✔ ANI vs AGI vs ASI
✔ ANI vs AGI vs ASI (Comparison with Examples)
Introduction
Artificial Intelligence is classified into three types based on capability:
• ANI – Artificial Narrow Intelligence
• AGI – Artificial General Intelligence
• ASI – Artificial Super Intelligence
1. Artificial Narrow Intelligence (ANI)
• Also called Weak AI
• Designed to perform one specific task
• Cannot think or act beyond its programmed scope
• Most AI today is ANI
Examples:
• Chatbots, voice assistants
• Face recognition
• Recommendation systems
2. Artificial General Intelligence (AGI)
• Also called Strong AI
• Has human-level intelligence
• Can learn, reason, and perform any intellectual task
• Does not exist yet (research stage)
Examples:
• Hypothetical human-like AI system
3. Artificial Super Intelligence (ASI)
• Intelligence beyond human intelligence
• Can outperform humans in all domains
• Self-improving and autonomous
• Exists only as a theoretical/future concept
Examples:
• Super-intelligent AI (science fiction concept)
Comparison Table (Very Important for Exam)
Basis ANI AGI ASI
Artificial Narrow Artificial General Artificial Super
Full form
Intelligence Intelligence Intelligence
Intelligence
Task-specific Human-level Beyond human
level
Learning ability Limited High Extremely high
Decision
Predefined Flexible Self-improving
making
Current status Exists today Not developed Theoretical
Basis ANI AGI ASI
Examples Chatbots, Alexa None Sci-fi AI
Conclusion
• ANI is widely used today
• AGI is the future goal of AI research
• ASI represents a hypothetical super-intelligent future
Short Revision Version (Quick Memory)
• ANI → One task
• AGI → Human-like intelligence
• ASI → Beyond human intelligence
If you want:
• Only 3 / 4 / 7 marks GTU answer
• Diagram-only explanation
• Examples-based short notes
Send next question
✔ Intelligent Agents & Environment
✔ Intelligent Agents & Environment
Intelligent Agent
Definition
An intelligent agent is an entity that perceives its environment through sensors and acts
upon that environment through actuators to achieve a goal.
Main Elements
• Sensors → collect information (camera, microphone, sensors)
• Actuators → perform actions (motors, speakers, display)
• Agent program → decides actions based on perceptions
Examples
• Self-driving car
• Robot vacuum cleaner
• Chatbot
Environment
Definition
The environment is everything outside the agent that it interacts with while performing
tasks.
Examples
• Road and traffic for a self-driving car
• Chessboard for a chess-playing AI
• Users and internet for a chatbot
Agent–Environment Interaction (Cycle)
1. Agent perceives environment
2. Agent decides action
3. Agent acts using actuators
4. Environment changes
5. Cycle repeats
Types of Environments (Key Points)
• Fully / Partially observable
• Deterministic / Stochastic
• Static / Dynamic
• Discrete / Continuous
• Single-agent / Multi-agent
Conclusion
An intelligent agent continuously observes and acts in an environment to achieve goals
efficiently. The nature of the environment strongly affects agent design.
Short Revision Version (Quick Memory)
• Agent = Perceive → Decide → Act
• Sensors = Input | Actuators = Output
• Environment = Agent’s surroundings
If you want:
• Characteristics of intelligent agents
• Rational agent explained
• Only 3 / 4 / 7 marks GTU answer
Send next question
✔ Expert Systems
✔ Expert Systems
Definition of Expert System
An Expert System is an AI-based computer system that imitates the decision-making ability
of a human expert in a specific domain using knowledge and reasoning.
In simple words:
It acts like a human expert to solve problems and give advice.
