ARTIFICIAL INTELLIGENCE AND MACHINE

LEARNING (IT3201)

(Sem: 6
(Section B))
th

Lecture 24-28
Instructor: Venkatesh Gauri Shankar
 Intelligent Agents in AI

• In the context of AI, intelligent agents are entities that are designed to
perceive their environment, reason about it, and take actions to
achieve specific goals.

• These agents use artificial intelligence techniques to mimic or exhibit

intelligent behavior.
 Agent v/s Software Program
• Agents are autonomous, decision-making entities with perception
capabilities, often using AI, while software programs follow
predetermined instructions for specific tasks.

• All agents are software programs, but not all software programs
are agents.

• An agent is a specialized type of software program that exhibits

autonomous behavior, typically involving perception, reasoning,
and decision-making.
 Agent v/s Software Program
Characteristic Agent Software Program

Typically follows a predetermined sequence of

Autonomy Exhibits autonomy, acting independently instructions

Perception Capable of perceiving its environment May not necessarily have sensing capabilities

Decision-Making Makes decisions based on perception and goals Follows predefined logic or algorithms

Learning May incorporate learning mechanisms May lack adaptive or learning capabilities

Complexity Can be complex, incorporating AI techniques Can vary in complexity, from simple to complex

Purpose Designed to achieve specific goals or tasks Developed for a particular function or task

Adaptability Can adapt to changing conditions May require manual updates for changes

Interaction Can interact with other agents or systems May or may not interact with external entities

Examples Intelligent agents in robotics, virtual assistants Text editors, calculators, basic programs
 Components of Intelligent Agents
• Perception (Sensors): Intelligent agents are equipped with sensors or
mechanisms to perceive information from their environment. These sensors
could include cameras, microphones, temperature sensors, or other input
devices depending on the nature of the agent and its application.

• Knowledge Base: Agents maintain a knowledge base that represents their

understanding of the world. This knowledge can be pre-programmed or
acquired through learning from data and experiences.

• Reasoning: Intelligent agents use reasoning mechanisms to interpret the

information they receive from sensors and their knowledge base. This involves
processing data, making inferences, and deriving conclusions.

• Decision-Making: Based on their reasoning, agents make decisions to achieve

specific goals. Decision-making involves selecting appropriate actions from a
range of possibilities, taking into account the current state of the environment
and the agent's goals.
 Components of Intelligent Agents
• Actuators (Effectors): Intelligent agents have actuators or effectors that
allow them to interact with the environment and execute actions. These
can include motors, speakers, displays, or other output devices.

• Autonomy: Intelligent agents operate autonomously, meaning they can

function without constant human intervention. Autonomy enables them
to adapt to changing conditions and make decisions independently.

• Learning: Many intelligent agents are designed to learn from experience

or data. Machine learning algorithms allow agents to improve their
performance over time by adjusting their behavior based on feedback
and new information.

• Communication: Intelligent agents may communicate with other agents,

humans, or external systems to exchange information, coordinate
activities, or achieve common goals.
 Examples of intelligent agents in AI include:
• Virtual Assistants: Agents like Siri, Google Assistant, or Amazon
Alexa that understand natural language, respond to user queries,
and perform tasks.

• Autonomous Vehicles: Agents that navigate and control vehicles

without human intervention, such as self-driving cars or drones.

• Game-Playing Agents: Agents that play games like chess, Go, or

video games, making decisions to optimize their chances of
winning.

• Robotics: Physical robots that use sensors, actuators, and AI

algorithms to perform tasks in the real world.
 Classification of Agents
• Simple Reflex Agents:
• These agents make decisions based solely on the current percept
(environmental input) without considering the history of past percepts.
They are reactive and lack memory.
• Model-Based Reflex Agents:
• These agents maintain an internal model of the world, allowing them to
consider the history of percepts. They use this model to make more
informed decisions.
• Goal-Based Agents:
• Goal-based agents have specific goals they aim to achieve. They take
actions that move them closer to these goals, considering the current state
and available actions.
• Utility-Based Agents:
• Utility-based agents evaluate different actions based on a utility or
preference function. They choose actions that maximize expected utility,
considering both goals and the desirability of outcomes.
 Classification of Agents
• Learning Agents:
• Learning agents have the ability to improve their performance over time by
learning from experience. They adapt their behavior based on feedback from
the environment.
• Rational Agents:
• Rational agents make decisions that maximize expected utility, given their
current knowledge and beliefs about the world. They are designed to act
logically and effectively.
• Belief-Desire-Intention (BDI) Agents:
• BDI agents model human-like reasoning by maintaining beliefs about the
world, desires (goals) to be achieved, and intentions (plans) to achieve those
goals.
• Perception-Based Agents:
• These agents focus on perception and interpretation of their environment.
They use sensory information to make decisions and take actions.
 Classification of Agents
• Collaborative Agents:
• Collaborative agents work together with other agents or entities to achieve
common goals. This may involve communication, coordination, and
cooperation.
• Reactive vs. Deliberative Agents:
• Reactive agents respond quickly to immediate stimuli without extensive
planning. Deliberative agents, on the other hand, engage in more thoughtful
planning and decision-making.
• Mobile Agents:
• These agents have the ability to move between different computational
environments, executing tasks while in transit.
• Intelligent Software Agents:
• General term encompassing various types of software entities that exhibit
intelligent behavior, including virtual assistants, recommendation systems, and
autonomous agents in robotics.
 Classification of Agents: Case study
• Simple Reflex Agent: Traffic Light Controller

Case Study:
A simple reflex agent is used to control traffic lights at an intersection. The
agent makes decisions based on the current state (percept) of the traffic,
without considering the history.

