Machines That Learn: Artificial Intelligence

Artificial intelligence is the ability of machines to perform cognitive tasks such as thinking, perceiving, learning, problem-solving and decision-making. For UPSC, AI sits at the meeting point of science and technology, governance and ethics: it powers crop advisories, cancer screening and fraud detection, yet forces hard questions about bias, privacy, jobs and even the nature of truth in public life.

The building blocks: how AI works

Machine learning is the subfield of AI in which computers learn patterns from data instead of following hand-written rules, so performance improves as more data arrives. A spam filter that gets better at catching junk mail the longer you use it is everyday machine learning at work.

Neural networks are computing models loosely inspired by the human brain, built from layers of interconnected nodes that detect patterns in images, speech and text. Deep learning is simply machine learning done with very large neural networks, and it powers modern image recognition, voice assistants and language models.

Natural language processing (NLP) is the branch of AI that lets machines understand and generate human language, from chatbots and translators to tools like Digital India Bhashini, which offers speech and translation services across 22 Indian languages. Computer vision is the branch that interprets visual input, enabling facial recognition, medical scan analysis and quality inspection on factory floors.

Generative AI is AI that creates new content, such as text, images, audio or video, after learning the patterns of its training data. Generative Adversarial Networks (GANs) are one engine behind it: two neural networks, a generator and a discriminator, compete against each other until the forgeries become nearly indistinguishable from reality. This same architecture powers deepfakes, which is why understanding it matters for governance.

Two textbook definitions are worth quoting verbatim. Deep learning is a form of machine learning that utilizes artificial neural networks (ANN) to acquire knowledge from data. An LLM, or large language model, is a type of artificial intelligence model that has been trained through deep learning algorithms to recognize, generate, translate, and/or summarize vast quantities of written human language and textual data. LLMs are foundation models: large models trained on broad data that can be adapted to many downstream tasks, which is why one model can write essays, code and translations.

Not all neural networks are alike. Shallow neural networks contain only one layer of neurons: simple to train but weak at complex patterns. Deep neural networks stack many layers and learn intricate patterns shallow nets cannot. Convolutional neural networks (CNNs) are designed for image recognition, learning spatial relationships between pixels to identify objects. Recurrent neural networks (RNNs) are tailored for sequence modelling, capturing temporal relationships between sequence elements, which makes them suited to language and time-series data before transformers took over.

The economic stakes most often cited are three dated estimates, and each should carry its label. A PwC study (2017) predicted AI could add 14 percent, about $15.7 trillion, to global GDP by 2030. NITI Aayog's National Strategy for AI (2018) estimated large-scale AI adoption could boost India's annual growth rate by 1.3 percentage points by 2035. And an MIT study (2023) found generative AI assistants raised worker productivity by about 14 percent in customer-support tasks. Treat all three as projections from their year, not as current measurements.

AI at work: from farms to hospitals

In healthcare, AI assists early diagnosis, personalised treatment and drug discovery. Indian startups illustrate the frontier: Niramai's Thermalytix screens for breast cancer using thermal imaging, Qure.ai reads chest X-rays for tuberculosis and pneumonia with very high accuracy, and the telemedicine platform e-Sanjeevani uses AI-enabled tools to bridge the rural-urban gap in doctor access.

In agriculture, AI advisories combine weather, soil and satellite data to tell farmers when to sow, irrigate and spray, while in governance the RBI Innovation Hub's MuleHunter.AI uses AI to detect suspicious money-mule transactions in real time. Retailers use recommendation engines, banks use credit-scoring models, and educators use adaptive learning platforms that personalise lessons to each student's pace.

In manufacturing, AI adoption in India jumped sharply as factories deployed predictive maintenance, in which sensors and AI forecast equipment failures before they happen, cutting unplanned downtime. AI-powered vision systems spot micro-defects on production lines in real time, digital twins, which are virtual replicas of physical factories, simulate layout and energy use, and collaborative robots (cobots) work safely alongside humans on repetitive or physically demanding tasks.

India's AI-in-healthcare story has its own institutional layer. In April 2024 the World Health Organization launched S.A.R.A.H., a generative AI prototype for digital health promotion, marking a milestone in public-health AI deployment. In India, iOncology.ai, developed jointly by AIIMS New Delhi and C-DAC, is an indigenous AI platform for early detection, diagnosis and treatment planning for breast and ovarian cancers, showcased in May 2026 and integrated with the Ayushman Bharat Digital Mission ecosystem. The Maharashtra government with NITI Aayog launched the International Centre for Transformational AI (ICTAI) focused on AI-driven rural healthcare solutions. And in March 2023 the Indian Council of Medical Research released its Ethical Guidelines for AI in Biomedical Research and Healthcare, India's home-grown rulebook for the field.

The concerns examiners probe are practical: data privacy and security, since AI needs large datasets that can leak; algorithmic bias from models trained on narrow data; the black box nature that hides how decisions are made; regulation and accountability gaps over who is liable for errors such as in AI-assisted robotic surgery; affordability and access for rural and underserved regions; and fear of job loss in administration and diagnostics. India's ICMR guidelines plus WHO's principles are the twin foundations mains answers should cite for safe, equitable AI in health.

Governing with AI

Governance with AI, sometimes called GovAI, is the use of artificial intelligence inside the state itself: to make policy, deliver welfare, detect fraud and talk to citizens. India is a natural laboratory because its Digital Public Infrastructure already gives AI rails to run on: Aadhaar for identity, UPI for payments, DigiLocker for documents, CoWIN for vaccination, e-Sanjeevani for telemedicine and DigiYatra for travel. AI layered on DPI turns registries and transaction trails into targeted, measurable governance.

The challenges are as Indian as the opportunity. Data privacy and security demand robust protection, since episodes like Aadhaar data leaks exposed vulnerabilities. The digital divide is stark: rural internet penetration stood at only about 37 percent (TRAI, 2023), so AI governance can bypass the very citizens it should include. There is a skill gap in the bureaucracy, with low digital literacy among lower-tier officials, and AI infrastructure is costly to build and maintain. The way forward mapped by policy: secure data infrastructure with encryption, workforce reskilling through programmes like FutureSkills PRIME, transparent and explainable AI mandated for governance use, cyber-regulation frameworks that keep IT law current with algorithmic accountability, and public-private collaboration so startups and academia co-create public-welfare tools.

India's big bet: the IndiaAI Mission

The IndiaAI Mission, approved by the Union Cabinet in March 2024 with an outlay of Rs 10,371.92 crore over five years, is India's flagship programme for artificial intelligence. Its guiding vision is summed up as making AI in India and making AI work for India, with an overarching goal of AI for All: democratising access to computing power, building indigenous models, and pushing AI adoption in agriculture, healthcare, governance, disaster management and climate resilience.

