AI MIDS THEORY

(a) Suppose a university wants to implement an automated system to assist students in

selecting their courses based on their academic history, preferences, and career goals. The
system needs to make personalized recommendations while also continuously learning and
adapting to changes. Among all five types of agent, what kind of agent this system is?
Select one which is more appropriate according to the given scenario. Briefly explain
and justify your answer.

The most appropriate type of agent for this system is a Learning Agent.

Justification:
The system needs to make personalized course recommendations based on each student's
academic history, preferences, and career goals. It also needs to continuously learn and
adapt to changes as the student’s academic progress and goals evolve over time. A
learning agent is specifically designed to improve its performance by learning from past
experiences and feedback, which makes it the best fit for this scenario.
In short, a Learning Agent is ideal because it can adapt and improve its recommendations
based on the student's changing needs and feedback.

ALL TYPES OF AGENTS:

1. Simple Reflex Agent

 What it does: Follows simple rules to take actions based only on what it currently sees or
perceives.
 Best for: Simple, repetitive tasks that don’t need thinking ahead (e.g., a light turning on
when it gets dark).

2. Model-based Reflex Agent

 What it does: Keeps track of the world’s state and uses that information, not just what it
sees right now, to make decisions.
 Best for: Tasks where the agent needs to remember things (e.g., a robot that navigates a
room and remembers where it’s been).

3. Goal-based Agent

 What it does: Makes decisions to achieve specific goals, planning its actions to reach
them.
 Best for: Tasks that need planning and problem-solving (e.g., finding the fastest route to
a destination).
4. Utility-based Agent

 What it does: Chooses actions that maximize its "happiness" or benefits, balancing
different goals.
 Best for: Situations where you need to make decisions between different goals or choices
(e.g., deciding how to invest money for the best returns).

5. Learning Agent

 What it does: Learns from experience and feedback to improve its actions over time.
 Best for: Tasks that need continuous improvement and adaptation (e.g., recommending
courses to students based on their preferences and past performance).

Quick Summary:

 Simple Reflex: For basic, rule-based actions.

 Model-based Reflex: For tasks that need memory or awareness of past states.
 Goal-based: For tasks that involve planning to achieve a goal.
 Utility-based: For tasks with multiple goals and choices.
 Learning Agent: For tasks that need to adapt and improve over time.

(b) Suppose a restaurant is looking to implement an AI system to take customer orders, provide
menu recommendations based on dietary preferences, and personalize the dining
experience. The system must understand customer inquiries and preferences while
optimizing the restaurant's menu offerings and sales. Which type of AI approach - acting
humanly, acting rationally, thinking humanly, or thinking rationally. Briefly explain
which would be most appropriate for this scenario?

The best AI approach for this scenario is acting rationally.

Why?

The system needs to make smart decisions like recommending the best dishes based on customer
preferences and optimizing sales. It doesn't need to act or think like a human but should focus on
making the best choices to meet the restaurant's goals.

In short, acting rationally is about making the right decisions to achieve the desired outcome.

Here’s a simple and short breakdown of the AI approaches:

1. Acting Humanly

 What it does: AI behaves like a human.

 Best for: Making AI interact like humans, e.g., chatbots or virtual assistants.

2. Thinking Humanly
  What it does: AI thinks like a human, mimicking human thought processes.
 Best for: Understanding or modeling human thinking, e.g., psychological studies or
brain-computer interfaces.

3. Acting Rationally

 What it does: AI makes decisions that achieve its goals effectively.

 Best for: Tasks where the AI needs to make the best decisions, e.g., recommendation
systems (like Netflix or Amazon), self-driving cars.

4. Thinking Rationally

 What it does: AI reasons logically to solve problems.

 Best for: Tasks that need logical problem-solving, e.g., chess engines, planning, or
robotics.

Quick Summary:

 Acting Humanly: AI that acts like humans (e.g., chatbots).

