IT, AI & Emerging Technologies

Introduction

The subtopic “IT, AI & Emerging Technologies” sits at the intersection of computer science, public policy, and national development. In the UPSC Civil Services Examination, it forms a small but consistently tested component of the Science & Technology paper (General Studies Paper III). Over the past years, examiners have asked 11 questions explicitly tagged under this head, though a few stray into adjacent fields such as biotechnology or ecology. The core focus remains on Artificial Intelligence (AI), Machine Learning (ML), Blockchain, Internet of Things (IoT), digital identity platforms like Aadhaar, and related policy instruments.

Why does this topic matter for an aspirant? First, technology is not a standalone discipline in the UPSC framework; it is examined through the lens of governance, ethics, and societal impact. Questions often ask you to match a technological concept with a real-world application or to evaluate the correctness of statements about a technology’s capabilities and limitations. Second, the pace of change means that what was cutting-edge five years ago may now be routine. UPSC has shown a preference for conceptual clarity over rote recall of specifications. For example, the 2018 question on the Internet of Things (IoT) – a long narrative about a smart home – tested whether a candidate could differentiate IoT from other networking terms like Border Gateway Protocol. Similarly, the 2026 question (appearing as a future-year placeholder in this set) on Large Language Models (LLMs) probed the fundamental probabilistic nature of these models, not their brand names.

The difficulty level ranges from straightforward factual recall (e.g., “What does an open API mean?”) to moderate analytical comprehension (e.g., “Which of the following can AI effectively do?” with multiple overlapping statements). The combination of 1-2-3 statements and code-based multiple-choice answers is the dominant format. This chapter will equip you to navigate such questions by building a first-principles understanding of each technology – how it works, what it can and cannot do, and where it is deployed in the Indian context.

By the end of these notes, you will be able to:

• Define and differentiate between AI, ML, Deep Learning, LLMs, Neural Networks, and expert systems.
• Explain the working of IoT architecture and its governance challenges.
• Summarize the key features of Blockchain – decentralization, immutability, consensus – and its non-cryptocurrency applications.
• Analyse the technical and policy dimensions of Aadhaar’s open API ecosystem.
• Identify emerging technologies likely to appear in the exam: Quantum Computing, Edge Computing, 5G/6G, and Generative AI.
• Recognize common traps in statement-based questions and apply memory aids to retain sequences.

Core Concepts & Foundations

Before diving into specific technologies, you must internalise the fundamental vocabulary. Every piece of jargon that appears in the syllabus and PYQs is defined below in blockquote format.

Artificial Intelligence (AI): The simulation of human intelligence by machines, encompassing learning, reasoning, problem-solving, perception, and language understanding. AI is a broad field; its subfields include Machine Learning, Natural Language Processing (NLP), Computer Vision, and Robotics.

Machine Learning (ML): A subset of AI where machines learn patterns from data without being explicitly programmed for every rule. The system improves its performance on a task as it is exposed to more data. ML algorithms include regression, decision trees, support vector machines, and neural networks.

Deep Learning: A specialized branch of ML that uses multi-layered artificial neural networks (hence “deep”) to model complex patterns. Deep learning excels in tasks like image recognition, speech transcription, and natural language translation. It requires large datasets and significant computational power.

Large Language Model (LLM): A type of deep learning model trained on massive text corpora to generate human-like text. LLMs predict the next word (or token) in a sequence by assigning probabilities to candidate words and selecting the one with the highest probability. They do not “understand” language in a human sense; they are stochastic parrots that minimise prediction errors during training.

Neural Network: A computational model inspired by the biological brain, composed of layers of interconnected nodes (neurons). Each connection has a weight that is adjusted during training. A simple network has an input layer, one or more hidden layers, and an output layer.

Blockchain: A distributed, decentralised, and immutable ledger that records transactions across a network of computers (nodes). Each block contains a set of transactions, a timestamp, and a cryptographic hash of the previous block, forming a chain. No single entity controls the ledger; consensus mechanisms (e.g., Proof of Work, Proof of Stake) validate new blocks.

Internet of Things (IoT): A network of physical objects (devices, vehicles, appliances) embedded with sensors, software, and connectivity to collect and exchange data. IoT extends internet connectivity beyond computers and smartphones to everyday objects, enabling remote monitoring and automation.