The mission rests on seven pillars. The IndiaAI Compute pillar provides subsidised high-end GPUs to startups and researchers: from an initial target of 10,000 GPUs, India has onboarded over 38,000 GPUs, available at about Rs 65 per hour. The foundation models pillar supports indigenous large models such as Sarvam AI and BharatGen, trained on Indian languages and priority sectors. AI Kosh is the national datasets platform hosting hundreds of curated non-personal datasets, while the application development, FutureSkills, startup financing and safe-and-trusted-AI pillars complete the ecosystem.

This builds on the National Strategy for Artificial Intelligence (2018) of NITI Aayog, which first articulated the AI for All vision for healthcare, agriculture and education, and on the IndiaAI portal run by MeitY as a one-stop platform for AI resources, datasets and startups. Together with 58 approved AI Centres of Excellence across states, the architecture treats AI as public infrastructure rather than a private luxury.

Two infrastructure facts anchor the mission's compute story. AIRAWAT, installed at C-DAC Pune under the National Supercomputing Mission, is India's AI-specific supercomputing system, built for large AI workloads rather than general scientific computing. And in February 2026, Sarvam AI, selected under the IndiaAI Mission to build a sovereign model, launched a 105-billion-parameter foundational model trained from scratch in India with zero external data dependency, the largest model trained indigenously at the time. Note the fact carefully: it is a 105-billion-parameter model, not a product called Sarvam-1B. On the global stage, India was a founding member of the Global Partnership on AI (GPAI) in 2020 alongside 15 other countries, set up to build frameworks for the responsible use of emerging technologies.

Trustworthy AI: the deepfake test

Trustworthy AI means AI systems that are transparent, accountable, safe and aligned with human values, so that citizens can rely on them in welfare delivery, policing, recruitment, credit and healthcare. For India, with its scale and diversity, a biased model deployed nationally could affect millions, which is why the IndiaAI Mission dedicates an entire pillar to safe and trusted AI.

Deepfakes are AI-generated media that realistically mimic the appearance, voice or behaviour of a real person or event. They are a sophisticated form of synthetically generated information (SGI), which Indian law now defines as information that appears reasonably authentic but has been created or altered using a computer resource in ways that could pass for the real thing. Beyond face-swaps and voice cloning, deepfakes create what experts call the liar's dividend: once fakes flood the public sphere, even genuine evidence can be dismissed as fake, eroding trust in news, elections and institutions.

India answered with the February 2026 amendment to the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021. The amendment writes SGI into law and requires intermediaries to ensure clear labelling and traceable metadata for AI-generated content, so users can identify synthetic material. Timelines were sharpened dramatically: unlawful content flagged by a court or government direction must go down within 3 hours, down from 36 hours, and complaints in sensitive categories such as non-consensual imagery and impersonation must be acted on within 2 hours. Platforms that miss these duties risk losing their safe harbour protection under Section 79 of the IT Act, 2000, which shields intermediaries from liability for user content.

Watermarking is the technique of embedding visible or invisible markers in AI-generated content to declare its synthetic origin, while provenance metadata is the durable record of how a file was made that travels with it across platforms. On the detection side, tools analyse biological signals that fakes often lack, such as blood flow patterns in skin, and experts recommend teaching media forensics in schools so citizens learn to spot artefacts like unnatural blinking and lip-sync errors. Technology, law and literacy must advance together, because no single fix can carry the load.

The standard definition is worth quoting verbatim: deepfakes are digital media, video, audio and images, edited and manipulated using artificial intelligence; basically hyper-realistic digital falsification. Their cheaper cousins, shallow fakes or cheap fakes, are pictures, videos and voice clips altered without AI, using simple editing tools such as Photoshop: easier to make and easier to spot, but still effective at mis-captioning and mis-contextualising real footage. Beyond pornography, which a 2019 Deeptrace study found accounted for about 96 percent of deepfake videos online at the time, deepfakes enable character assassination, erosion of trust in media, threats to national security and use by non-state actors, plus the liar's dividend already covered above.

The response toolkit has five parts: enhanced media literacy so citizens pause before sharing; meaningful regulation built with industry and civil society, which India began with the SGI labelling rules; technology solutions for detection, authentication and watermarking; individual responsibility in consumption and sharing; and dedicated detection research, on the lines of DARPA's deepfake-detection programmes in the United States, so that forensic tools keep pace with generation tools.

The world writes the rulebook

Globally, AI governance is converging around risk-based regulation. The European Union's AI Act (2024) classifies AI systems by the risk they pose and bans unacceptably risky uses such as social scoring. The OECD AI Principles were the first international standard for responsible AI, the UNESCO Recommendation on the Ethics of AI (2021) was adopted by 193 member states, and the Global Partnership on AI (GPAI), in which India is a member, coordinates collaborative research and policy.

India moved the conversation from safety to development by hosting the AI Impact Summit 2026 in New Delhi in February 2026, the first global AI summit hosted in the Global South. Organised by MeitY at Bharat Mandapam, it adopted the New Delhi Declaration on AI Impact, endorsed by 92 countries and international organisations and guided by the principle of Sarvajan Hitaya, Sarvajana Sukhaya (welfare for all, happiness for all). The Declaration is structured around seven chakras: democratising AI resources, economic growth and social good, secure and trusted AI, AI for science, access for social empowerment, human capital development, and resilient, efficient and innovative AI systems. The summit drew investment commitments over $200 billion and launched shared platforms such as the Global AI Impact Commons, positioning India as a bridge between technology-rich and technology-seeking nations.

At the same summit, Prime Minister Modi unveiled the MANAV vision, India's human-centric framework for AI, with MANAV meaning human. The acronym unpacks as: M for moral and ethical systems; A for accountable governance with transparent rules and robust oversight; N for national sovereignty, meaning whose data, his right; A for accessible and inclusive AI, a multiplier not a monopoly; and V for valid and legitimate systems that are lawful and verifiable. For mains answers, MANAV is the one-line summary of India's AI governance philosophy.

Pax Silica is the US-led international collaboration, launched in December 2025, for secure and resilient AI ecosystems across the full silicon stack: critical minerals, semiconductor fabrication, AI systems and deployment infrastructure. India formally joined at the AI Impact Summit in February 2026, and a second summit in Washington in June 2026 brought 35 nations together around a Joint Statement on AI Opportunity. It matters for India because it links AI ambition to supply-chain security and trusted manufacturing partnerships.

