UNIT 2

1. How is the term 'State of the Art' defined, and what are some current examples of
AI applications?

State of the Art in AI refers to the most advanced and latest developments in artificial
intelligence at a given time. It includes cutting-edge systems that solve complex problems,
often surpassing human capabilities in specific tasks.

Examples:
1. HITECH (Chess Program) – HITECH was the first AI to defeat a grandmaster, Arnold
Denker, in chess. It analyzed the board and determined the best moves, proving AI
could challenge top human players
2. PEGASUS (Speech Understanding Program) – PEGASUS helped a traveler book a
flight by understanding voice commands. Despite misinterpreting one out of ten
words, it successfully completed the booking and saved $894 over the regular fare
3. MARVEL (Space Monitoring System) – MARVEL, a real-time expert system,
monitored data from the Voyager spacecraft. It detected an anomaly near Neptune
and alerted analysts, preventing a possible issue that might have been overlooked
4. Self-Driving Car System – A robotic system controlled a van for 90 miles without
human input. It used cameras, sonar, and laser sensors to analyze the road and steer
the vehicle safely
5. Medical Expert System – An AI expert system assisted a pathologist in diagnosing a
difficult case. Initially doubting the diagnosis, the expert accepted the AI’s reasoning
after it explained the interactions between symptoms
6. Traffic Monitoring System – A streetlight-mounted camera in Paris monitored traffic
movements. It detected vehicles, reported incidents, and even made emergency calls
when a speeding van collided with a motorcyclist

2. Could you explain the concepts of percept sequence, performance measure, ideal
mapping, and utility function?

1. Percept Sequence – It is the complete history of all inputs (percepts) received by an

AI agent from its environment. The agent makes decisions based on this sequence.
Example: A self-driving car detects traffic lights, pedestrians, and road signs
continuously. The sequence of these inputs helps it decide when to stop or move.

2. Performance Measure – It defines how well an AI agent is achieving its goal. It is a

numerical or qualitative measure that evaluates the agent’s success in performing
tasks.
Example: A chess-playing AI’s performance measure could be its win rate, accuracy of
moves, or time taken to make decisions.
 3. Ideal Mapping – It refers to the perfect relationship between percepts and actions
that an agent should follow to achieve the best possible outcome. It represents the
optimal decision-making process.
Example: In a spam filter, the ideal mapping would perfectly classify emails as spam
or not spam without making mistakes.

4. Utility Function – It is a function that assigns a value to each possible state of the
environment, helping the AI agent choose the best action to maximize success. It
ensures better decision-making under uncertainty.
Example: A recommendation system assigns higher utility to movies that match a
user's preferences, ensuring better suggestions based on past ratings.

3. Can you develop an agent program for a simple reflex agent and provide an
explanation?

A Simple Reflex Agent is an AI system that selects actions based only on the current
percept, without considering past experiences or future consequences. It follows
predefined condition-action rules to react instantly to environmental changes.

1. Function Definition – SIMPLE-REFLEX-AGENT(percept) returns action

• The function takes a percept (input from the environment) and returns an
appropriate action based on predefined rules.
2. Static Rules – A Set of Condition-Action Pairs
• The agent has a fixed set of rules that determine what action to take in response to a
given percept.
• Example:
o If the floor is dirty → clean it
o If the floor is clean → move to another location
3. Interpreting the Percept – state ← INTERPRET-INPUT(percept)
• The agent processes the percept to understand the current state of the environment.
• Example:
o The agent receives a "dirty" percept, so it understands that the floor needs
cleaning.
4. Rule Matching – rule ← RULE-MATCH(state, rules)
• The agent checks its predefined rules to find the best action for the given state.
• Example:
o If the state is "dirty", the matching rule will be "clean".
5. Action Execution – action ← RULE-ACTION[rule]
• The agent performs the selected action using its effectors (e.g., robotic arms,
motors).
• Example:
o If the rule says "clean", the vacuum robot starts cleaning the area.
6. Returning the Action – return action
• The agent executes the action and updates the environment.

4. Could you depict the operation of simple reflex agents using a diagram?

A Simple Reflex Agent is an AI system that selects actions based only on the current
percept, without considering past experiences or future consequences. It follows
predefined condition-action rules to react instantly to environmental changes.

Example: A vacuum cleaner detects dirt (percept) and cleans (action) without
remembering past states.

Step-by-Step Explanation

1. Perception through Sensors ("What the world is like now")

o The agent observes the environment using its sensors.
o Example: A vacuum cleaner detects dirt on the floor.
2. Condition-Action Rules ("Condition-action rules")
o The agent processes the percept and applies a predefined rule to decide the
action.
o Example: If the percept indicates "dirty", the rule says "clean".
 3. Decision Making ("What action I should do now")
o Based on the rule, the agent determines the appropriate action.
o Example: The vacuum starts cleaning.
4. Action Execution through Effectors ("Effectors")
o The agent performs the action using its effectors.
o Example: The vacuum cleaner’s motor activates the brush and suction.
5. Environment Interaction ("Environment")
o The action modifies the environment, and the cycle repeats.