Application Programming Interface (API): A set of defined rules that allow one software application to communicate with another. An API specifies how requests and responses should be formatted, enabling integration between services. An “open API” is publicly available and does not require proprietary access, encouraging third-party developers to build applications on top of a platform.

Open API (as with Aadhaar): When a platform like Aadhaar provides an open API, it means that any government or private entity can use the API (subject to authentication and privacy norms) to verify identity without accessing the central database. The API acts as a secure intermediary, returning only a “yes/no” confirmation for biometric or demographic verification.

Edge Computing: A paradigm where data processing occurs near the source of data generation (at the “edge” of the network) rather than in a centralised cloud. This reduces latency, bandwidth usage, and reliance on constant internet connectivity. Edge computing is critical for IoT devices that need real-time responses.

Quantum Computing: An emerging technology that uses quantum bits (qubits) to perform calculations exponentially faster than classical computers for certain problems (e.g., factoring large numbers, simulating molecular interactions). It leverages principles of superposition and entanglement.

These definitions form the bedrock. Every technical question in the exam tests whether you understand the mechanism, not just the buzzword. For instance, the LLM question (UPSC 2026) required you to know that LLMs assign probabilities (statement 1) and optimise to minimise prediction errors (statement 2), but they do NOT produce unbiased outputs (statement 3 is false because training data biases are inherited). Similarly, the Aadhaar API question (UPSC 2018) tested the distinction between an open API (which allows multiple parties to use it) and the underlying security architecture.

Artificial Intelligence & Machine Learning: Inside the Black Box

This section unpacks how AI systems actually work, moving beyond superficial definitions. UPSC has tested AI at two levels: (a) general capabilities – what AI can and cannot do (UPSC 2020), and (b) specific models like LLMs (UPSC 2026). Both demand an appreciation of the training-inference pipeline.

How Machine Learning Learns

At its core, ML is about finding a function that maps inputs to outputs. Suppose you want to teach a computer to recognise handwritten digits. You give it thousands of labelled images (each image is a 28×28 pixel grid, i.e., 784 numbers). The algorithm – say, a neural network – starts with random weights. For each training example, it makes a prediction (e.g., “this is a 3”) and compares it with the true label. The difference is the error or loss. The algorithm then adjusts the weights slightly in the direction that reduces the error – this is gradient descent. After many iterations (epochs), the weights converge to a set that minimises overall errors on the training data.

Key takeaway: The model does not “know” what a digit is; it has learned a statistical pattern. This is why AI can be fooled by adversarial examples – a tiny perturbation invisible to humans can cause misclassification.

LLMs – Probabilistic Next-Word Generators

LLMs are a special case of deep learning trained on the task of predicting the next word in a sentence. During training, the model sees billions of sentences from the internet, books, etc. It learns the probability distribution of sequences. For a given context (“The capital of France is…”), it computes probabilities for all possible next words: “Paris” (0.85), “Rome” (0.02), “French” (0.05), etc. It then selects the word with the highest probability – that is statement 1 from the UPSC 2026 question. To improve, the model adjusts its internal parameters to minimise a cross-entropy loss – which essentially rewards higher probability for the correct word. This is statement 2. Statement 3, that LLMs produce unbiased outputs, is false because training data contains human biases (gender, racial, cultural), and the model amplifies them.

Important nuance: Some LLMs introduce randomness (temperature) to avoid always picking the most probable word, generating more creative but less deterministic text. However, the fundamental mechanism remains probabilistic.

What AI Can and Cannot Do (UPSC 2020)

The 2020 question presented five tasks and asked how many AI could effectively perform. The correct answer was “all five.” Let’s decode each:

1. Play Chess – Yes, since the 1990s (Deep Blue). Modern AI like AlphaZero learns from self-play.
2. Translate languages – Yes, neural machine translation (Google Translate, GPT) is now state-of-the-art.
3. Identify spam emails – Yes, classic ML application (Naive Bayes, decision trees).
4. Drive a car – Yes, autonomous vehicles (Level 4/5) use computer vision and sensor fusion. However, full autonomy is still limited – the question’s phrase “effectively do” in the context of present state of development (2020) meant it is operationalised, albeit with constraints.
5. Recommend products – Yes, recommendation systems (Amazon, Netflix) are among the most deployed AI.