For AI in warfare, the Responsible AI in the Military Domain (REAIM) process sets the emerging norms: develop international standards for military AI, demand transparency and accountability in development and use, ensure systems are not biased or discriminatory, protect the privacy and security of training and operational data, and promote responsible research and development. Together with the GPAI and the New Delhi Declaration, REAIM completes the governance alphabet examiners expect.

The human capital question

AI shifts rather than simply destroys work, and that makes skills the decisive policy lever. Building locally relevant AI needs dual expertise that combines algorithmic understanding with software engineering, which is why the IndiaAI Mission funds 543 FutureSkills Data and AI labs in tier-2 and tier-3 cities and 686 AI fellowships to democratise deep-tech learning. Programmes such as FutureSkills Prime by MeitY and the Skill India Mission retrain workers in AI, data and robotics, while the National Education Policy 2020 pushes flexible curricula, multiple entry-exit options and industry apprenticeships.

The deeper shift is in what humans are valued for. As AI absorbs routine retrieval and summarisation, human contribution moves upward toward judgement, direction, domain expertise and synthesis: the doctor who frames the right diagnostic question, the engineer who decomposes a complex problem into steps a model can execute. Policymakers must also protect experience-intensive physical roles, from nursing to advanced metalwork, that AI cannot easily replicate. The World Economic Forum's Future of Jobs Report 2025 estimates automation could displace about 92 million jobs globally by 2030 while creating about 170 million new ones, which is why reskilling is treated as infrastructure, not welfare.

Challenges and the way forward

Four challenges recur across every AI debate. Bias and discrimination arise when models trained on skewed data misidentify or disadvantage particular groups, a documented problem in facial recognition. The black box problem means many deep-learning systems cannot explain their decisions, which undermines accountability in lending, hiring and policing. Privacy and surveillance risks grow as AI tracks movement and behaviour at population scale. And the concentration of compute and talent in a few advanced economies threatens to turn the Global South into consumers rather than creators of the technology.

The way forward, as mapped by Indian policy, has five strands: implement AI governance through the IndiaAI Mission with clear accountability; adopt explainable-AI tools and ethical guidelines aligned with global frameworks; protect privacy through the Digital Personal Data Protection Act, 2023 and techniques like federated learning; expand skilling so the workforce transitions with the technology; and deepen global partnerships through GPAI and the New Delhi Declaration framework. The measure of success is not machine capability but human upliftment.

One philosophical lens enriches ethics answers. Immanuel Kant's ethics centres on human autonomy, rationality and moral duty: persons must be treated as ends, never merely as means. Applied to AI in governance, it raises a sharp question: when an opaque model decides welfare eligibility or policing priorities, is the citizen still being treated as a rational agent owed reasons, or merely as data to be processed? The Kantian answer demands explainability and human oversight as moral requirements, not just technical features, which is why accountability and the right to an explanation recur in every responsible-AI framework.

AI speaks Indian: language as infrastructure

One of India's most distinctive AI bets is on language. Digital India Bhashini, the national language-translation mission, builds speech and translation tools across 22 Indian languages so that a farmer in Odisha or a shopkeeper in Madurai can use digital services in her own tongue. Project Vaani is assembling a 150,000-hour dataset of Indian speech to capture the country's linguistic diversity, because models trained only on English fail most Indians. When AI is layered onto Digital Public Infrastructure such as UPI, ONDC and Aadhaar, its benefits flow through rails citizens already use: voice-based payments, AI crop advisories over basic phones, and automated grievance redressal in local languages.

The agentic turn: AI that acts

Agentic AI is artificial intelligence that does not just answer but acts: systems that perceive their environment, plan multi-step tasks and execute them with minimal human direction. A conventional chatbot waits for a prompt; an agentic system breaks a goal like preparing a field survey into steps, calls tools, checks results and adjusts. The 2026 mains question on agentic AI, its working, applications, advantages and risks, shows how fast this has moved from lab to syllabus.

The working loop has four stages:

• Perception: sensing data and context from the environment.
• Reasoning and planning: decomposing the goal into ordered steps.
• Action: using tools, APIs or physical actuators to execute each step.
• Reflection: evaluating outcomes and correcting course.

Applications already span customer-service agents that resolve tickets end to end, coding agents that write and test software, and research agents that run literature reviews. The risks scale with autonomy: an agent that can act can also err at machine speed, raising questions of accountability when no human approved each step, security when agents are manipulated through their inputs, and concentration when only a few firms can train capable agents. Two newer risks sharpen the picture: red-teaming, the deliberate practice of trying to break or misuse an AI system in a controlled setting to expose blind spots before attackers do; and social sycophancy, the tendency of models to over-affirm a user's views, even unethical ones, to win trust, a perverse incentive that deepens manipulation. The convergence of AI-designed genomic sequences with accessible DIY biology kits adds a biosecurity dimension that links AI governance directly to biotechnology.

Open or closed: who should own AI's brains

The Economic Survey 2025-26 framed India's AI debate around a fork in the road: open versus proprietary models. Open models publish their weights for anyone to download, adapt and build on, lowering cost and dependence; proprietary models keep weights closed, offering control, safety filters and commercial support at the price of dependence on the provider.

The Survey's policy takeaways: strong domestic demand and digital inclusion should steer the choice, with compute, data and talent as the binding constraints. For India, the question is which mix builds domestic capability fastest without locking public systems into foreign-controlled infrastructure.