5. Could you design an agent program for a goal-based agent and explain its
functionality?

A Goal-Based Agent extends the reflex agent by incorporating goals to determine the best
action. Instead of just reacting to conditions, it considers future consequences and selects
actions that lead toward achieving a desired goal.

Static Components
• State → Describes the current world condition.
• Rules → A set of condition-action pairs that guide the agent’s behavior.

Explanation Based on Figure 2.10

1. Updating State (state ← UPDATE-STATE(state, percept))
o The agent updates its knowledge based on the current percept.
o Example: A robot vacuum detects dirt on the floor and updates its internal
state.
2. Matching Rules (rule ← RULE-MATCH(state, rules))
o The agent selects a rule that applies to the current situation.
o Example: If dirt is present, the rule "clean the dirty area" is selected.
3. Executing Action (action ← RULE-ACTION[rule])
o The agent performs the action associated with the selected rule.
o Example: The vacuum starts cleaning the detected dirt spot.
 4. Updating State Again (state ← UPDATE-STATE(state, action))
o The agent modifies its state based on the performed action.
o Example: After cleaning, it updates the state to "area is now clean."
5. Returning Action (return action)
o The final action is executed based on the defined condition-action rules.

6. Can you illustrate the functioning of a utility-based agent with the help of a
diagram?

A utility-based agent selects actions based on their expected utility (happiness or

effectiveness) rather than just reaching a goal. It evaluates how desirable a particular state
is before making a decision.

Functioning of a Utility-Based Agent (Using Figure 2.12)

1. Perception through Sensors ("What the world is like now")
o The agent observes the environment using its sensors to gather information.
o State: Represents the current world condition based on percepts.
o Example: A self-driving car detects traffic congestion on its current route.
2. Predicting Future States ("What it will be like if I do action A")
o The agent evaluates the possible consequences of different actions.
o Example: The car considers an alternate route and predicts whether it will
reduce travel time.
3. Utility Evaluation ("How happy I will be in such a state")
o The agent uses a utility function to measure the desirability of different
outcomes.
o Example: The car assigns a higher utility score to the route that minimizes
travel time and fuel consumption.
4. Selecting the Best Action ("What action I should do now")
o The agent compares all possible actions and chooses the one with the highest
utility value.
o Example: The car switches to the alternate route since it offers the best travel
efficiency.
 5. Execution through Effectors
o The agent performs the selected action using its effectors.
o Example: The car adjusts its steering and speed to follow the new path.

7. How would you describe the working of goal-based agents, supplemented by a

diagram?

A Goal-Based Agent extends the reflex agent by incorporating goals to determine the best
action. Instead of just reacting to conditions, it considers future consequences and selects
actions that lead toward achieving a desired goal.

Functionality Explanation (Based on Figure 2.10)

1. Perception through Sensors ("What the world is like now")
o The agent receives input from its sensors about the environment.
o Example: A self-driving car detects traffic signals and road conditions.
2. State Maintenance ("State")
o The agent stores and updates information about the environment.
o Example: The car remembers that a red light means stop and green means go.
3. Goal Consideration ("What my actions do")
o The agent evaluates which actions will bring it closer to its goal.
o Example: The car chooses a route to reach its destination fastest.
4. Decision Making Based on Goals ("Condition-Action Rules")
o The agent selects the best action by comparing outcomes with its goal.
o Example: The car slows down when it detects a pedestrian crossing.
5. Action Execution through Effectors ("What action I should do now")
o The agent performs the chosen action in the environment.
o Example: The car accelerates when the road is clear.
 8. In the context of a conveyor belt system with parts, where a part-picking robot acts
as the agent, how would you describe its Performance measure, Environment,
Actuators, and Sensors (PEAS) description?

In a conveyor belt system where a part-picking robot is the agent, we can describe its PEAS
(Performance measure, Environment, Actuators, and Sensors) as follows:

1. Performance Measure (P):

o Accuracy: Picking the correct parts.
o Speed: Picking parts quickly to match the conveyor belt speed.
o Efficiency: Minimizing errors and wasted motion.
o Safety: Avoiding collisions or dropping parts.
2. Environment (E):
o The conveyor belt system (moving parts).
o The workspace where the robot operates.
o Lighting conditions affecting visual sensors.
o Obstacles such as other machines or human workers.
3. Actuators (A):
o Robotic arm (for picking and placing parts).
o Grippers (for holding parts securely).
o Motors (for arm movement and gripping).
4. Sensors (S):
o Cameras/Visual Sensors (to detect and identify parts).
o Proximity Sensors (to measure distance from parts).
o Force/Torque Sensors (to adjust grip strength).
o Infrared Sensors (for detecting obstacles).

