UPSC GS Paper III: Science Tech Artificial Intelligence ML Deep Learning AI Policy India

Must‑Know (20–25 detailed bullets)

• Artificial Intelligence (AI) – simulation of human intelligence processes by machines, especially computer systems; includes learning, reasoning, problem-solving, perception, and language understanding; exemplified by India’s AI in healthcare via NITI Aayog’s pilot projects in Andhra Pradesh and Maharashtra.
• Machine Learning (ML) – subset of AI that enables systems to learn from data without explicit programming; uses algorithms like decision trees and support vector machines; deployed in Aadhaar’s biometric authentication systems for fraud detection.
• Deep Learning – subset of ML based on artificial neural networks with multiple layers; enables image and speech recognition; used in BharatNet for monitoring rural broadband infrastructure via satellite imagery analysis.
• Neural Network – computational model inspired by human brain; consists of input, hidden, and output layers; foundational to deep learning models used in DRDO’s autonomous surveillance drones.
• Natural Language Processing (NLP) – enables machines to understand and generate human language; applied in Bhashini platform under Digital India to support multilingual communication across 22 Scheduled Languages.
• Computer Vision – field of AI enabling machines to interpret visual data; used in Smart Cities Mission for traffic management and crowd monitoring via CCTV analytics.
• Supervised Learning – ML technique where models are trained on labeled datasets; used by RBI in detecting fraudulent banking transactions using historical fraud data.
• Unsupervised Learning – ML technique identifying patterns in unlabeled data; applied by CSIR in drug discovery by clustering molecular structures.
• Reinforcement Learning – ML where agents learn by trial and error using rewards/penalties; explored by ISRO in optimizing satellite orbital maneuvers.
• Generative AI – AI capable of creating text, images, or videos; includes models like ChatGPT; Indian startups like Hanooman AI are developing Indic-language generative models.
• AI Bias – systematic error in AI outputs due to skewed training data; highlighted in 2021 Delhi Police facial recognition system misidentifying minors as criminals due to poor dataset diversity.
• Explainable AI (XAI) – methods to make AI decisions interpretable to humans; mandated in EU’s AI Act; under discussion in India’s draft Digital India Act for algorithmic accountability.
• Turing Test – proposed by Alan Turing in 1950 to evaluate machine intelligence; no AI has passed it conclusively; referenced in UPSC 2020 as a benchmark for strong AI.
• Backpropagation – algorithm for training neural networks by adjusting weights based on error; essential for deep learning; used in AI-based crop yield prediction models by ICAR.
• Overfitting – ML model performs well on training data but poorly on new data; mitigated using regularization techniques; observed in early versions of AI-based credit scoring by fintech firms.
• Edge AI – AI processing done locally on devices rather than cloud; enhances privacy and reduces latency; used in AI-enabled stethoscopes by Indian startup Remidio for rural diagnostics.
• AI in Agriculture – used for soil health monitoring, pest detection, and yield prediction; Soil Health Card Scheme integrates AI via mobile apps like CropIn.
• National Strategy for Artificial Intelligence – released by NITI Aayog in 2018; identifies five sectors: healthcare, agriculture, education, smart cities, and mobility.
• Responsible AI for Social Empowerment (RAISE) 2020 – national summit organized by MeitY to frame India’s AI vision; led to formation of Inter-Ministerial Committee on AI.
• INDIAai – national AI portal launched by MeitY and NASSCOM in 2022; serves as repository for AI resources, startups, and research.
• AI Task Force – constituted by MeitY in 2017; recommended establishing Centres of Excellence (CoEs) in AI; led to AI CoE in Bengaluru under Karnataka government.
• Global Partnership on Artificial Intelligence (GPAI) – India joined in 2020; multilateral initiative to guide responsible AI development; co-chaired by France and Canada.
• Centre for Artificial Intelligence and Robotics (CAIR) – DRDO lab in Bengaluru; develops AI for defense applications like autonomous navigation and surveillance.
• Digital India Corporation (DIC) – formerly Media Lab Asia; supports AI R&D under MeitY; funds projects on voice recognition for regional languages.
• AI in Education – used in adaptive learning platforms like DIKSHA; personalizes content delivery based on student performance analytics.

Difficulty Level

Intermediate – AI concepts are conceptually abstract but frequently tested in combination with government schemes and ethical dimensions; requires clarity on distinctions between AI, ML, and DL.