Key Terms

• Artificial intelligence: Artificial intelligence is the field of computer science concerned with building systems that perform tasks requiring human-like intelligence, such as learning from data, reasoning, perceiving images and speech, and making decisions. Its main approaches include machine learning, deep learning, and generative models. AI is now a strategic technology with applications across governance, health, agriculture, and defence, raising questions of ethics, jobs, and regulation. Example: India's INDIAai Mission, approved in 2024 with an outlay of over Rs 10,000 crore, aims to build domestic AI compute infrastructure, datasets, and startup support.
• Machine learning: A branch of artificial intelligence in which algorithms learn patterns from data and improve their performance without being explicitly programmed for every rule. Its main families are supervised learning (training on labelled examples), unsupervised learning (finding structure in unlabelled data) and reinforcement learning (learning by trial and reward). Example: Satellite imagery classified with machine learning is used to map mangrove cover change, crop health and land degradation at national scale.
• Neural networks: Computing systems inspired by the brain, made of layers of interconnected nodes (neurons) that transform inputs through weighted connections. They learn by iteratively adjusting those weights, typically through backpropagation, to minimise prediction error, and stacked deep networks of this kind are the engine of modern deep learning. Example: Image recognition systems that identify objects in photographs and large language models that generate text are both built on neural networks.
• Deep learning: Deep learning is a branch of machine learning that uses artificial neural networks with many layers (deep networks) to learn patterns directly from large datasets. Unlike traditional models that need hand-crafted features, deep networks discover their own representations, which is why they excel at tasks like image recognition, speech processing and language understanding. Its recent breakthroughs rest on three ingredients: massive datasets, powerful GPUs, and architectures such as convolutional networks and transformers. Example: AlphaFold, which predicts protein structures with near-experimental accuracy, and the transformer models behind modern language AI are both deep learning systems.
• Digital India Bhashini: Digital India Bhashini is the government's AI-led language technology mission under the Digital India programme, building open speech and text translation tools for India's languages. It aims to make digital services and content available in all 22 scheduled languages through shared datasets, open-source models and APIs that startups and government apps can plug into. The mission rests on the idea that language should not be a barrier to accessing the internet, governance or education. Example: A farmer can speak a query in Marathi and receive a government scheme's answer translated in real time through a Bhashini-powered voice bot.
• Computer vision: Computer vision is the branch of artificial intelligence that enables machines to extract meaningful information from images and video. Its core tasks include image classification, object detection, image segmentation and facial recognition, powered largely by convolutional neural networks and other deep-learning models. It underpins applications from medical imaging and autonomous vehicles to industrial quality inspection and satellite-based crop monitoring. Example: Automatic number-plate recognition cameras in Indian cities use computer vision to read vehicle plates for traffic enforcement and tolling.
• generative AI: Generative AI is artificial intelligence that produces new content, text, images, audio or video, from patterns learned across vast training data; it powers chatbots like ChatGPT and image generators. For governance it cuts both ways: it can draft, translate and summarise public services at scale, yet enables deepfakes, misinformation and copyright disputes. India regulates it through the IT Act, the DPDP Act, 2023 and IT Rules rather than a dedicated AI statute, a live UPSC debate. Example: ChatGPT-style chatbots for citizen services on one side, and AI-generated deepfake videos in election campaigns on the other.
• healthcare: Healthcare is the organised system of services for maintaining and improving people's health, covering prevention, diagnosis, treatment, and rehabilitation. It is one of the sectors most transformed by new technology: artificial intelligence now assists in medical imaging, drug discovery, and hospital management, while robots assist in surgery and patient care. For public policy, healthcare sits at the intersection of technology, cost, and equitable access. Example: AI-based tools that read X-rays and scans to help radiologists detect diseases at an earlier stage.
• Niramai's Thermalytix: An AI-powered breast cancer screening solution developed by the Bengaluru health-tech startup Niramai, founded by Geetha Manjunath. It combines a high-resolution thermal sensing device with cloud-hosted artificial intelligence analytics that detect abnormal temperature patterns on the chest, offering a portable, radiation-free, non-contact and privacy-sensitive alternative to mammography. Example: Clinical studies indicate the system can detect tumours as small as 4 millimetres and has received the European CE mark for use across Europe.
• Qure.ai: Qure.ai is a Mumbai-based health-technology company that uses artificial intelligence to interpret medical scans. Its deep-learning software reads chest X-rays to detect tuberculosis, pneumonia and other lung conditions with high accuracy. It is an example of AI being deployed for affordable healthcare in settings where trained radiologists are scarce. Example: Qure.ai's chest X-ray software screens patients for tuberculosis, supporting TB detection drives in India.
• e-Sanjeevani: e-Sanjeevani is the national telemedicine platform of the Ministry of Health and Family Welfare, described as the world's largest telemedicine deployment in primary healthcare. It has two variants: eSanjeevani AB-HWC, a doctor-to-doctor service linking Health and Wellness Centres (spokes) with specialist hubs, and eSanjeevaniOPD, a patient-to-doctor service scaled up during the COVID-19 pandemic. It crossed 10 crore teleconsultations by early 2023, with a majority of beneficiaries being women, and is a pillar of India's digital health ecosystem. Example: A patient at a rural Health and Wellness Centre consulting a specialist at a district hospital hub through a paramedic-assisted video call, with an e-prescription issued at the end of the session.
• Agriculture: For UPSC, agriculture is India's largest employer, engaging close to half the workforce, while contributing roughly 18 percent of gross value added, a structural gap that explains rural distress and the push for allied sectors and food processing. It depends heavily on the monsoon and is shaped by MSP, subsidies, and irrigation policy. For UPSC, it is the core of GS-3 economy and connects to environment and social issues. Example: null.
• governance: Governance refers to the structures, rules, and processes through which decisions are made, implemented, and overseen, covering both how governments manage public affairs and how organisations regulate themselves. In the technology context, it means the policies, standards, and accountability mechanisms that keep new systems safe, fair, and transparent. Good governance balances innovation with responsibility, so that harms are prevented without blocking progress. Example: The European Union's AI Act (2024), the world's first comprehensive law on artificial intelligence, uses a risk-based framework to govern AI systems.
• manufacturing: Manufacturing is the sector of the economy that converts raw materials and components into finished goods through industrial processes, from factories making cars and chips to plants producing medicines. It is central to economic development because it creates large-scale employment, drives exports and pulls along services and logistics. For UPSC, manufacturing links industrial policy, employment, technology and self-reliance, which is why programmes like Make in India and the PLI schemes target it. Example: AI adoption in Indian manufacturing jumped as factories deployed predictive maintenance, where machine-learning models analyse sensor data to warn of equipment failure before a production line stops.
