Ch 17: Artificial Intelligence (AI)

Artificial Intelligence / कृत्रिम बुद्धिमत्ता

Artificial Intelligence (AI) is presented as a branch of computer science that aims to create machines that can think and act like humans.
It enables machines to learn from experience, adjust to new inputs, and perform tasks such as decision making, problem solving and language understanding.
The chapter later connects AI with robotics, Machine Learning, Deep Learning, NLP, Computer Vision and Reinforcement Learning.

हिंदी सार

AI कंप्यूटर विज्ञान की वह शाखा है जिसका उद्देश्य ऐसी मशीनें बनाना है जो मानव जैसी सोच, सीखने, निर्णय लेने, समस्या समाधान और भाषा समझने जैसी क्षमताएँ प्रदर्शित कर सकें।

Major Coverage

HistoryRobotsAI TypesDeep Learning

AI Winters & modern era
Famous world & Indian robots
Capabilities + functionalities
Neural networks and training process

Source-derived visual from the AI history milestones listed in the PDF.

History of AI / AI का इतिहास

Period / Year	Source Points
Before 1940	Greek philosophers such as Aristotle and Plato discussed logic and reasoning; the idea of artificial brains appeared in literature and early science fiction.
1943	McCulloch and Pitts created the first mathematical model of a neuron.
1950	Alan Turing proposed the Turing Test.
1951	Marvin Minsky built the first neural-network machine called SNARC.
1956	John McCarthy coined the term Artificial Intelligence during the Dartmouth Conference. The source calls 1956 the Birth Year of AI.
1956–1974	Growth period; rapid research and high expectations. Early programs solved simple logic and mathematical problems.
1955 / 1957	Logic Theorist (1955) — first AI program; General Problem Solver (1957).
1974–1980	First AI Winter — funding and research declined because of limited computer power and failed expectations.
1980–1987	Revival with Expert Systems; example: MYCIN for medical diagnosis.
1987–1993	Second AI Winter — high costs and low efficiency reduced interest and funding.
1997	IBM Deep Blue defeated world chess champion Garry Kasparov.
2011	IBM Watson won the Jeopardy! quiz show.
2012	Rise of Deep Learning and Neural Networks.
2016	Google DeepMind's AlphaGo beat Lee Sedol in Go.
Present	AI is used in self-driving cars, healthcare, finance, education and robotics.

Quick Timeline Hook

1943 neuron model → 1950 Turing Test → 1956 AI term/Dartmouth → AI Winters → 1997 Deep Blue → 2011 Watson → 2016 AlphaGo.

AI & Robots / AI और रोबोट

AI gives robots the ability to see, hear, understand and interact with the world.
A robot is described as a programmable machine that performs tasks automatically or semi-automatically, often controlled by AI, sensors or human input.

Source-derived visual using the robot names and facts provided in the chapter.

Robot	Developer / Place	Type / Key Source Facts	Use
Sophia	Hanson Robotics (Hong Kong)	Humanoid; source calls it the world's first robot to get citizenship — Saudi Arabia, 2017; AI-powered facial expressions and human-emotion recognition.	Education, research, media
Manav	A-SET Training & Research Institute, New Delhi	Humanoid; walk, talk, dance; smartphone/Wi-Fi controlled; built using 3D-printed parts. Source also says it was developed in late Dec 2014 by Diwakar Vaish.	Education, research, exhibitions
Mitra	Invento Robotics, Bengaluru	Recognizes faces and speech; multilingual; source notes the Global Entrepreneurship Summit, Hyderabad.	Banks, hospitals, events
Rashmi	Ranjit Shrivastava, Ranchi	Indian humanoid; source describes it as India's first lip-synchronizing humanoid; Hindi, English, Bhojpuri, Marathi.	Chatting, customer service, social interaction
Shalu	Dinesh Patel	Humanoid made from waste materials; source says 47 languages (9 Indian + 38 foreign); mentioned in World Book of Records.	Education, AI teaching assistant
Asimov & RoboCop	DRDO	Military robots; remotely controlled; sensors and cameras.	Bomb disposal, surveillance, defence operations
Robot Nurse / Robot Waiter	Used in Indian hospitals/restaurants	Voice-controlled and AI-enabled; source example: “Seva” robot.	Food/medicine delivery and patient assistance

Major Techniques, Goals & Applications

Major Techniques

Neural Networks
Fuzzy Logic
Genetic Algorithms
Search Algorithms
Heuristic Methods

Goals of AI

Automate human tasks
Enhance efficiency
Improve decision making
Enable adaptive learning
Create intelligent systems

Applications

Virtual assistants
Self-driving cars
Healthcare diagnosis
Facial recognition
Financial fraud detection
Robotics & smart homes

AI Core Concepts / AI के प्रमुख सिद्धांत

Concept	Source Description
Machine Learning (ML)	AI systems learn from data and improve performance.
Deep Learning (DL)	Subfield of ML based on neural networks such as CNN and RNN.
Natural Language Processing (NLP)	AI capable of understanding and processing human language.
Computer Vision	AI capable of understanding and making decisions from visual data.
Reinforcement Learning (RL)	Learning process in which AI learns from experience and acts to obtain a reward.
CNN	Deep-learning neural network mainly used for image recognition and computer vision.
RNN	Neural network for sequential or time-series data, where previous output affects the next.