Main Components of an Expert System
1. Knowledge Base
• Stores facts, rules, and domain knowledge
• Usually represented using IF–THEN rules
Example:
IF fever AND cough → THEN flu
2. Inference Engine
• The brain of the expert system
• Applies rules from the knowledge base
• Draws logical conclusions
3. User Interface
• Interaction between user and system
• Takes user input and shows output
4. Explanation Facility
• Explains how and why a decision was made
• Increases trust and understanding
5. Knowledge Acquisition System
• Collects knowledge from human experts
• Updates and improves the knowledge base
Working of Expert System (Steps)
1. User enters problem
2. Inference engine applies rules
3. System reaches conclusion
4. Output + explanation given to user
Applications of Expert Systems
• Medical field – Disease diagnosis
• Banking – Loan approval
• Industry – Fault detection
• Education – Intelligent tutoring
• Agriculture – Crop disease advice
Advantages
• Fast and consistent decisions
• Available 24×7
• Reduces human error
Limitations
• Expensive to develop
• Limited to specific domain
• Cannot think creatively
Conclusion
Expert systems use knowledge and logical reasoning to provide expert-level solutions,
making them useful in many real-world applications.
Short Revision Version (Quick Memory)
• Expert System = Human expert + AI
• Core parts → Knowledge base + Inference engine
• Used in medical, banking, industry
If you want:
• Only 3 / 4 / 7 marks GTU answer
• Examples like MYCIN, DENDRAL
• Expert system vs Human expert
Send next question
✔ Supervised vs Unsupervised Learning
✔ Supervised Learning vs Unsupervised Learning
Introduction
Machine Learning methods are classified based on whether labeled data is available during
training.
Supervised Learning
Meaning
Supervised Learning trains a model using labeled data (input with correct output).
Key Points
• Input + output (labels) given
• Learns under guidance
• Used for prediction & classification
Examples
• Email spam detection
• House price prediction
Common Algorithms
• Linear Regression
• Logistic Regression
• SVM
• k-NN
Unsupervised Learning
Meaning
Unsupervised Learning trains a model using unlabeled data (only input).
Key Points
• No predefined output
• Finds hidden patterns
• Used for clustering & grouping
Examples
• Customer segmentation
• Market basket analysis
Common Algorithms
• k-Means Clustering
• Hierarchical Clustering
• Apriori
Comparison Table (Exam-Favorite)
Basis Supervised Learning Unsupervised Learning
Training data Labeled Unlabeled
Output Known Unknown
Goal Predict / Classify Discover patterns
Human involvement Required Not required
Accuracy Generally higher Depends on data
Examples Spam detection Customer grouping
Conclusion
• Supervised Learning learns with guidance
• Unsupervised Learning learns by itself
Both are essential for real-world ML tasks.
Short Revision Version
• Supervised → Labeled → Predict
• Unsupervised → Unlabeled → Discover
If you want:
• Reinforcement Learning comparison
• Only 3 / 4 / 7 marks GTU answer
• More real-world examples
Send next question
✔ Linear Regression / SVM
✔ Linear Regression / Support Vector Machine (SVM)
(Exam-oriented, simple & point-wise)
Linear Regression
Definition
Linear Regression is a supervised learning algorithm used to predict continuous values by
establishing a linear relationship between input and output.
Model Equation
𝑦 = 𝑚𝑥 + 𝑐
Key Points
• Predicts numerical output
• Finds best-fit straight line
• Minimizes error (Mean Squared Error)
Examples
• House price prediction
• Marks vs study hours
Advantages
• Simple and fast
• Easy to interpret
Limitations
• Works only for linear relationships
• Sensitive to outliers
Support Vector Machine (SVM)
Definition
Support Vector Machine (SVM) is a supervised learning algorithm that classifies data by
finding an optimal hyperplane with maximum margin.