 Thinking Humanly: AI that thinks like humans (e.g., cognitive modeling).
 Acting Rationally: AI that makes the best decisions (e.g., recommendations, self-driving
cars).
 Thinking Rationally: AI that reasons logically (e.g., solving puzzles or playing chess).

(c) Describe PEAS and Properties of Environment for “Medical Diagnostic System”.

PEAS for Medical Diagnostic System:

1. Performance Measure:
o Accuracy of diagnosis, speed of diagnosis, patient satisfaction, and treatment
effectiveness.
2. Environment:
o Patients, medical records, hospitals/clinics, and medical knowledge (diseases,
symptoms, treatments).
3. Actuators:
o Provides diagnosis, recommends treatments, sends alerts, and generates reports.
4. Sensors:
o Collects patient data (symptoms, history), test results, imaging data (X-rays,
MRIs), and inputs from medical devices.

Properties of Environment:
1. Partial Observability: The system may not have all information about the patient's condition.
2. Stochastic: Diagnoses can have uncertainty due to rare diseases or incomplete data.
3. Sequential: Diagnosis and treatment are a series of steps over time.
4. Dynamic: Patient conditions may change, requiring updates to the diagnosis.
5. Discrete: The outcomes (diagnoses) are distinct (e.g., "this disease" or "that disease").
6. Multi-agent: The system interacts with doctors, patients, and other healthcare tools.
In short:
 PEAS defines the goals, tools, and inputs of the system.
  Environment: Includes patients, medical data, and devices.
 Key Properties: The system works with partial info, uncertainty, and changes over time.

JUST READ:
OTHER EXAMPLESOF PEAS :

1. Self-Driving Car:

 Performance Measure: What criteria would you use to evaluate the success of a self-
driving car (e.g., safety, speed, fuel efficiency)?
 Environment: What are the key factors in the environment that affect a self-driving car
(e.g., road conditions, traffic, pedestrians)?
 Actuators: What actions would the self-driving car take (e.g., steering, braking,
accelerating)?
 Sensors: What sensors would the car use to gather information (e.g., cameras, LiDAR,
radar)?

2. Robot Vacuum Cleaner:

 Performance Measure: How would you measure the performance of a robot vacuum
(e.g., cleanliness, battery life, coverage area)?
 Environment: What factors of the environment influence the robot's cleaning process
(e.g., room layout, floor type, obstacles)?
 Actuators: What actions does the robot take (e.g., moving, vacuuming, returning to
charging station)?
 Sensors: What sensors help the robot navigate and clean (e.g., proximity sensors, dirt
sensors, bumpers)?

3. Online Recommendation System (e.g., Netflix):

 Performance Measure: How would you evaluate the success of a recommendation

system (e.g., accuracy of recommendations, user satisfaction)?
 Environment: What is the environment the system interacts with (e.g., user preferences,
movie database, viewing history)?
 Actuators: What actions does the system take (e.g., recommending movies, showing
personalized content)?
 Sensors: What data does the system collect to make recommendations (e.g., user ratings,
viewing history, demographic data)?

4. Smart Home System:

 Performance Measure: What metrics would indicate the success of a smart home
system (e.g., energy savings, convenience, security)?
 Environment: What elements are part of the home environment (e.g., temperature,
lighting, security devices)?
 Actuators: What actions can the system take to manage the home (e.g., adjusting
thermostat, turning on lights, locking doors)?
  Sensors: What sensors does the system use (e.g., motion detectors, temperature sensors,
door/window sensors)?

5. Chatbot for Customer Service:

 Performance Measure: How would you measure the chatbot’s performance (e.g.,
response time, customer satisfaction, issue resolution rate)?
 Environment: What factors affect the chatbot's interactions (e.g., customer queries,
database of FAQs)?
 Actuators: What actions does the chatbot take (e.g., responding to inquiries, offering
solutions, escalating to a human)?
 Sensors: What input data does the chatbot use (e.g., text input from the user, past
conversation history)?