• predictive maintenance: Predictive maintenance uses artificial intelligence and sensor data to forecast when industrial equipment is likely to fail, so repairs happen before breakdowns occur. Machine-learning models detect subtle patterns in vibration, temperature or current draw that precede faults. It reduces unplanned downtime and maintenance costs compared with fixed-schedule servicing or run-to-failure strategies. Example: Airlines analysing jet-engine sensor data to replace components before in-flight failures occur, avoiding costly unscheduled grounding.
• digital twins: A digital twin is a virtual replica of a physical object, system or process that is continuously updated with real-time data from sensors, allowing simulation, monitoring and prediction of behaviour. Unlike a static model, it mirrors its physical counterpart live, so engineers can test changes, predict failures and optimise performance before acting in the real world. Applications range from manufacturing plants and aircraft engines to city-scale twins used for urban planning and disaster management. Example: A digital twin of a city's traffic network, fed by live sensor data, used to test the effect of new signal timings before changing them on the streets.
• IndiaAI Mission: The IndiaAI Mission is the national programme approved by the Cabinet in March 2024 with an outlay of 10,371.92 crore rupees over five years to build India's artificial intelligence ecosystem. Its seven pillars include shared GPU compute capacity, indigenous foundational models, datasets, startup financing, and safe AI, implemented by the IndiaAI division under the Digital India Corporation. UPSC relevance: technology sovereignty and digital public infrastructure. Example: The compute pillar's target of 10,000-plus GPUs offers startups subsidized access to high-end AI infrastructure.
• AI for All: The guiding vision of the IndiaAI Mission (2024), summarised as making AI in India and making AI work for India. It means democratising access to computing power, building indigenous models and deploying AI in agriculture, healthcare, governance, disaster management and other priority sectors. Example: the vision builds on NITI Aayog's 2018 National Strategy for Artificial Intelligence, which first articulated AI for All for healthcare, agriculture and education. Example: Rooted in NITI Aayog's 2018 National Strategy for Artificial Intelligence, the first policy document to articulate AI for All.
• IndiaAI Compute pillar: One of the seven pillars of the IndiaAI Mission, it envisions building a high-end, scalable AI computing ecosystem through public-private partnership so that Indian startups, researchers and enterprises get affordable access to computing power. The pillar originally targeted infrastructure of 10,000 or more graphics processing units (GPUs), offered at subsidised rates to mission beneficiaries. Example: By mid-2026 over 38,000 GPUs had been empanelled through cloud service providers under the compute pillar and made available to startups and researchers at subsidised hourly rates.
• foundation models pillar: One of the seven pillars of the IndiaAI Mission, the government's flagship programme to build India's artificial intelligence ecosystem. This pillar funds the development of indigenous foundation models, large AI models trained on vast datasets that can be adapted for many tasks, including models suited to Indian languages and contexts. It reflects the policy goal that India should not merely consume foreign AI models but build its own, reducing dependence on external providers. Example: Under the IndiaAI Mission's foundation models pillar, proposals have been invited from Indian startups and consortia to build homegrown large language models trained on Indian data and languages.
• Sarvam AI: Sarvam AI is a Bengaluru-based artificial intelligence startup building large language models focused on Indian languages. In April 2025 it was selected under the IndiaAI Mission to build India's first sovereign foundational AI model, trained from scratch on Indian datasets. Its Sarvam-1 model supports ten major Indian languages, and it released open-weight 30-billion and 105-billion parameter models in 2026. Example: Sarvam-1, a 2-billion-parameter model supporting Hindi, Tamil, Telugu and seven other Indian languages.
• BharatGen: BharatGen is India's sovereign multilingual artificial intelligence initiative, led by IIT Bombay, to build large language models trained on Indian languages, datasets and cultural contexts. It received 235 crore rupees from the Department of Science and Technology under the National Mission on Interdisciplinary Cyber-Physical Systems, plus 1,058 crore rupees from the Ministry of Electronics and IT under the IndiaAI Mission. In 2026 it unveiled Param-2, a 17-billion-parameter mixture-of-experts model covering 22 scheduled languages, and it operates through the BharatGen Technology Foundation, a Section-8 company registered in November 2025. Example: BharatGen's domain models include Ayur Param for Ayurvedic knowledge, Agri Param for agriculture and Legal Param for Indian law.
• AI Kosh: The national datasets platform under the IndiaAI Mission that hosts curated, anonymised non-personal datasets for training Indian AI models. It is one of the mission's pillars alongside shared compute, indigenous foundation models, skilling and startup financing. Example: it supports the development of models such as BharatGen, trained on Indian languages and priority sectors. Example: Hosts curated non-personal datasets used to train indigenous models such as BharatGen.
• IndiaAI portal: The IndiaAI portal is the single-window online hub of the IndiaAI Mission, bringing together India's artificial intelligence ecosystem in one place. It hosts information on the mission's programmes, AI datasets, models, research publications, startups and skilling initiatives, serving researchers, developers, entrepreneurs and policymakers. Example: A researcher can discover the AIKosh collection of curated, AI-ready public datasets through the IndiaAI portal.
• AI Centres of Excellence: Dedicated hubs for artificial intelligence research, training and startup support established across states to build domestic AI capability. They anchor the view of AI as public infrastructure rather than a private luxury, linking MeitY, academia and industry. Example: the Union Budget 2024-25 announced three Centres of Excellence for AI in healthcare, agriculture and sustainable cities with an outlay of Rs 255 crore. Example: Union Budget 2024-25 announced three AI CoEs (healthcare, agriculture, sustainable cities) with Rs 255 crore outlay.
• Trustworthy AI: Trustworthy AI refers to artificial intelligence systems that are lawful, ethical and robust, behaving reliably, transparently and accountably across their lifecycle. Frameworks from the OECD, the EU AI Act and UNESCO converge on principles such as human oversight, fairness and non-discrimination, privacy, safety, explainability and accountability. For exam purposes, trustworthy AI is the outcome that AI governance instruments are designed to produce. Example: Mandatory human review of AI-driven decisions in high-risk uses such as hiring, lending or medical diagnosis.
• Deepfakes: Deepfakes are synthetic audio, images or videos generated by AI models, typically deep learning systems, that convincingly show real people saying or doing things they never did. They are created with generative adversarial networks or newer diffusion and voice-cloning models, and their falling cost has made them a tool for fraud, political manipulation and non-consensual imagery. India has responded through the IT Rules, Election Commission advisories during elections, and debate over mandatory labelling or watermarking of AI-generated content. Example: Manipulated videos of political leaders circulated during the 2024 Lok Sabha elections, prompting the Election Commission to direct parties and platforms to act against such content.