Key Concepts
• Hyperplane → Decision boundary
• Support Vectors → Closest data points
• Margin → Distance between classes
Key Points
• Used mainly for classification
• Can handle high-dimensional data
• Uses kernels for non-linear problems
Examples
• Face recognition
• Text classification
Advantages
• High accuracy
• Effective for complex data
Limitations
• Slow for large datasets
• Kernel selection is complex
Comparison Table (Very Important for Exam)
Basis Linear Regression SVM
Type Supervised Supervised
Output Continuous value Class label
Basis Linear Regression SVM
Main idea Best-fit line Maximum margin hyperplane
Use Prediction Classification
Complexity Low High
Examples Price prediction Face recognition
Conclusion
• Linear Regression is best for predicting numerical values
• SVM is best for classification with clear boundaries
Short Revision Version (Quick Memory)
• Linear Regression → Predict numbers
• SVM → Classify with margin
If you want:
• Only 3 / 4 / 7 marks GTU answer
• Numerical problem
• Difference with Logistic Regression
Send next question
✔ Neural Network architecture
✔ Neural Network Architecture
Introduction
A Neural Network architecture describes the structure and arrangement of layers and
neurons used to process data and learn patterns.
Main Layers of Neural Network
1. Input Layer
• First layer of the network
• Receives raw input features
• No computation is done here
Example: pixels of an image, marks, age
2. Hidden Layer(s)
• One or more layers between input and output
• Perform main computation
• Extract patterns and relationships
More hidden layers → Deep Neural Network
3. Output Layer
• Final layer of the network
• Produces final prediction
• Can be:
o Class label (Yes/No)
o Numerical value (price, score)
Neuron Structure (Core Unit)
Each neuron performs:
𝑦 = 𝑓(∑𝑤𝑖 𝑥𝑖 + 𝑏)
Where:
• 𝑥𝑖 = inputs
• 𝑤𝑖 = weights
• 𝑏= bias
• 𝑓= activation function
• 𝑦= output
Activation Functions (Examples)
• Sigmoid – output between 0 and 1
• ReLU – removes negative values
• Tanh – output between −1 and 1
Data Flow in Architecture
Input → Weighted sum → Activation → Next layer → Output
Learning in Neural Network
• Error is calculated using loss function
• Weights are updated using backpropagation
• Optimization using gradient descent
Advantages
• Learns complex, non-linear patterns
• High accuracy for large data
Limitations
• Requires large datasets
• High computation cost
Conclusion
Neural Network architecture consists of input, hidden, and output layers, where neurons
process data using weights, bias, and activation functions to learn and predict effectively.
Short Revision Version (Quick Memory)
• Layers → Input → Hidden → Output
• Neuron → 𝑤𝑥 + 𝑏→ activation
• Learning → Backpropagation
If you want:
• Only 3 / 4 / 7 marks GTU answer
• ANN vs CNN vs RNN architecture
• Numerical example
Send next question
✔ CNN vs RNN
✔ CNN vs RNN (Comparison)
Introduction
Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) are deep
learning models designed for different data types and problem domains.
Convolutional Neural Network (CNN)
• Best suited for image and spatial data
• Uses convolution filters to extract local features (edges, shapes)
• Processes data in parallel
• No memory of previous inputs
Examples: Image classification, face recognition
Recurrent Neural Network (RNN)
• Best suited for sequential and time-series data
• Uses feedback (memory) to remember past information
• Processes data step-by-step (sequence-wise)
• Output depends on previous inputs
Examples: Speech recognition, text prediction
Comparison Table (Very Important for Exam)
Basis CNN RNN
Full form Convolutional Neural Network Recurrent Neural Network
Data type Image / spatial data Sequential / time-series data
Memory No memory Has memory (hidden state)
Processing Parallel Sequential
Core operation Convolution Recurrence (feedback loop)
Feature focus Spatial features Temporal (time) dependency
Training speed Faster Slower
Common uses Vision tasks NLP, speech, forecasting
Conclusion
• CNN is ideal for image-related problems
• RNN is ideal for sequence-based problems
Short Revision Version (Quick Memory)
• CNN → Images, no memory
• RNN → Sequences, has memory
If you want:
• Only 3 / 4 / 7 marks GTU answer
• CNN vs RNN vs ANN table
• Examples-based differences
Send next question
✔ LLMs + ChatGPT + Bhashini
✔ LLMs + ChatGPT + Bhashini (Combined Explanation – Exam Oriented)
4
1. Large Language Models (LLMs)
Meaning / Definition
Large Language Models (LLMs) are advanced AI models trained on huge volumes of text
data to understand, generate, and respond to human language.