Types of Artificial Intelligence / AI के प्रकार

Two classifications used in the PDF: by capabilities and by functionalities.

Based on Capabilities

Narrow AI (Weak AI): Designed for specific tasks; cannot perform tasks outside its programming; source says most current AI systems are Narrow AI. Examples: Siri, Google Search, recommendation systems.
General AI (Strong AI): Can perform intellectual tasks like a human; can learn, reason and adapt; still theoretical in the source.
Super AI: Surpasses human intelligence in all fields; purely hypothetical at present.

Based on Functionalities

Reactive Machines: React to specific inputs; no memory of past experiences. Example: IBM Deep Blue.
Limited Memory: Uses past data for decisions. Examples: self-driving cars and virtual assistants.
Theory of Mind: AI that can understand emotions, thoughts and beliefs; still in research.
Self-aware AI: Consciousness and self-awareness; can think about itself; theoretical/future concept.

Advantages & Challenges

Advantages

Reduces human error
Works in risky environments
Available 24×7
Automates repetitive tasks
Faster decision making

Challenges

High development cost
Ethical issues
Risk of unemployment
Data privacy concerns
Bias in decision making

Deep Learning / डीप लर्निंग

Deep Learning is described as a subset of Machine Learning using neural networks with many layers to model complex patterns in data.
It is inspired by the human brain and is used for tasks such as image recognition and natural language processing.
A deep neural network has multiple hidden layers between input and output; more layers allow the network to learn more complex representations.

Source-derived deep neural network structure: input, hidden layers and output.

Key Concepts of Deep Learning

Concept	Meaning in the Source
Neuron / Perceptron	Basic unit of a neural network; receives inputs, multiplies them with weights, adds bias and applies an activation function.
Input Layer	Takes raw input data.
Hidden Layers	Intermediate layers that extract patterns/features.
Output Layer	Produces the final result.
Weights & Bias	Parameters the model learns during training to make predictions.
Activation Function	Adds non-linearity; source lists ReLU, Sigmoid and Tanh.
Forward Propagation	Moves inputs through the network to obtain the output.
Backpropagation	Updates weights to improve accuracy.

Types of Deep Learning Networks

FNN, CNN, RNN, LSTM and GAN — all listed in the source chapter.

Network	Source Description / Use
FNN	Simplest neural network; information flows only forward; no loops/cycles; used in simple classification tasks.
CNN	Used specifically for image processing and computer vision; convolutional layers perform feature extraction. Examples: face recognition, object detection.
RNN	Used for sequential data such as text, speech and time-series; has memory to store past information. Examples: language translation, speech recognition.
LSTM	Advanced version of RNN for long-term dependencies; useful for chatbots, sentiment analysis and stock prediction.
GAN	Generates new and realistic-looking data using two networks: a generator and a discriminator.

Activation Functions & Training Process

Activation Functions

Sigmoid: Used for binary classification; source states output is between 0 and 1.
ReLU: Most common; if input < 0 output = 0, otherwise output = input.
Softmax: Source lists it for multi-class classification. The PDF wording about its output range is retained only as source wording and not silently corrected.

Training Process

Forward Propagation: Data flows through layers.
Loss Calculation: Measures the error.
Backpropagation: Adjusts weights using optimization algorithms such as Gradient Descent.
Optimization: Methods such as Adam and RMSProp are used to minimize errors.

Deep Learning Applications

Computer Vision — Face Recognition/Object Detection • NLP — Chatbots/Translation • Healthcare — Disease Prediction/Medical Imaging • Autonomous Vehicles — Self-Driving Cars • Robotics — Smart Robots.

Quick Recall / त्वरित पुनरावृत्ति

One-Line Revision

AI: Machines that learn, adapt and perform human-like intelligent tasks.
1956: John McCarthy coined “Artificial Intelligence” at Dartmouth; source calls it AI's birth year.
AI Winters: 1974–1980 and 1987–1993.
Modern milestones: Deep Blue (1997), Watson (2011), Deep Learning rise (2012), AlphaGo (2016).
Capabilities: Narrow → General → Super.
Functionalities: Reactive → Limited Memory → Theory of Mind → Self-aware.
AI core: ML • DL • NLP • Computer Vision • RL.
Deep Learning: many-layer neural networks.
CNN: image/computer vision.
RNN: sequential/time-series data.
LSTM: long-term dependencies.
GAN: generator + discriminator.
Training: Forward → Loss → Backpropagation → Optimization.

Exam Traps

1956 ≠ Turing Test year. Turing Test is listed at 1950; 1956 is the Dartmouth/AI-term milestone.
Narrow AI ≠ General AI. Narrow AI is task-specific; General AI is described as human-level general intellectual capability and theoretical in the source.
CNN vs RNN: CNN → images/vision; RNN → sequential/text/speech/time-series.