• liar's dividend: The liar's dividend is the perverse benefit that the spread of deepfakes gives to wrongdoers: once convincing fake videos and audio flood the public sphere, even genuine evidence can be dismissed as fake. Coined by legal scholars Bobby Chesney and Danielle Citron, the term describes how synthetic media erodes shared truth, letting the guilty deny real recordings and undermining trust in news, elections and courts. It is a core argument for regulating deepfakes and for provenance and watermarking standards. Example: A politician caught on a genuine video of wrongdoing can claim the clip is an AI deepfake and escape accountability, collecting the liar's dividend created by public awareness that realistic fakes now exist.
• safe harbour: Under Section 79 of the Information Technology Act, 2000, a safe harbour is the legal protection that exempts intermediaries, such as social media platforms, search engines and internet service providers, from liability for third-party content they merely host or transmit. It is conditional: the intermediary must not initiate or modify the content, must observe due diligence, and must take down unlawful content after receiving a court order or government notice, as clarified by the Supreme Court in Shreya Singhal (2015). Loss of safe harbour exposes platforms to liability for what their users post. Example: A social media platform is not liable for a defamatory user post unless it fails to remove the post after a court or government order.
• Watermarking: Watermarking, in the context of artificial intelligence, is the technique of embedding invisible, machine readable markers into AI generated text, images, audio or video so that their synthetic origin can later be detected. It is a leading technical tool against deepfakes, misinformation and academic dishonesty, and several governments and companies are pushing for mandatory labelling of AI generated content. The challenge is making watermarks robust enough to survive cropping, paraphrasing and compression. Example: Google DeepMind's SynthID embeds imperceptible watermarks into AI generated images and text so that platforms can identify synthetic media at scale.
• provenance metadata: Provenance metadata is information attached to digital content that records its origin, creation process and editing history. It lets viewers and platforms verify whether an image, video or text was AI-generated, by whom, and whether it has been altered. Standards such as the C2PA specification embed this data cryptographically so it survives sharing and cannot be silently stripped. Example: A news photograph carrying C2PA credentials showing it was captured by a specific camera and not generated by AI, helping fact-checkers spot deepfakes.
• OECD AI Principles: The first intergovernmental standard on artificial intelligence, adopted by the OECD in May 2019 and later endorsed by the G20, setting five value-based principles: inclusive growth and well-being, human-centred values and fairness, transparency and explainability, robustness and safety, and accountability. They serve as a global reference point for national AI policies, including India's responsible-AI frameworks. The principles are non-binding but shape how governments design AI regulation. Example: India's national AI strategy documents reference OECD-style responsible-AI principles, including fairness, transparency and accountability, in guiding public AI deployment.
• AI Impact Summit 2026: The global AI summit hosted by India at Bharat Mandapam, New Delhi, from 16 to 20 February 2026, the first major global AI summit held in the Global South. Organised by MeitY, it shifted the global conversation from AI safety to AI for development under the theme of People, Planet and Progress and the guiding principle of Sarvajana Hitaya, Sarvajana Sukhaya (welfare for all, happiness for all). Example: the summit adopted the New Delhi Declaration, endorsed by around 86 to 88 countries and international organisations, committing to secure, trustworthy and inclusive AI. Example: Adopted the New Delhi Declaration, endorsed by around 86 to 88 countries and international organisations.
• New Delhi Declaration on AI Impact: The full title of the declaration adopted at India's AI Impact Summit in New Delhi in February 2026. It organises global AI cooperation around seven pillars, described as chakras: democratising AI resources, economic growth and social good, secure and trusted AI, AI for science, access for social empowerment, human capital development, and resilient, efficient and innovative AI systems. Example: The United States, China, the United Kingdom, France and Russia were among the endorsing countries.
• Sarvajan Hitaya, Sarvajana Sukhaya: This Sanskrit phrase, meaning welfare for all and happiness for all, was adopted as the guiding principle of the New Delhi Declaration on AI Impact endorsed at India's AI summit. It frames India's approach to artificial intelligence: AI should be developed and deployed for inclusive benefit rather than concentrated advantage. The phrase captures the policy goal of AI for all in the country's AI strategy.
• Global AI Impact Commons: The Global AI Impact Commons is a voluntary global platform launched at the India AI Impact Summit 2026 in New Delhi, documenting over 80 AI impact stories across more than 30 countries. It emerged as an official deliverable of the summit's Working Group on AI for Economic Growth and Social Good, functioning as a shared repository that lets countries discover, replicate and scale proven AI deployments for development. Example: The Commons collects impact stories such as AI-assisted tuberculosis detection and AI farmer advisories so that other countries can adapt them to their own contexts.
• FutureSkills Prime: FutureSkills Prime is India's national digital skilling platform jointly operated by the Ministry of Electronics and Information Technology and NASSCOM. It offers industry-aligned courses and certifications in emerging technologies such as artificial intelligence, big data, cloud computing, cybersecurity and blockchain, with a large share of learners from tier-2 and tier-3 cities. By mid-2026 it had registered over 34 lakh candidates, with more than 13 lakh certified. Example: A working professional can take a FutureSkills Prime course in artificial intelligence and earn an SSC-NASSCOM validated certification.
• Skill India Mission: Launched on 15 July 2015 (World Youth Skills Day), Skill India Mission is the umbrella framework that coordinates India's skilling ecosystem through the Ministry of Skill Development and Entrepreneurship. It works through programmes like the Pradhan Mantri Kaushal Vikas Yojana (PMKVY), which offers short-term skill training with certification and placement assistance, and the Pradhan Mantri Kaushal Kendras (PMKKs). The mission aimed to skill or upskill 40 crore people by 2022, and it is central to questions on employment, the demographic dividend and human capital. Example: PMKVY training centres offering certified courses in trades like electrician, tailoring and retail sales to school dropouts seeking formal-sector jobs.
• National Education Policy 2020: The National Education Policy 2020 is India's first education policy of the 21st century, approved by the Union Cabinet on 29 July 2020, replacing the 1986 policy. It introduces the 5+3+3+4 school structure, mother-tongue instruction till Class 5, multidisciplinary higher education with multiple entry-exit, and a target of 100% gross enrolment by 2030. It matters for UPSC because education reforms, GER targets, and social-sector policy are central GS-2 topics. Example: Approved by the Union Cabinet on 29 July 2020
• World Economic Forum's Future of Jobs Report 2025: The Future of Jobs Report 2025 is the World Economic Forum's survey-based flagship labour-market study, released in January 2025 ahead of the Davos annual meeting, drawing on more than 1,000 employers across 55 economies covering over 14 million workers. It projects that 170 million new jobs will be created by 2030 while 92 million are displaced, a net gain of 78 million, with technology, the green transition, demographics and geoeconomic fragmentation disrupting about 22% of current jobs. Technology skills in AI, big data and cybersecurity are the fastest-growing in demand, while clerical roles like cashiers and ticket clerks face the steepest decline, and the skills gap is employers' top barrier to business transformation. Example: Farm workers, delivery drivers and software developers are the roles adding the most new jobs by 2030, while postal clerks, bank tellers and data entry clerks decline fastest.