6. Warehouse Management System (Robot):

 Performance Measure: How would you measure the performance of a warehouse robot
(e.g., speed, accuracy of item retrieval, energy consumption)?
 Environment: What factors impact the robot’s work environment (e.g., layout of the
warehouse, inventory items)?
 Actuators: What actions can the robot take (e.g., picking items, navigating the
warehouse, placing items in designated areas)?
 Sensors: What sensors help the robot navigate and identify items (e.g., cameras, barcode
scanners, proximity sensors)?

7. Medical Diagnostic System:

 Performance Measure: What performance metrics would be important for evaluating a

medical diagnostic system (e.g., diagnostic accuracy, treatment effectiveness)?
 Environment: What is the environment for a medical diagnostic system (e.g., patient
data, hospital records, medical devices)?
 Actuators: What actions does the system perform (e.g., generating a diagnosis,
suggesting treatments)?
 Sensors: What inputs does the system use (e.g., patient medical history, test results,
symptoms)?

8. Smart Traffic Light System:

 Performance Measure: How would you evaluate the success of a smart traffic light
system (e.g., traffic flow efficiency, reduced waiting times, safety)?
 Environment: What environmental factors influence the traffic system (e.g., traffic
volume, weather conditions, time of day)?
 Actuators: What actions does the system take (e.g., changing traffic light signals,
controlling pedestrian lights)?
 Sensors: What sensors does the system use to monitor traffic (e.g., cameras, infrared
sensors, pressure sensors)?

9. Autonomous Drone Delivery System:

 Performance Measure: How would you evaluate the performance of a drone delivery
system (e.g., delivery speed, accuracy, reliability)?
  Environment: What are the environmental factors the drone operates in (e.g., weather
conditions, obstacles, delivery locations)?
 Actuators: What actions does the drone perform (e.g., flying, delivering packages,
returning to base)?
 Sensors: What sensors help the drone navigate and avoid obstacles (e.g., GPS, cameras,
ultrasonic sensors)?

10. Voice Assistant (e.g., Amazon Alexa):

 Performance Measure: How would you measure the performance of a voice assistant
(e.g., response accuracy, user satisfaction, task completion rate)?
 Environment: What environmental factors influence the voice assistant (e.g., user
requests, connected devices, internet access)?
 Actuators: What actions does the assistant take (e.g., answering questions, controlling
devices, providing reminders)?
 Sensors: What inputs does the system use (e.g., voice commands, environmental
sounds)?

ALL ALGROITHMS

1. Naive Bayes

 Best for: Text classification (e.g., spam filtering, sentiment analysis), medical diagnosis.
 Why: Assumes features are independent, works well with high-dimensional data.

2. K-Nearest Neighbors (KNN)

 Best for: Classification or regression in small datasets, recommendation systems.

 Why: Non-parametric, simple, and works well when decision boundaries are non-linear.

3. K-Means Clustering

 Best for: Customer segmentation, market segmentation, and image compression.

 Why: Groups data into clusters based on similarity, good for exploratory analysis.

4. Apriori Algorithm

 Best for: Market basket analysis, association rule mining (e.g., finding frequently bought
item sets).
 Why: Identifies relationships and patterns in transactional data.

5. Decision Tree (ID3 Algorithm)

 Best for: Classification tasks (e.g., loan approval, medical diagnosis), simple prediction
rules.
 Why: Provides clear decision-making paths and handles categorical data well.

6. Q-Learning

 Best for: Reinforcement learning scenarios like game-playing bots, robotics, or

pathfinding.
  Why: Learns optimal actions in an environment based on rewards.

7. Linear Regression

 Best for: Predicting continuous variables (e.g., house prices, sales forecasting).
 Why: Establishes a linear relationship between the target and features.

8. Multiple Regression

 Best for: Predicting a dependent variable influenced by multiple factors (e.g., impact of
ads and pricing on sales).
 Why: Models interactions between multiple independent variables.

Each algorithm is optimal when matched to the specific nature of the data and problem at hand.