept sequence to an action

agent function - abstract mathematical description specifying which action to take in response to any given percept sequence

anything an agent uses to act upon its environment

effectors

Ideal agent takes the action that is...

expected to maximize the performance measure, given its percepts

Learning agents

makes improvements

A _______ _______ is one that does the right thing (to be most successful)

rational agent

anything that agent uses to perceives its environment through

sensors

Ideal Agent

takes the action that is expected to maximize the performance measure, given its percepts