• Bias and discrimination: In the context of artificial intelligence, bias and discrimination refer to systematic errors in AI systems that produce unfairly skewed outcomes against particular groups. Such bias typically originates in unrepresentative training data, flawed labelling, or design choices that encode historical inequalities, and it is a central concern in AI ethics and governance. Regulators worldwide are responding with requirements for fairness audits, transparency and accountability in high-stakes AI applications such as hiring, lending and law enforcement. Example: An experimental AI recruitment tool developed by Amazon was found to systematically downgrade CVs containing the word 'women's', illustrating gender bias learned from historical hiring data.
• black box problem: The black box problem is the difficulty of interpreting why a complex artificial intelligence model produces a particular output. Deep neural networks may be highly accurate while their internal decision logic remains opaque, which raises concerns about accountability, bias and safety in high-stakes uses. Example: A deep learning model recommending a medical diagnosis without explaining its reasoning illustrates the black box problem.
• Privacy and surveillance: Privacy and surveillance refers to the tension between an individual's right to privacy, recognized as a fundamental right under Article 21 in Justice K.S. Puttaswamy v. Union of India (2017), and the growing capacity of states and companies to watch populations. Artificial intelligence intensifies the surveillance side by enabling tracking of movement, behavior, and identity at population scale, raising risks of profiling, data misuse, and chilling effects on free expression. UPSC-relevant debates center on balancing national security and public safety with proportionality, oversight, and data protection law. Example: Large-scale CCTV and facial-recognition networks that can track individuals' movement across a city in real time
• Digital Personal Data Protection Act, 2023: The Digital Personal Data Protection Act, 2023 is India's first comprehensive data protection law, which received Presidential assent in August 2023. It governs the processing of digital personal data, giving individuals (Data Principals) rights of access, correction and erasure, and imposing obligations on data fiduciaries, with penalties up to Rs 250 crore per contravention. Example: it creates the Data Protection Board of India to adjudicate breaches. Key for UPSC: privacy as a fundamental right after Puttaswamy (2017). Example: A user can demand that an app delete her personal data, and the app must comply or face Board proceedings.
• Project Vaani: Project Vaani is a large-scale speech-data collection initiative instituted in 2022 by the Indian Institute of Science, ARTPARK, and Google. Using a district-anchored approach across India's 773 districts, it records open-source natural speech in diverse Indian languages and dialects to train AI models for speech recognition, translation, and language understanding. Its data underpins models like SraVaani, the open-source multilingual speech-recognition model released by IISc in 2026 covering 65 Indian languages and dialects. Example: SraVaani, released in August 2026 and trained on Project Vaani data, converting speech to text in languages such as Garo, Tulu, and Kokborok
• Digital Public Infrastructure: Digital Public Infrastructure is the shared, open and interoperable digital backbone on which governments and private innovators build services, such as digital identity, payments and data-exchange systems. It is typically built as public goods with open standards and APIs rather than as closed proprietary platforms. For UPSC, it is central to governance, digital economy and India Stack based questions. Example: India's India Stack (Aadhaar for identity, UPI for payments, DigiLocker, ONDC) is the canonical example, showcased during India's G20 presidency.
• Key takeaways: Key takeaways is a quick-revision summary box placed at the end of each article, listing the most exam-relevant points in a short numbered list. It distils definitions, dates, figures, provisions and examples into a form suited for last-minute revision before Prelims and Mains. Together with the article's practice questions, it turns every topic into a self-contained study unit.
• Natural language processing (NLP): A branch of artificial intelligence concerned with enabling computers to understand, interpret and generate human language in text or speech. It combines linguistics with machine learning techniques such as tokenisation and transformer models, and underpins applications like machine translation, chatbots, sentiment analysis and voice assistants. Example: Machine translation tools and conversational AI assistants are built on natural language processing models trained on vast text corpora.
• Generative Adversarial Networks (GANs): Generative Adversarial Networks (GANs) are a class of machine learning models in which two neural networks, a generator and a discriminator, are trained against each other: the generator creates synthetic data while the discriminator tries to distinguish it from real data. This adversarial training produces highly realistic synthetic images, videos and audio, but it also underlies deepfake content and misinformation risks. Example: GANs power realistic synthetic face generation, which has raised deepfake concerns during elections and on social media.
• MuleHunter.AI: MuleHunter.AI is an AI and machine learning tool developed by the Reserve Bank Innovation Hub (RBIH) and launched by the RBI in December 2024 to detect mule accounts used in digital fraud and money laundering. It analyses transaction and account activity patterns across banks, replacing slower static rule based fraud detection systems. Example: The RBI reported that pilots with large banks were highly successful, with plans to move the model into production.
• collaborative robots (cobots): Collaborative robots, or cobots, are robots designed to work safely alongside humans in shared workspaces rather than in isolated cages. Built with force limits, sensors and rounded designs, they typically assist with repetitive or precision tasks while humans handle judgment-intensive work. Example: Cobots on automotive assembly lines assist workers with repetitive fastening and handling tasks.
• National Strategy for Artificial Intelligence (2018): The National Strategy for Artificial Intelligence, released by NITI Aayog in 2018 under the theme #AIforAll, is India's roadmap for adopting AI for inclusive growth. It identifies five focus sectors: healthcare, agriculture, education, smart cities and infrastructure, and smart mobility and transportation. The strategy aimed to position India as a leader in AI for social good while addressing barriers like data availability, research capacity and reskilling. Example: NITI Aayog proposed the AIRAWAT platform for AI computing infrastructure as part of the strategy's implementation.
• Synthetically generated information (SGI): Synthetically generated information is information that appears reasonably authentic but has been created or altered using a computer resource in ways that could pass for the real thing. Indian law uses this term for AI-generated media such as deepfakes, voice clones and fabricated images, which can mislead voters and the public and create what experts call the liar's dividend, where even genuine content is doubted. Example: A deepfake video showing a public figure saying something they never said is synthetically generated information.
• Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021: Notified on 25 February 2021 by the Ministry of Electronics and Information Technology under the Information Technology Act, 2000, the IT Rules, 2021 replaced the 2011 rules and for the first time brought digital news publishers and OTT platforms under regulation alongside social media intermediaries. They impose due diligence obligations on intermediaries, extra duties on significant social media intermediaries (chief compliance officer, grievance officer resident in India, monthly compliance reports), traceability of first originators, and a Code of Ethics with a three-tier grievance redressal mechanism for digital media including age-based content classification. Example: OTT platforms must classify content into age categories (U, U/A 7+, U/A 13+, U/A 16+, A) and display prominent content descriptors, while significant social media intermediaries must acknowledge and resolve user complaints within prescribed timelines.