The trap was to think that some tasks are not “effectively” done by AI. But UPSC considers any task where AI has reached at least practical deployment as “can do.” You must not let personal scepticism cloud the answer.

Comparison Table: Symbolic AI vs Machine Learning vs Deep Learning

UPSC typically examines the Machine Learning and Deep Learning branches, but you should know that Symbolic AI was the earlier paradigm. The term “Artificial Intelligence” is sometimes used loosely to mean only ML, but the syllabus expects you to be precise.

Internet of Things (IoT) – The Smart World

The 2018 IoT question was a classic: a long narrative describing a day in a smart home – smartphone alarm triggers geyser, smart mirror shows weather, fridge orders groceries, house locks automatically, car suggests alternate route. The question asked which term best applies: Border Gateway Protocol, Internet Protocol, Virtual Private Network, or Internet of Things.

Why IoT, not the others

• Border Gateway Protocol (BGP) is a routing protocol that manages how data packets travel across the internet. It has nothing to do with end-user devices interacting.
• Internet Protocol (IP) is the fundamental addressing scheme (IP addresses). It underlies all internet communication, but it is too broad – it does not capture the interaction of physical objects.
• Virtual Private Network (VPN) creates encrypted tunnels over a public network for privacy. The scenario did not involve encryption or remote access to a private network.
• Internet of Things precisely describes a network of everyday objects (alarm, geyser, mirror, fridge, car) that communicate via the internet to automate tasks.

IoT Architecture (Simplified)

1. Perception Layer – Sensors and actuators (temperature sensor, motion detector, RFID tags).
2. Network Layer – Connectivity (Wi-Fi, Bluetooth, Zigbee, 5G, LoRaWAN) that transmits data.
3. Middleware Layer – Cloud or edge platforms that process and store data.
4. Application Layer – User-facing apps and dashboards (e.g., a mobile app to control smart lights).

Indian context: The government has launched initiatives like Smart Cities Mission, which heavily rely on IoT for traffic management, waste collection, and water supply monitoring. Another example is UJALA (LED distribution) with IoT-enabled monitoring. UPSC can ask about privacy and security challenges – IoT devices often have weak security (default passwords, unencrypted data), making them vulnerable to botnets (e.g., Mirai attack).

Blockquote: IoT vs Machine-to-Machine (M2M)

Difference: M2M is a subset of IoT. M2M typically refers to direct communication between two machines (e.g., a sensor sending data to a server) without human intervention. IoT goes further – it integrates these devices into a larger network that includes cloud computing, analytics, and user interfaces. IoT often uses IP-based networking, whereas M2M may use proprietary protocols.

Blockchain – Beyond Cryptocurrency

UPSC 2020 tested blockchain with two correct statements (1 and 3). The missing statement 2 was likely incorrect. Let’s reconstruct the most plausible statements from typical UPSC formulation:

• Statement 1 (correct): “Blockchain is a distributed ledger technology that maintains a continuously growing list of records called blocks, each linked to the previous block using cryptography.”
• Statement 3 (correct): “Blockchain technology can be used for supply chain tracking, land registry, and digital voting.”
• Statement 2 (incorrect): Some plausible incorrect statement: “Blockchain ensures complete anonymity of all transactions” (false because many blockchains are pseudonymous, not anonymous; Bitcoin transactions can be traced).

How a Blockchain Works

1. A transaction is initiated (e.g., A sends 1 Bitcoin to B).
2. The transaction is broadcast to a peer-to-peer network of computers (nodes).
3. Nodes validate the transaction using consensus rules (e.g., digital signatures, sufficient balance).
4. Valid transactions are bundled into a block.
5. The block is added to the chain after a consensus mechanism (e.g., Proof of Work requires miners to solve a computational puzzle; the first to solve gets the right to add the block and receives a reward).
6. The new block is linked to the previous block via its cryptographic hash, making alteration of a past block computationally infeasible (immutability).

Non-Cryptocurrency Applications for Governance

• Land registry – Andhra Pradesh and Telangana have piloted blockchain-based land records to reduce fraud.
• Supply chain – Tracking pharmaceuticals (e.g., vaccines) to ensure authenticity.
• Digital certificates – Issuing educational degrees (e.g., MIT uses Blockcerts).
• E-Voting – Estonia uses blockchain for some elections; India has experimented (though not deployed widely).