Common UPSC Traps (3–5 factual traps)

Trap: AI and Machine Learning are synonymous – Fact: AI is the broader field; ML is a subset focused on learning from data (NITI Aayog’s National Strategy for AI, 2018).
Trap: Deep Learning requires only large data, not specialized hardware – Fact: Deep Learning relies on GPUs/TPUs for training neural networks (CAIR, DRDO documentation).
Trap: India has a dedicated AI Act – Fact: India does not have a standalone AI law; regulatory framework is evolving under Digital India Act and IT Rules, 2021 (MeitY, RAISE 2020 outcomes).
Trap: Bhashini is an AI model developed by IITs – Fact: Bhashini is a government platform under Digital India; built using collaborative inputs from CDAC, IITs, and startups.

Practice MCQs (5–7 questions)

Question: Which of the following best describes the primary objective of India’s RAISE 2020 summit?
A) To launch a national AI-powered education platform
B) To establish a legal framework for AI patents
C) To position India as a global hub for responsible AI development
D) To create a central AI regulatory authority
Answer: C
Explanation: RAISE 2020 (Responsible AI for Social Empowerment) was organized by MeitY to showcase India’s AI vision and foster international collaboration.
Why others fail: Option D is tempting as regulation is debated, but no central authority was created post-RAISE.

Question: In the context of artificial intelligence, what is ‘backpropagation’ primarily used for?
A) Data labeling in supervised learning
B) Reducing bias in training datasets
C) Training neural networks by minimizing error
D) Deploying AI models on edge devices
Answer: C
Explanation: Backpropagation adjusts weights in neural networks during training by propagating errors backward; fundamental to deep learning.
Why others fail: Option A is incorrect as data labeling is a preprocessing step, not related to backpropagation.

Question: Which of the following is a key application of computer vision in Indian smart cities?
A) Predicting monsoon patterns using satellite data
B) Monitoring traffic flow using CCTV analytics
C) Generating multilingual government notices
D) Automating tax filing through chatbots
Answer: B
Explanation: Computer vision enables real-time analysis of video feeds for traffic management and surveillance in Smart Cities Mission projects.
Why others fail: Option A involves remote sensing and meteorology, not computer vision per se.

Question: The Bhashini platform, often seen in news, is primarily associated with:
A) AI-based crop monitoring
B) Natural Language Processing for Indian languages
C) Facial recognition for national security
D) Blockchain-based land records
Answer: B
Explanation: Bhashini uses NLP to enable speech-to-speech translation across 22 Indian languages under Digital India.
Why others fail: Option C is incorrect as facial recognition is handled by other systems like those of Delhi Police.

Question: Which of the following correctly pairs an AI technique with its application in Indian governance?
A) Reinforcement Learning – Aadhaar biometric authentication
B) Unsupervised Learning – Clustering of agricultural zones by soil type
C) Generative AI – Real-time air quality monitoring
D) Supervised Learning – Earthquake prediction using seismic data
Answer: B
Explanation: Unsupervised learning identifies hidden patterns; used by ICAR to cluster regions for crop planning based on unlabeled soil data.
Why others fail: Option A is wrong because Aadhaar uses pattern recognition, not reinforcement learning.

Last‑Minute Revision (20–25 one‑liners)

• ⚠️ AI ≠ ML: AI is broad; ML is a subset focused on data-driven learning.
• Deep Learning uses neural networks with multiple hidden layers.
• NITI Aayog released National Strategy for AI in 2018.
• RAISE 2020 was organized by MeitY.
• INDIAai is the national AI portal launched in 2022.
• India joined GPAI in 2020.
• CAIR is a DRDO lab for AI in defense.
• Bhashini enables multilingual NLP under Digital India.
• Edge AI processes data on-device, not on cloud.
• Turing Test proposed in 1950 by Alan Turing.
• Backpropagation trains neural networks by error minimization.
• Overfitting occurs when ML model fails on new data.
• Supervised Learning uses labeled datasets.
• Unsupervised Learning finds patterns in unlabeled data.
• Reinforcement Learning uses reward-penalty mechanism.
• Explainable AI (XAI) ensures transparency in decisions.
• AI bias in Delhi Police FRT system led to wrongful identifications.
• Soil Health Card Scheme uses AI for nutrient recommendations.
• DIKSHA platform uses AI for personalized learning.
• ⚠️ No standalone AI Act in India as of 2023.
• MeitY constituted AI Task Force in 2017.
• Digital India Corporation supports AI R&D.
• Generative AI creates content (text, images, audio).
• Computer Vision used in Smart Cities for traffic monitoring.
• ⚠️ NLP is key to Bhashini, not blockchain.