Scenario:

If the sensors detect heavy traffic on one road, the traffic light for that road
stays green for a longer duration to allow vehicles to clear the congestion.
 Classification of Agents: Case study
• Simple Reflex Agent: Traffic Light Controller

Case Study:
A simple reflex agent is used to control traffic lights at an intersection. The
agent makes decisions based on the current state (percept) of the traffic,
without considering the history.

Scenario:

If the sensors detect heavy traffic on one road, the traffic light for that road
stays green for a longer duration to allow vehicles to clear the congestion.
 Classification of Agents: Case study
• Model-Based Reflex Agent: Chess Playing Agent
▪ Case Study:
▪ A model-based reflex agent plays chess. It maintains an internal model of
the chessboard to make decisions based on the current state and history
of moves.

▪ Scenario:
▪ The agent considers the positions of all pieces on the board and uses a
model to determine the best move based on learned strategies and past
game data.
 Classification of Agents: Case study
• Goal-Based Agent: Personal Assistant App
• Case Study:
• A goal-based agent in the form of a personal assistant app helps users
manage tasks and appointments.

• Scenario:
• The user sets a goal of scheduling a meeting. The agent analyzes the
user's calendar, considers the goal, and suggests available time slots for
the meeting.
 Classification of Agents: Case study
• Utility-Based Agent: Online Retail Recommendation System
• Case Study:
• A utility-based agent is employed in an online retail recommendation
system, such as Amazon's.

• Scenario:
• The agent evaluates the utility or preference of different products for a
user based on their browsing and purchase history, recommending items
with the highest predicted utility.
 Classification of Agents: Case study
• Learning Agent: Email Spam Filter
• Case Study:
• A learning agent is used as an email spam filter.

• Scenario:

• The agent learns from the user's actions (marking emails as spam or not)
and continuously adapts its filtering criteria to improve accuracy over
time.
 Classification of Agents: Case study
• Rational Agent: Stock Trading Algorithm
• Case Study:
• A rational agent operates as a stock trading algorithm.

• Scenario:

• The agent analyzes market data, news, and historical trends to make buy
or sell decisions, aiming to maximize returns while considering risk.
 Classification of Agents: Case study
• Belief-Desire-Intention (BDI) Agent: Social Robot in Elderly Care
• Case Study:
• A BDI agent is implemented in a social robot designed to assist elderly
individuals.

• Scenario:

• The agent maintains beliefs about the environment, desires to provide

companionship and assistance, and forms intentions to perform specific
tasks, such as reminding the user to take medication.
 Classification of Agents: Case study
• Perception-Based Agent: Image Recognition System
• Case Study:
• A perception-based agent is employed in an image recognition system.

• Scenario:

• The agent analyzes images using computer vision techniques, perceiving

and identifying objects or patterns within the images.
 Classification of Agents: Case study
• Collaborative Agent: Swarm Robotics
• Case Study:
• Collaborative agents in the form of swarm robotics work together to
accomplish tasks, such as exploring an unknown environment.

• Scenario:

• Each robot in the swarm communicates with others to share information

about the terrain, optimizing exploration efficiency.
 Classification of Agents: Case study
• Reactive vs. Deliberative Agent: Robotic Vacuum Cleaner
• Case Study:
• A robotic vacuum cleaner can be modeled as a reactive agent.

• Scenario:

• The vacuum cleaner reacts to immediate dirt detection by changing its

cleaning pattern or direction.
 Classification of Agents: Case study
• Mobile Agent: Search and Rescue Drone Fleet
• Case Study:
• Mobile agents in the form of search and rescue drones are deployed to
locate and assist disaster victims.

• Scenario:

• Drones move between different areas affected by a disaster, searching for

survivors and delivering essential supplies.
 Classification of Agents: Case study
• Intelligent Software Agent: Siri (Virtual Assistant)
• Case Study:
• Siri, Apple's virtual assistant, serves as an intelligent software agent.