Key Points
• Based on Deep Learning (Transformer architecture)
• Learn language patterns using attention mechanism
• Can perform multiple NLP tasks with one model
Examples
• ChatGPT
• Gemini
• Krutrim
2. ChatGPT
Concept
ChatGPT is a conversational AI system built on LLMs that can interact with users in natural
language and generate human-like responses.
Key Features
• Understands questions and context
• Generates clear, human-like answers
• Supports learning, writing, coding, and problem-solving
Applications
• Education (exam preparation, concept explanation)
• Customer support chatbots
• Content creation (emails, reports)
• Programming help
3. Bhashini
Concept
Bhashini is an Indian government-backed AI initiative aimed at language inclusivity by
enabling digital services in Indian regional languages.
Role in Inclusivity
• Breaks English-language barrier
• Supports speech-to-text, text-to-speech, translation
• Enables rural and non-English users to access digital services
Applications
• E-governance services
• Education in regional languages
• Voice-based digital access for citizens
Comparison Table (Very Important for Exam)
Basis LLMs ChatGPT Bhashini
Language AI
Type Language model Conversational AI
platform
Core NLP + Speech +
Transformer, Deep Learning LLM-based chatbot
Technology LLMs
Language understanding & Human-like
Main Purpose Language inclusivity
generation conversation
Language Indian regional
Global languages Mostly global
Support languages
Students,
Target Users Developers, researchers Indian citizens
professionals
Govt services,
Example Use Text generation Q&A, coding
education
Conclusion
• LLMs form the foundation of modern language AI
• ChatGPT uses LLMs for intelligent conversation and learning
• Bhashini applies AI for inclusive access to technology in Indian languages
Short Revision Version (Quick Memory)
• LLMs → Core language models
• ChatGPT → Conversational AI using LLMs
• Bhashini → Indian language inclusivity using AI
If you want:
• Only 7-mark GTU answer
• One-page revision notes
• Diagram-based explanation
Send next question
✔ AI Issues & Challenges
✔ AI Issues & Challenges
4
Introduction
While Artificial Intelligence offers many benefits, it also creates serious issues and
challenges related to ethics, privacy, security, and society. Addressing these is essential for
responsible AI use.
Major Issues & Challenges of AI
1. Data Privacy
• AI systems use large amounts of personal data
• Risk of data misuse, leakage, and surveillance
Example: Face recognition without user consent
2. Bias & Discrimination
• AI may learn biased behavior from training data
• Leads to unfair decisions
Example: Biased hiring or loan approval systems
3. Lack of Transparency (Black Box AI)
• AI decisions are difficult to explain
• Users don’t know why a decision was made
4. Ethical Issues
• AI may violate human values
• Questions of right and wrong use
Example: Use of AI in weapons or mass surveillance
5. Job Displacement
• Automation replaces human jobs
• Creates unemployment and skill gap
6. Security Threats
• AI can be misused for:
o Cyber-attacks
o Deepfakes
o Fraud
7. Accountability & Legal Issues
• Unclear who is responsible for AI mistakes
• Lack of strong AI laws and regulations
8. High Cost & Energy Consumption
• AI requires expensive hardware (GPU/TPU)
• High power consumption affects environment
Ways to Address AI Challenges
• Fair and diverse data collection
• Explainable AI models
• Strong data protection laws
• Human supervision
• Ethical AI guidelines
Conclusion
AI issues and challenges include privacy, bias, ethics, jobs, and security. Solving these is
necessary to build safe, fair, and trustworthy AI systems.
Short Revision Version (Quick Memory)
• Privacy
• Bias
• Transparency
• Ethics
• Job loss
• Security
• Cost & laws
If you want:
• Only 4 / 7 marks GTU answer
• Issues vs Challenges table
• Case-study based answer
Send next question