Module 1

What Is AI?

• Arti cial Intelligence (AI) enables machines to replicate human intelligence and cognitive
capabilities.
• It is a broad eld of computer science focused on creating smart machines.
• AI systems are designed to perform tasks that usually require human intelligence, such as
reasoning, learning, and problem-solving.
• AI is increasingly integrated into everyday life, with examples including:
• Self-driving cars that navigate and make decisions without human input.
• Smart assistants like Siri and Alexa that understand voice commands and provide helpful
responses.
• The eld continues to grow rapidly, impacting industries like healthcare, nance, education, and
transportation.
fi
fi
fi
fi
Approaches of AI
1. Acting humanly: The Turing Test approach
• Proposed by Alan Turing in 1950.
• The goal is to make a machine act like a human.
• A machine passes the test if a human can’t tell it’s a machine.
• It involves natural language processing to communicate.
• Needs knowledge, reasoning, and learning.
• Also requires perception (like seeing or hearing) and movement.
• Focus is on human-like behavior, not how the machine works inside.
2. Thinking humanly: The cognitive modeling approach
• Focuses on making machines think like humans.
• Tries to model the human mind and mental processes.
• Based on how humans solve problems and learn.
• Uses findings from psychology and cognitive science.
• Often tested by comparing a machine’s thinking to human behavior.
• Involves building programs that simulate human thought steps.
3. Thinking rationally: The “laws of thought” approach
•Focuses on making machines think logically and correctly.
•Based on rules of reasoning (like in mathematics and logic).
•Aims to follow clear, rational steps to solve problems.
•Uses formal logic to guide thinking.
•Assumes the machine should always make the right decision.
•Sometimes hard to apply when real-world problems are unclear or complex.
4. Acting rationally: The rational agent approach
•Focuses on making machines act to achieve the best outcome.
•A rational agent chooses actions that give the most success.
•It doesn't have to think like humans, just act logically and effectively.
•Uses perception and reasoning to decide what to do.
•Works even when the environment is uncertain.
•Goal is to take the best possible action in any situation.
The State of Art or Applications of AI
• Robotic Vehicles: Self-driving cars like STANLEY and BOSS can navigate rough terrain and city traffic safely.
• Speech Recognition: AI systems handle full voice conversations, such as booking flights.
• Autonomous Planning: Space missions use AI to plan and control operations without human input.
• Game Playing: AI has beaten world champions in chess and other games.
• Spam Filtering: AI filters billions of emails daily to block spam.
• Logistics Planning: AI plans large-scale transport and operations efficiently.
• Robotics: Robots like Roomba and PackBot perform tasks at home and in hazardous zones.
• Machine Translation: AI translates languages automatically using data and learning models.
Agents and Environments
• An agent perceives its environment through sensors and acts using actuators.

• An agent's behavior is de ned by the agent function, which maps percept sequences to actions.

• The agent function is an abstract concept; the agent program is its concrete implementation.

• A percept is an agent’s input at a moment; a percept sequence is the complete history of inputs.

• Intelligent agents choose actions based on their percept sequence, not unseen data.

• In simple environments (e.g., vacuum-cleaner world), agent functions can be explicitly tabulated.

• The quality of an agent depends on how well its actions align with its percept history to achieve
goals.
fi
Concept of Rationality
• A rational agent chooses actions that are expected to achieve the best outcome based on what it knows.
• Rationality depends on:
• The performance measure (what defines success),
• The agent’s percepts (what it has observed),
• The agent’s knowledge,
• The actions available, and
• The built-in capabilities of the agent.
• Rationality is about doing the best possible action at the moment, not being perfect or always right.
• An agent is not expected to be omniscient (all-knowing) — it doesn’t know the future or all outcomes.
• Even if the result is bad, the agent can still be rational if the decision was reasonable at the time.
• Exploration is necessary: agents must sometimes try new actions to learn useful things.
• A rational agent should be able to learn from past experiences to improve its performance.
• Learning helps the agent adapt to unknown or changing environments.
• An agent with high autonomy relies mostly on its own learning, not just on hardcoded rules.
• More autonomous agents are generally more flexible and intelligent in real-world situations.
The nature of Environments
1. An agent operates in an environment, and its behavior depends on how well it handles the task environment.