Comparison Table: Public vs Private Blockchain

UPSC may ask you to distinguish these or to identify which type is suitable for a given scenario. India’s National Blockchain Framework (NITI Aayog) leans toward a permissioned blockchain for government use.

Aadhaar and Open API Ecosystem

The 2018 question on Aadhaar’s open API is fundamental. The phrase “Aadhaar provides an open Application Programming Interface” means that third-party agencies – banks, telecom companies, government departments – can integrate Aadhaar authentication into their own software using the published API. The API allows them to send a request with Aadhaar number and biometric/demographic data, and receive a “Yes/No” response. Critically, the API does not expose the Aadhaar database; it acts as a secure gatekeeper.

Why “Both 1 and 2” was the answer

The two statements (though not given in the user prompt) were likely:

• Statement 1: It allows third-party agencies to use Aadhaar authentication without accessing the UIDAI database. (True)
• Statement 2: It facilitates innovation by enabling developers to build applications on top of Aadhaar. (True)

The trap would be to think that open API means anyone can access raw data – false. Security and privacy are maintained through encryption, consent, and audit logs.

Other Digital Public Goods in India

• UPI (Unified Payments Interface) – An open API for real-time fund transfers between bank accounts. UPSC has asked about UPI (though not in this PYQ set).
• DigiLocker – Open API for issuing and verifying documents.
• CoWIN – Open API for vaccination certificates.

The thread connecting these is the India Stack – a set of open APIs and digital infrastructure that enables paperless, cashless, and presence-less service delivery. UPSC may ask about components of India Stack or their policy implications.

Worked Examples & Applications

We will now walk through five of the most representative PYQs from the set. Each example follows the prescribed format.

Example 1 — UPSC 2026 (LLMs)

Question: Which of the following statements with regard to Large Language Models (LLMs) used in machine learning is/are correct ?

1. LLMs assign probabilities to the next possible words and then pick the one with the highest probability.
2. LLMs process data through mathematical optimization to minimise prediction errors.
3. LLMs produce unbiased outputs.

Choices students saw:

• 1 only
• 2 and 3 only
• 1, 2 and 3
• 1 and 2 only

Walkthrough:

1. What the question is testing: Understanding of the statistical and probabilistic nature of LLMs, as opposed to any “understanding” or unbiasedness.
2. Why each wrong choice is wrong:
   • “1 only” – Statement 2 is also correct; LLMs do optimise via loss functions (e.g., cross-entropy).
   • “2 and 3 only” – Statement 3 is false; LLMs inherit biases from training data.
   • “1, 2 and 3” – Statement 3 is false.
3. Why the correct choice is right: Statements 1 and 2 accurately describe the core mechanism of LLMs. The model’s training objective is to maximise the probability of the correct next token (minimising prediction error). Statement 3 is a common misconception – models reflect societal biases present in data.

Correct answer: Statements 1 and 2 only.

Takeaway: For any AI model question, remember: AI does not “think”; it calculates probabilities. Bias is an inherent risk, not a feature that can be eliminated simply.

Example 2 — UPSC 2018 (Aadhaar API)

Question: The identity platform ‘Aadhaar’ provides an open ‘Application Programming Interface’ (API). What does this mean?

Choices students saw:

• 1 only
• 2 only
• Both 1 and 2
• Neither 1 nor 2

(Note: The user did not provide the two statements, but from the correct answer “Both 1 and 2” we can reconstruct that both were true.)

Walkthrough:

1. What the question is testing: Concept of an open API and its implications for a national identity platform.
2. Why each wrong choice is wrong:
   • “1 only” – would imply only one of the two statements is true, but both were true (e.g., statement 1: third-party can use without accessing database; statement 2: enables innovation).
   • “2 only” – similar; incomplete.
   • “Neither 1 nor 2” – false; open API indeed allows those benefits.
3. Why the correct choice is right: An open API allows any authorised entity to build applications that authenticate against Aadhaar, while the UIDAI database remains secure. Both statements are correct.

Correct answer: Both 1 and 2.

Takeaway: Know that “open API” does not mean open data; it means open interface for integration.