9. How would you formulate a well-defined problem for the Vacuum World scenario?

Based on the given image, the Vacuum World scenario can be formulated as a well-defined
problem as follows:
1. Initial State
The vacuum can be in either the left or right room, and each room can either be clean or
dirty. This creates 8 possible states, as shown in the image.

2. Actions (Possible Moves)

• Move Left (if in the right room).
• Move Right (if in the left room).
• Suck Dirt (if the current location has dirt).

3. Transition Model
Each action changes the state:
• If the vacuum sucks dirt, the room becomes clean.
• If it moves left/right, it changes its position.

For example:
• State 1 (Vacuum at Left, Both Dirty)
o Action: Suck Dirt → Moves to State 3 (Left Clean, Right Dirty)
o Action: Move Right → Moves to State 2 (Vacuum at Right, Both Dirty)
• State 3 (Vacuum at Left, Left Clean, Right Dirty)
o Action: Move Right → Moves to State 4 (Vacuum at Right, Left Clean, Right
Dirty)

4. Goal State
• The goal is to reach a state where both rooms are clean, i.e., State 7 or State 8.
5. Performance Measure
• Minimize number of actions taken to clean both rooms.
• Reduce energy consumption by avoiding unnecessary moves.
• Ensure full cleaning by covering all dirty rooms.

10. Select a domain you are familiar with and compose a PEAS description for an agent
operating in that environment. How would you characterize the environment
regarding accessibility, determinism, episodic nature, dynamism, and continuity?
Which agent architecture would be most suitable for this domain?

PEAS Description for a Self-Driving Car

A self-driving car operates in a dynamic environment and must make real-time decisions.
The PEAS (Performance Measure, Environment, Actuators, and Sensors) framework
describes its components as follows:

PEAS Components:
1. Performance Measure: Safe navigation, minimal collisions, efficient routes, and traffic
law compliance.
2. Environment: Roads, vehicles, pedestrians, traffic signals, and weather conditions.
3. Actuators: Steering, acceleration, braking, indicators, and wipers.
4. Sensors: Cameras, LiDAR, RADAR, and GPS for perception and navigation.
Characterization of the Environment:
1. Accessibility: Partially accessible
o The agent can perceive a large portion of its surroundings but may have blind
spots due to occlusions (e.g., objects hidden behind trucks).
2. Determinism: Stochastic (Non-deterministic)
o The environment is unpredictable due to random pedestrian movements,
traffic congestion, and sudden obstacles.
3. Episodic vs. Sequential: Sequential
o Every decision affects future states (e.g., taking a wrong turn increases travel
time).
4. Static vs. Dynamic: Highly dynamic
o The road conditions, other vehicles, and weather constantly change, requiring
real-time decision-making.
5. Discrete vs. Continuous: Continuous
o The agent operates in a continuous space, adjusting speed, direction, and
braking smoothly.

Most Suitable Agent Architecture:

A Utility-Based Agent is ideal as it evaluates multiple factors (e.g., shortest route, safety, fuel
efficiency) and selects the best possible action in real-time.

11. Can you elaborate on the properties of environments in detail and specify which
attributes are most conducive for an agent's performance?
Properties of Environments
1. Accessible vs. Inaccessible
o Accessible: The agent gets complete information about the environment
through its sensors.
▪ Example: Chess, where the full board is always visible.
o Inaccessible: Some aspects of the environment are hidden from the agent.
▪ Example: A self-driving car cannot see around corners or past obstacles.
2. Deterministic vs. Nondeterministic
o Deterministic: The next state is fully predictable based on the current state
and action.
▪ Example: A calculator always gives the same result for the same input.
o Nondeterministic: The next state is affected by random factors or external
conditions.
▪ Example: Weather forecasting depends on unpredictable atmospheric
changes.
3. Episodic vs. Nonepisodic
o Episodic: Each action is independent, and past actions do not influence future
decisions.
▪ Example: Image recognition systems – Each image is analyzed
separately.
o Nonepisodic: Actions are linked, meaning past decisions affect future
outcomes.
▪ Example: Chess, where earlier moves impact later game situations.
 4. Static vs. Dynamic
o Static: The environment remains the same while the agent makes decisions.
▪ Example: A crossword puzzle does not change while being solved.
o Dynamic: The environment keeps changing over time, even if the agent does
nothing.
▪ Example: A self-driving car must adjust to changing traffic conditions.
5. Discrete vs. Continuous
o Discrete: The environment has a finite number of possible actions and states.
▪ Example: Chess, where each piece has a limited number of moves.
o Continuous: The environment has an infinite range of possibilities.
▪ Example: A robotic arm moving smoothly in any direction.