2. A task environment includes everything the agent interacts with to complete its task.

3. To design a rational agent, we need to specify the task environment clearly using the PEAS framework:

• P: Performance measure – what de nes success

• E: Environment – what the agent perceives and acts upon
• A: Actuators – what tools/actions the agent uses
• S: Sensors – how the agent gathers information
4. Task environments differ greatly depending on the application (e.g., self-driving car vs. chess program).
fi
Key Properties of Task Environments/ Types of Environments

• Fully Observable vs. Partially Observable:

◦ Fully observable: the agent has access to the complete state of the environment at each time step.
◦ Partially observable: the agent has limited or incomplete information (e.g., poker, driving in fog).
• Single Agent vs. Multi-Agent:
◦ Single-agent: only one agent is acting (e.g., crossword solver).
◦ Multi-agent: others are present and can cooperate or compete (e.g., chess, autonomous traf c).
• Deterministic vs. Stochastic:
◦ Deterministic: next state is completely determined by current state and action. (e.g., solving maths problem).
◦ Stochastic: includes randomness or uncertainty (e.g., weather forecasting).
• Episodic vs. Sequential:
◦ Episodic: agent’s experience is divided into independent episodes (e.g., image classi cation).
◦ Sequential: current decisions affect future outcomes (e.g., driving, chess).

fi
fi
Static vs. Dynamic:
• Static: environment doesn’t change while the agent is thinking.
• Dynamic: environment changes over time (e.g., real-time systems, robotic navigation).
Discrete vs. Continuous:
• Discrete: limited number of states, percepts, and actions (e.g., chess).
• Continuous: infinitely many possible values (e.g., speed control, robotic arms).
Known vs. Unknown:
• Known: the agent knows the rules and outcomes of its actions.
• Unknown: the agent must learn the rules through interaction or exploration.
The structure of Agents
• An agent consists of an architecture (hardware) and an agent program (software).
• The agent program maps percepts to actions based on what it perceives.
• The structure defines how the agent processes input and chooses actions.
• Good structure helps the agent behave rationally and effectively.
• Different types of agents use different structures depending on task complexity.
Agent Programs
An agent program is the software that implements the agent function.
It runs on the agent's architecture and decides actions from percepts.
The simplest agent program uses a condition-action rule (if percept, then action).
More advanced programs store internal state or goals.
Agent programs vary in complexity depending on the environment.
Types of Agent Programs/ Kinds of Agent Programs
1. Simple reflex agent
• Respond only to the current percept, ignoring the rest of the world.
• Use condition-action rules to choose actions (e.g., "if dirty → suck").
• Work well in fully observable and simple environments.
• Cannot handle partial observability or past knowledge.
• Can be inefficient or blind to important context.
2. Model-based reflex agents
•Maintain some internal state or memory about the world.
•Use a model to keep track of what is unobserved (past events).
•Better than simple re ex agents in partially observable environments.
•Can update beliefs about the world using percepts and a model.
•More exible and effective in dynamic settings.
fl
fl
3. Goal-Based Agents
• Have goals that describe desirable outcomes.
• Choose actions that help them achieve those goals.
• Can plan ahead by considering sequences of actions.
• More flexible than reflex agents as they evaluate outcomes.
• Require more computation and reasoning to make decisions.
4. Utility-based agents
• Use a utility function to measure how good a state is.
• Aim to choose the most desirable action, not just reach any goal.
• Help in situations with multiple conflicting goals.
• Can compare different outcomes to pick the best.
• Are more intelligent in complex, uncertain environments.
5. Learning agents
• Can learn from experience and improve over time.
• Consist of four parts: learning element, performance element, critic, and problem generator.
• The learning element improves decisions using feedback.
• The critic gives feedback on performance.
• Can adapt to new tasks and unknown environments.
How the components of agent programs work
• Agent programs include components like perception, decision-making, learning, and action execution.
• Sensors gather percepts; actuators perform actions.
• Internal data structures store knowledge or history.
• The program uses rules, models, goals, or utilities to decide what to do.
• Learning helps the agent adjust and improve these components over time.
Environment State Representations
• Environmental state representation de nes how the state of the environment is described within an AI system.
• The way a state is represented affects how an agent perceives, reasons, and acts in the environment.
• There are three main types of representations: Atomic, Factored, and Structured.
• Each type offers a different level of detail, expressiveness, and reasoning power.
• Choosing the right representation depends on the complexity of the environment and the task the agent must
perform.

fi
Types of Environment State Representations
1. Atomic Representation
• The environment is viewed as indivisible or a single unit.
• States have no internal structure—just labels (e.g., S1, S2).
• Agents can't access parts of the state; they only deal with the whole.
• Common in search problems and classical planning.
• Example: In a maze, each cell is treated as a distinct, unstructured state.

2. Factored Representation
• States are described by a set of variables or attributes (facts).
• Each state is a combination of variable values (like "door=closed", "light=on").
• Supports logical reasoning, probability, and learning.
• Used in Bayesian networks, Markov models, logical agents.
• Example: A home security system tracking "window=closed", "motion=detected", "alarm=on".
3. Structured Representation
• The environment includes objects with relationships and properties.
• Enables reasoning about individual entities and how they interact.
• More expressive and supports complex inference.
• Used in knowledge representation, ontologies, semantic networks.
• Example: In a medical system, objects like "patient", "disease", and "treatment" are connected via relationships.