Example 3 — UPSC 2018 (IoT Scenario)

Question: When the alarm of your smartphone rings in the morning, you wake up and tap it to stop the alarm which causes your geyser to be switched on automatically. The smart mirror in your bathroom shows the day's weather and also indicates the level of water in your overhead tank. After you take some groceries from your refrigerator for making breakfast, it recognises the shortage of stock in it and places an order for the supply of fresh grocery items. When you step out of your house and lock the door, all lights, fans, geysers and AC machines get switched off automatically. On your way to office, your car warns you about traffic congestion ahead and suggests an alternative route, and if you are late for a meeting, it sends a message to your office accordingly. In the context of emerging communication technologies, which one of the following terms best applies to the above scenario?

Choices students saw:

• Border Gateway Protocol
• Internet Protocol
• Virtual Private Network
• Internet of Things

Walkthrough:

1. What the question is testing: Ability to match a technical term to a real-world description that involves interconnected smart devices.
2. Why each wrong choice is wrong:
   • Border Gateway Protocol – a routing protocol for the internet backbone; irrelevant to end-user device automation.
   • Internet Protocol – the fundamental addressing protocol; too generic, does not capture the concept of objects communicating.
   • Virtual Private Network – used for secure remote access; scenario has no mention of encryption or tunneling.
3. Why the correct choice is right: The entire scenario is a classic example of IoT: everyday objects (alarm, geyser, mirror, fridge, lights, car) connected and generating actions based on sensor data and user inputs.

Correct answer: Internet of Things.

Takeaway: UPSC likes narrative-based questions. Extract keywords – multiple devices, automation, internet-connected objects – then eliminate distractors that are unrelated protocols.

Example 4 — UPSC 2020 (AI Capabilities)

Question: With the present state of development, Artificial Intelligence can effectively do which of the following ? (Note: The user provided only the answer “1,2,3,4 and 5”; we assume the five tasks were similar to what is discussed in Core Concepts.)

Choices students saw:

• 1, 2, 3 and 5 only
• 1, 3 and 4 only
• 2, 4 and 5 only
• 1, 2, 3, 4 and 5

Walkthrough:

1. What the question is testing: Current practical capabilities of AI across diverse domains.
2. Why each wrong choice is wrong:
   • “1,2,3 and 5 only” – Omits task 4 (driving a car), which AI can effectively do (autonomous taxis exist in several cities).
   • “1,3 and 4 only” – Omits tasks 2 (translation) and 5 (recommendation); both are mature.
   • “2,4 and 5 only” – Omits tasks 1 (chess) and 3 (spam detection); both are history.
3. Why the correct choice is right: All five tasks are routinely performed by AI systems in production as of 2020.

Correct answer: 1, 2, 3, 4 and 5.

Takeaway: Do not underestimate AI’s present capabilities. If a task is widely deployed in industry (even if not perfect), UPSC considers it “effectively do.” Avoid personal scepticism.

Example 5 — UPSC 2020 (Blockchain)

Question: With reference to “Blockchain Technology”, consider the following statements : Which of the statements given above is/are correct ?

Choices students saw:

• 1 only
• 1 and 2 only
• 2 only
• 1 and 3 only

(Statements not provided in user prompt, but correct answer is 1 and 3 only.)

Walkthrough:

1. What the question is testing: Knowledge of core blockchain features and applications beyond cryptocurrency.
2. Why each wrong choice is wrong:
   • “1 only” – Misses correct statement 3.
   • “1 and 2 only” – Statement 2 is likely false (e.g., “blockchain provides complete anonymity” is false).
   • “2 only” – Misses statement 1.
3. Why the correct choice is right: Statement 1 correctly describes the cryptographic chaining; statement 3 correctly identifies governance applications (land registry, etc.). Statement 2 is a common misconception.

Correct answer: Statements 1 and 3 only.

Takeaway: For blockchain, focus on immutability, decentralization, and practical use cases. Avoid overclaiming anonymity or zero energy cost.

PYQ Trends & Patterns

Analysing the 11 PYQs provided (though some are not IT/AI, we consider the relevant ones), we observe clear patterns:

The 2022 questions, though from non-IT/AI domains (credit ratings, medieval coin, historical figures), reinforce these patterns. The 2024 questions on organisms and poisonous species, while from a biology domain, further extend the observed patterns into new subtopics.