12. Could you discuss the characteristics of environments in depth and provide
examples of real-world environments along with their properties?
SAME ANS. OF Q11 WRITE HERE

13. How would you construct a basic environment simulator program and explain its
components?

Basic Environment Simulator Program

An environment simulator models how an agent interacts with its surroundings.

Components of the Environment Simulator (from Figure 2.14)

1. State: Represents the current condition of the environment.
2. UPDATE-FN: A function that modifies the environment based on agent actions.
3. Agents: A set of intelligent entities that perceive and act.
4. Termination: A condition that checks when the simulation should stop.
Working of the Simulator
1. Perception Phase:
o Each agent receives a percept based on the current state using GET-
PERCEPT(agent, state).
2. Decision Phase:
o Each agent decides an action based on its percept using
PROGRAM[agent](PERCEPT[agent]).
3. Action Execution:
o Actions modify the state through UPDATE-FN(actions, agents, state).
4. Repeat Until Termination:
o The process continues until a stopping condition is met.

Example: Vacuum Cleaner Environment

• State: Dirty or Clean rooms.
• Percept: The agent senses if the room is dirty or clean.
• Action: If dirty → clean; else move to another room.
• Update: Room status changes after cleaning.
• Termination: All rooms are clean.

14. Can you develop an environment simulator program that monitors the
performance measure for each agent and elucidate its functionality?

Environment Simulator with Performance Monitoring

This simulator (Figure 2.15) not only simulates the environment but also tracks the
performance of each agent.
Components of the Simulator
1. State: Represents the current status of the environment.
2. UPDATE-FN: Updates the environment based on agents' actions.
3. Agents: A set of entities interacting with the environment.
4. Termination Condition: Defines when the simulation ends.
5. PERFORMANCE-FN: Evaluates and updates agents' scores based on actions.

Working of the Simulator

1. Initialize Scores:
o A score vector (size = number of agents) is set to zero.
2. Perception Phase:
o Each agent perceives the environment state.
3. Decision Phase:
o Agents choose actions based on perception.
4. Execution & Update:
o The environment updates based on actions.
5. Performance Evaluation:
o PERFORMANCE-FN(scores, agents, state) calculates and updates each agent’s
performance score.
6. Repeat Until Termination:
o The process continues until the stopping condition is met.
7. Return Scores:
o The function returns agents' final performance scores.

Example: Vacuum Cleaner Scenario

• State: Rooms are dirty or clean.
• Percept: The agent detects if a room is dirty.
• Action: If dirty → cleans, else moves.
• Performance: Score increases for cleaning but decreases for unnecessary
movements.
 15. What defines problem-solving agents, and what steps do they undertake to address
AI challenges?

Problem-Solving Agents in AI
A problem-solving agent is an AI system designed to find solutions for given problems using
search and planning techniques. These agents follow a systematic process to make
decisions and achieve goals.

Steps of a Problem-Solving Agent

1. Problem Formulation:
o The agent defines the goal and identifies the initial state, possible actions,
and goal state.
o Example: A robot vacuum sets its goal to clean all dirty rooms.
2. Search for Solutions:
o The agent explores different possible sequences of actions.
o It may use uninformed search (e.g., BFS, DFS) or informed search (e.g., A*).
3. Plan Execution:
o Once a solution is found, the agent follows the sequence of actions.
4. Action Execution & Monitoring:
o The agent performs actions and monitors changes in the environment.
5. Goal Achievement & Optimization:
o The agent checks if the goal is reached and may optimize for better efficiency.

Example: Self-Driving Car

• Problem Formulation: Reaching a destination safely.
• Search for Solutions: Finding the best route using maps.
• Plan Execution: Following the planned path.
• Monitoring: Adjusting based on traffic signals.
• Goal Achievement: Successfully reaching the target.
 16. The sequence "Formulate, Search, and Execute" pertains to which steps, and who is
responsible for performing them? Could you provide an explanation?

This sequence represents the three key steps performed by a problem-solving agent to
achieve its goal.

Steps and Explanation

1. Formulate (Problem Formulation)
o The agent defines the problem, including:
▪ Initial state (starting condition)
▪ Actions (possible moves)
▪ Goal state (desired outcome)
o Example: A delivery robot formulates a problem to reach a target location.
Search (Finding a Solution)
o The agent searches for the best sequence of actions to reach the goal.
o It may use search algorithms like BFS, DFS, A*.
o Example: The delivery robot finds the shortest route to its destination.
2. Execute (Plan Execution)
o The agent executes the selected plan step by step.
o It continuously monitors the environment and adapts if needed.
o Example: The robot follows the chosen path, avoiding obstacles if necessary.

Who Performs These Steps?

The problem-solving agent performs these steps autonomously, using AI techniques to
analyze, plan, and act efficiently.

This process is used in robotics, navigation, and automated decision-making!