• Scenario:

• Siri responds to user voice commands, providing information, setting

reminders, and initiating various tasks on the user's device.
 PEAS Description
• Performance Measure defines the goal and the metrics used to evaluate
the success of the agent.
• Environment outlines the context and entities with which the agent
interacts.
• Actuators describe how the agent takes actions to influence the
environment.
• Sensors detail how the agent perceives and gathers information about its
surroundings.
 PEAS Description
1.Performance Measure:
1. Goal: Specifies what the agent is intended to achieve or optimize.
2. Measure: Defines how the success or failure of the agent is quantitatively evaluated. It
could be a numerical value or a set of criteria.
2.Environment:
1. Components: Describes the entities or objects the agent interacts with in its
surroundings.
2. Dynamics: Explains how the environment changes over time and the factors that
influence these changes.
3.Actuators:
1. Actions: Identifies the mechanisms or processes through which the agent affects its
environment.
2. Strategy: Describes the decision-making or control mechanisms the agent employs to
determine its actions.
4.Sensors:
1. Inputs: Specifies the sources of information or data the agent uses to perceive and
understand its environment.
2. Feedback: Describes how the agent receives information about the consequences of its
actions and the changes in the environment.
 PEAS Description: Case Study
• 1. PEAS Description for a Chess-Playing Agent:
• Performance Measure:
• Goal: Win the game.
• Measure: Number of games won, time taken for decision-making, quality of
moves.
• Environment:
• Components: Chessboard, chess pieces, opponent player.
• Dynamics: Turn-based moves, legal and illegal moves, game rules.
• Actuators:
• Actions: Moving chess pieces on the board, capturing opponent pieces.
• Strategy: Decision-making based on the current board state.
• Sensors:
• Inputs: Current state of the chessboard, positions of all pieces.
• Feedback: Opponent moves, overall game state.
 PEAS Description: Case Study
• PEAS Description for a Recommender System:
• Performance Measure:
• Goal: Maximize user satisfaction and engagement.
• Measure: Accuracy of recommendations, user click-through rates, user feedback.
• Environment:
• Components: User profiles, item database (movies, products), historical user
interactions.
• Dynamics: Changing user preferences, new items, user feedback.
• Actuators:
• Actions: Recommending items based on user preferences.
• Strategy: Machine learning algorithms, collaborative filtering.
• Sensors:
• Inputs: User preferences, browsing history, item features.
• Feedback: User interactions, ratings, feedback on recommendations.
 PEAS Description: Case Study
• PEAS Description for an Autonomous Drone:
• Performance Measure:
• Goal: Efficient exploration and data collection in a designated area.
• Measure: Coverage of the area, time taken to complete the mission, quality of
collected data.
• Environment:
• Components: Physical space, obstacles, waypoints, other drones (if collaborative).
• Dynamics: Changing weather conditions, obstacles, real-time sensor data.
• Actuators:
• Actions: Movement in three-dimensional space, capturing images or data.
• Strategy: Path planning, obstacle avoidance, collaborative coordination.
• Sensors:
• Inputs: GPS data, altitude, obstacle detection, camera images.
• Feedback: Environmental changes, proximity to obstacles, status of data
collection.
 Single and Multi-Agent System
• In a single-agent system, there is only one autonomous entity or intelligent
agent that operates in its environment to achieve specific goals.

• This agent is responsible for perceiving the environment, making decisions, and
taking actions without interacting with other agents.

• Single-agent systems are common in many AI applications, such as game-

playing agents, expert systems, and autonomous robots.

• In a multi-agent system, multiple intelligent agents exist and operate within the
same environment.
• These agents may have their own goals, capabilities, and decision-making
processes. The agents can interact with each other, sharing information and
influencing each other's behavior.
• Multi-agent systems are prevalent in applications where collaboration,
coordination, or competition between agents is essential.
 Single and Multi-Agent System
• Example: Chess-Playing Agent
• Agent: The chess-playing program that makes decisions based on the
current state of the chessboard and its goal to win the game.
• Environment: The chessboard, opponent, and game rules.
• Performance Measure: Winning the game or making optimal moves.
• Example: Multi-Agent Robotic Swarm
• Agents: A group of small robotic drones working together to explore an
unknown environment.
• Environment: The physical space with obstacles, potential targets, and
other drones.
• Performance Measure: Efficient exploration, coverage of the entire area.
 Single and Multi-Agent System
• Key Differences:
1.Interaction:
1. Single-Agent: Operates independently without interaction with other agents.
2. Multi-Agent: Involves interactions, collaborations, or competitions between multiple agents.
2.Goals:
1. Single-Agent: Has its own set of goals and objectives.
2. Multi-Agent: Each agent may have individual goals, and there may be collective goals for the
group.
3.Decision-Making:
1. Single-Agent: Makes decisions based on its own perception and goals.
2. Multi-Agent: Decision-making may involve considering the actions and influence of other agents.
4.Complexity:
1. Single-Agent: Generally simpler, as there's only one agent to consider.
2. Multi-Agent: Can be more complex due to interactions and coordination between multiple agents.
5.Applications:
1. Single-Agent: Found in applications where individual entities operate independently.
2. Multi-Agent: Used in scenarios requiring collaboration, distributed problem-solving, or
competition.
Agents in AI