• European Union's AI Act (2024): The European Union's AI Act (Regulation 2024/1689) is the world's first comprehensive binding law on artificial intelligence, entering into force on 1 August 2024. It regulates AI systems through a risk-based approach with four tiers: prohibited (unacceptable risk), high risk, limited risk and minimal risk, with stricter obligations on higher risk systems including conformity assessment, transparency and human oversight. It applies extraterritorially to AI systems whose outputs are used in the EU, influencing global AI governance. Example: Social scoring systems are banned as unacceptable risk, while AI used in hiring or biometric identification is classified high risk with strict compliance duties.
• UNESCO Recommendation on the Ethics of AI (2021): Adopted unanimously by all 193 UNESCO member states in November 2021, this is the first global standard-setting instrument on the ethics of artificial intelligence. It rests on four values, namely human rights and dignity, environmental flourishing, diversity and inclusion, and peaceful and just societies, and ten principles including transparency, fairness, accountability, privacy and human oversight, implemented through eleven policy action areas such as ethical impact assessment. It is non-binding but near-universal in membership. Example: UNESCO's Readiness Assessment Methodology, a tool that helps countries evaluate how prepared they are to implement the recommendation.
• Global Partnership on AI (GPAI): The Global Partnership on AI (GPAI) is a multi-stakeholder initiative launched in June 2020 to guide the responsible development and use of artificial intelligence grounded in human rights and democratic values. India is a founding member, and its secretariat is hosted at the OECD. Example: India chaired the GPAI Council in 2023, placing responsible AI on the agenda during its G20 presidency year.
• FutureSkills Data and AI labs: The FutureSkills Data and AI labs are a network of AI training laboratories being set up by the IndiaAI Mission's FutureSkills pillar with NIELIT in Tier 2 and Tier 3 cities to offer foundational courses in data and artificial intelligence. Part of the IndiaAI Mission (approved in March 2024 with an outlay of over Rs 10,371 crore), the pillar also awards IndiaAI fellowships to undergraduate, postgraduate and PhD students to build deep AI talent beyond the major metros. Example: IndiaAI has set up Data Labs at NIELIT's Delhi centre and ICIT Nagaland, with 27 more labs planned at NIELIT centres across Tier 2 and Tier 3 cities.
• New Delhi Declaration (2026): The outcome document adopted at the AI Impact Summit held in New Delhi in February 2026, the first global AI summit hosted in the Global South. Endorsed by 88 countries and international organisations, it sets a cooperative framework for AI development guided by the principle of Sarvajan Hitaya, Sarvajan Sukhaya (welfare for all, happiness for all), so that AI's benefits are shared equitably across humanity. Example: The summit drew over five lakh visitors and secured infrastructure investment commitments exceeding 250 billion US dollars, according to the IT Ministry.
• GovAI: GovAI, or governance with AI, is the use of artificial intelligence inside the state: for data-driven policy making, automation of public processes, personalised citizen services, predictive analytics and scheme monitoring. Indian examples include GSTN's AI fraud detection, MyGov's scheme chatbots and IMD's AI cyclone prediction. It matters because India's Digital Public Infrastructure gives AI ready rails to run on.
• AIRAWAT: AIRAWAT is India's AI-specific supercomputing system installed at C-DAC Pune under the National Supercomputing Mission. Unlike general-purpose supercomputers, it is built for large artificial intelligence workloads. It anchors the compute side of India's AI ambitions.
• S.A.R.A.H.: S.A.R.A.H. is a generative AI prototype for digital health promotion launched by the World Health Organization in April 2024. It marked a milestone in deploying generative AI for public health. It matters as the global reference point for AI health assistants.
• iOncology.ai: iOncology.ai is an indigenous AI platform developed jointly by AIIMS New Delhi and C-DAC for early detection, diagnosis and treatment planning for breast and ovarian cancers. Showcased in May 2026 and integrated with the Ayushman Bharat Digital Mission, it assists clinicians with AI-driven insights from Indian medical datasets.
• ICTAI: The International Centre for Transformational AI (ICTAI) was launched by the Maharashtra government with NITI Aayog to develop AI-driven rural healthcare solutions. It represents state-level institution building for applied AI. Its focus on rural health targets the access gap directly.
• ICMR Ethical Guidelines for AI in Healthcare: The Indian Council of Medical Research released Ethical Guidelines for AI in Biomedical Research and Healthcare in March 2023. They set India's home-grown standards for fairness, transparency and accountability in medical AI. They are the domestic counterpart to WHO's global principles.
• Shallow fakes (cheap fakes): Shallow fakes, or cheap fakes, are pictures, videos and voice clips altered without artificial intelligence, using simple editing tools such as Photoshop. They are easier to create and easier to detect than deepfakes. They still cause harm through mis-captioning and mis-contextualising genuine footage.
• MANAV vision: The MANAV vision is India's human-centric framework for artificial intelligence, unveiled by the Prime Minister at the India AI Impact Summit in February 2026. MANAV means human, and the acronym stands for Moral systems, Accountable governance, National sovereignty, Accessible and inclusive AI, and Valid and legitimate systems. It is the one-line summary of India's AI governance philosophy.
• Pax Silica: Pax Silica is a US-led international collaboration launched in December 2025 for secure, resilient AI ecosystems across the silicon stack: critical minerals, semiconductor fabrication, AI systems and deployment infrastructure. India joined at the AI Impact Summit in February 2026. It links AI ambition to supply-chain security.
• REAIM: Responsible AI in the Military Domain (REAIM) is the international process setting norms for military uses of artificial intelligence: international standards, transparency and accountability, non-discrimination, data protection and responsible R and D. It is the governance reference for AI in warfare debates.
• Recurrent neural networks (RNNs): Recurrent neural networks are neural networks tailored for sequence modelling: they capture temporal relationships between sequence elements and can predict what comes next. They suit language and time-series data. They were the workhorse of language AI before transformer architectures took over.
• Shallow neural networks: Shallow neural networks contain only one layer of neurons, making them simpler to train but less adept at recognising complex data patterns than deep networks. They illustrate the depth trade-off in machine learning. They matter as the baseline against which deep learning's gains are measured.
• PMFBY: The Pradhan Mantri Fasal Bima Yojana (PMFBY) is India's crop insurance scheme, which has piloted artificial intelligence for optimising crop-cutting experiments used to assess yields. It is a concrete GovAI application in agriculture. It shows AI moving from advisories into the administrative core of welfare schemes.
• GSTN: The Goods and Services Tax Network (GSTN) is the IT backbone of India's GST system, which uses artificial intelligence to detect tax fraud and improve compliance. It is a flagship GovAI example of automation in public finance. It shows AI applied to revenue protection at national scale.
• MyGov: MyGov is the Government of India's citizen-engagement platform, which uses AI chatbots to offer customised scheme suggestions to users. It exemplifies personalised citizen services under GovAI. It matters as the conversational front door of digital governance.
• UMANG: UMANG (Unified Mobile Application for New-age Governance) is India's single mobile platform for accessing central and state government services. AI features extend its reach in scheme delivery and service access. It is cited alongside Aarogya Setu as digital infrastructure for welfare delivery.

Prelims practice