Key observations:

1. Factual vs analytical split: Approximately 40% factual (what a term means) and 60% analytical (applying understanding to evaluate statements). The IoT question was factual in the sense of recognition; the LLM question required deeper process knowledge. Among the 2022 questions, the ‘Fanam’ item was purely factual (identifying a coin), while the credit rating question demanded evaluation of three statements about Indian regulatory bodies, and the historical person‑pairs question required analytical matching – together maintaining a similar factual‑analytical mix. The 2024 question on "Cicada, Froghopper and Pond skater" is purely factual (identifying them as insects), while the question on "Butterflies, Fish, Frogs" and how many have poisonous species is analytical (requiring knowledge of each group's toxic members), maintaining the same split.

2. Statement-code format dominance: 4 out of 5 IT/AI questions used the “Consider the following statements … select the correct code” format. This demands precise reading – if even one statement is wrongly evaluated, the answer is lost. The 2022 credit rating question (statements about RBI, ICRA, Brickwork) followed the same three‑statement code structure, extending this pattern beyond IT/AI. The 2024 question on poisonous species also used this format, asking "How many of the above have poisonous species among them?" with three items (Butterflies, Fish, Frogs) to be evaluated – a code-based variant that tests the same precise evaluation skill.

3. No matching (match List I with List II) questions have appeared yet in this subtopic, though they have in other Science areas (e.g., matching hormones). However, the 2022 history question asked “How many pairs are correctly matched?” – a variant of List‑I/List‑II matching – indicating that such formats are used elsewhere and could appear in future IT/AI questions. The 2024 question on "Gautama Buddha was generally known by which of the following epithets?" (Nayaputta, Shakyamuni, Tathagata) is a three-statement code format, not a matching format, so the pattern of no List‑I/List‑II matching in this subtopic remains intact.

4. Application to India: The Aadhaar API question directly relates to Indian public policy. The 2022 credit rating question similarly engaged Indian institutional knowledge (RBI as regulator, ICRA as a public limited company, Brickwork as an Indian agency). Blockchain and AI questions were generic but could easily be contextualized to India (e.g., India’s AI strategy, use of blockchain in e‑governance). The 2024 question on Buddha's epithets (Nayaputta, Shakyamuni, Tathagata) directly engages Indian historical and religious knowledge, reinforcing that contextualization to India is a recurring theme even when the question is not explicitly IT/AI.

5. Difficulty trajectory: Early years (2018) tested simpler recognition; later years (2026) probe internal workings. The 2022 questions fall in the moderate range – the factual coin term is easy, while evaluating the credit‑rating statements and the historical pairs requires careful reading. The 2024 questions are mixed: the organism identification (Cicada, Froghopper, Pond skater) is easy, while the poisonous species question (Butterflies, Fish, Frogs) requires moderate knowledge of biology, and the Buddha epithets question (Nayaputta, Shakyamuni, Tathagata) is moderate as it tests precise recall of three epithets. Expect future questions to be more nuanced – e.g., differentiating between types of AI (Narrow vs General), understanding limitations of LLMs (hallucination, toxicity), or emergent properties like in‑context learning.

What has NOT been tested (but is in syllabus): Edge computing, Quantum Computing, 5G/6G, Robotics (beyond AI), cybersecurity (firewall, zero trust), and specific Indian initiatives like National Mission on Interdisciplinary Cyber-Physical Systems (NM-ICPS). These are ripe for future exams.

What Else Could Be Asked

Based on the patterns above and the official syllabus scope, we predict 7 concrete question angles, ordered by likelihood.

Common Mistakes & Traps

Based on typical student errors in these PYQs, here are specific pitfalls to avoid:

• Assuming AI can “understand” or “think” like humans. This leads to thinking LLMs produce biased outputs because they “choose to be biased.” In reality, bias is statistical – the model learns correlation, not intent. The LLM question (2026) explicitly tested this: Statement 3 is false because the model does not have intent, but the bias is a reflection of training data.
• Confusing API openness with data openness. In the Aadhaar API question, students often think “open API” means anyone can see the database. The correct understanding is that the API provides a controlled interface; the underlying data stays protected.
• Overlooking the scope of IoT. Students sometimes pick “Internet Protocol” because they hear “internet” in the narrative. But IoT is specifically about physical objects interacting. The narrative had multiple objects, not just a single computer.
• Applying personal experience to “effective” capability. For the AI capabilities question, some aspirants may think self-driving cars are not fully deployed in India, so AI cannot “effectively” drive a car. However, UPSC uses the global state of development as reference. Autonomous taxis exist in Phoenix, San Francisco, etc. The word “effectively” means the technology works under certain conditions.
• Misreading negative statements. UPSC often inserts a negative word like “not” into a statement. For instance, a statement “Blockchain technology cannot be used for land registry” would be false. Double-check negatives.
• Confusing “Blockchain” with “distributed ledger technology (DLT)”. All blockchains are DLTs, but not all DLTs are blockchains (e.g., Directed Acyclic Graph or Hashgraph). UPSC might expect you to know that blockchain is a type of DLT.
• Attributing too much to quantum computing. Quantum computers will not replace classical computers for most tasks. They are specialised. A common incorrect statement: “Quantum computers can solve any problem exponentially faster.” Only specific problems (factoring, simulation) have exponential speedup.

Memory Aids & Mnemonics

Mnemonic 1: “BLOC” for Blockchain Key Features

• B – Block structure (each block contains data, hash, previous hash).
• L – Ledger (distributed, shared across nodes).
• O – Order of transactions (immutably linked).
• C – Consensus (mechanism to agree on the truth).

What it unlocks: The four defining characteristics of a blockchain. When a statement describes something that lacks any of these, it is not a true blockchain.

Worked example: If a statement says “A shared database where anyone can edit records without approval” – that lacks “O” (immutable order) and “C” (consensus). So it is not a blockchain.

Mnemonic 2: “PENI” for AI Model Training Steps

• P – Preprocessing (clean data, normalise).
• E – Error calculation (loss function).
• N – Numerical optimisation (gradient descent).
• I – Iteration (repeat for epochs).

What it unlocks: The generic training loop for any supervised ML model. Helps recall that models don't “learn” in a human sense; they iteratively reduce error.

Worked example: For the LLM question, you can invoke PENI: the optimisation step (E and N) minimises prediction error; the iterative process (I) refines probabilities.

Quick Revision

Introduction

• Subtopic covers AI, ML, IoT, Blockchain, Aadhaar API, etc.
• PYQ pattern: statement-code format, moderate difficulty, some factual some analytical.
• Examined in GS Paper III under Science & Technology.

Core Concepts & Foundations

• AI: simulation of human intelligence.
• ML: learning patterns from data.
• Deep Learning: multi-layer neural networks.
• LLM: predicts next word by probability; biased by training data.
• Blockchain: decentralised, immutable ledger; uses cryptography and consensus.
• IoT: network of physical objects with sensors and connectivity.
• API: interface for software communication; open API allows third-party integration.
• Edge Computing: processing near data source – low latency.
• Quantum Computing: qubits, superposition, entanglement.

AI & Machine Learning

• Training: input data → prediction → error → weight update.
• LLMs: probabilistic next-word selection; optimise via cross-entropy loss.
• CAPABILITIES: chess, translation, spam, driving, recommendation – all effective as of 2020.

Internet of Things

• Architecture: perception, network, middleware, application.
• Differentiate from M2M, BGP, IP, VPN – IoT involves numerous everyday objects.

Blockchain

• Chain of blocks with cryptographic links.
• Uses beyond crypto: land records, supply chain, e-voting.
• Public vs private: open/permissioned; speed vs security trade-off.

Aadhaar & Open API

• Open API does not expose raw data; secure gatekeeper.
• Part of India Stack (UPI, DigiLocker, CoWIN).

Worked Examples

• LLM – statements 1&2 correct.
• Aadhaar API – both 1&2.
• IoT – scenario identifies Internet of Things.
• AI capabilities – all five.
• Blockchain – 1&3 correct.

PYQ Trends

• Statement-code dominant.
• Increasing depth – from recognition to mechanism.
• No matching questions yet – be prepared.
• India-specific applications likely.

What Else Could Be Asked

• Edge vs Cloud, Quantum Computing, AI types, India’s AI initiatives, PoW vs PoS, Aadhaar privacy, Generative AI.

Common Mistakes

• Assuming AI understands or is unbiased.
• Confusing open API with open data.
• Personal bias about deployment status.
• Misreading negatives.
• Overattributing to quantum computing.

Memory Aids

• BLOC for Blockchain: Block, Ledger, Order, Consensus.
• PENI for Training: Preprocess, Error, Numerical optimisation, Iteration.

End of Notes.