AI governance

Explanation

The IMF AI Preparedness Index is an established measure that assesses how ready a country is to adopt and benefit from AI — encompassing infrastructure, legal frameworks, human capital, and institutional capacity. India's score on this index is significantly below that of the United States, which constrains its potential growth dividend from AI. Labour productivity is correctly described: AI raises output per worker by helping workers complete tasks more efficiently, not by eliminating labour wholesale. The third pair is incorrect because 'first-order exposure to generative AI' refers specifically to jobs where workers can perform their existing tasks more efficiently with the help of chatbots or AI tools — it is an augmentation concept, not a displacement concept. It does not mean AI fully replaces all human roles; rather, it identifies occupations where AI can assist workers in doing their work better. The conflation of 'exposure' with 'full replacement' is a common misconception in AI governance discourse.

Explanation

Pair 1 is correctly matched: the Cambridge Analytica scandal involved the mass harvesting of Facebook users' personal data — including political preferences — via a personality quiz, which was then transferred to a political consultancy that used it to build voter profiles and deliver targeted political ads during the 2016 US presidential campaign. This is the canonical example of platform data misuse in electoral contexts. Pair 2 is incorrectly matched: surrogate or dummy political ads are precisely those run by so-called third-party accounts, NOT by the official party account. The defining feature of such ads is that they fall outside the electoral code of conduct that applies to official party handles — they can promote a party or attack opponents without disclosing political affiliations. Saying they are 'run under the official party account and subject to the full electoral code' inverts the concept. Pair 3 is correctly matched: content moderation refers to platform-level review, restriction, or removal of posts and content. Concerns about political bias in this process — platforms allegedly suppressing one political side's content — are a core issue in AI governance and platform accountability debates globally.

Explanation

Statement 1 is correct. GPUs that power Large Language Models are extremely power-hungry and require very large quantities of freshwater for cooling — this is a well-established engineering reality of AI compute infrastructure and is central to why water availability is now treated as a critical site-selection variable for data centres. Statement 2 is incorrect on the principle of cost assignment. Sound data-centre governance frameworks hold that grid assets built exclusively to serve a private data-centre campus should be funded by the developer, not by the State. The rationale is straightforward: if infrastructure is dedicated to a single private user, the beneficiary-pays principle applies; public subsidy would effectively transfer a private cost to taxpayers. Treating such dedicated assets as public utility investments conflates shared-use infrastructure (which does justify State funding) with private-use infrastructure (which does not). Statement 3 is correct. Moving from project-by-project approvals to planning at the campus or infrastructure-corridor level is a recognised governance improvement in data-centre policy because it enables cumulative resource demands — power draw, water consumption, land footprint, grid stress — to be evaluated holistically before any individual approval locks in commitments. Granular project-level review misses compounding effects across multiple co-located facilities.

Explanation

Statement 1 is correct: hyperscale data centres are energy-intensive and water-intensive facilities. Responsible policy frameworks require developers to demonstrate grid readiness (through joint power ramp-up plans with DISCOMs and state transmission utilities) and mandate long-term water availability assessments — reflecting the core principle that AI compute infrastructure must be planned within the carrying capacity of shared physical resources. Statement 2 is incorrect: a foundational principle of sustainable AI infrastructure policy is that claims about renewable energy or battery storage reducing grid draw should be recognised only after the relevant infrastructure is operational and independently verified — not at the proposal stage. Counting unbuilt clean-energy capacity as a grid offset would incentivise greenwashing and misstate actual demand on the grid. Statement 3 is correct: tying incentives to verifiable, milestone-based outcomes (employment, domestic procurement, energy efficiency, water efficiency) rather than to headline investment figures is the cornerstone of evidence-based industrial policy for data-centre ecosystems, ensuring public resources are disbursed against demonstrated, not promised, economic benefits.

Explanation

Both statements reflect the standard economic framework for analysing a major general-purpose technology like AI. In the near term, the massive capital expenditure associated with AI infrastructure (data centres, chips, software) constitutes a demand-side investment boom. This surge in aggregate demand, without a matching immediate increase in supply capacity, puts upward pressure on prices — a classic short-term inflationary impulse. In the long term, AI is expected to raise total factor productivity by enabling economies to produce more output with the same or fewer inputs — a supply-side expansion that pushes the aggregate supply curve outward, which is disinflationary or growth-enhancing. Central banks such as the US Federal Reserve must therefore navigate a transition in which near-term inflation risks may require tighter monetary policy, even as long-term productivity gains eventually ease price pressures.

Explanation

Statement 1 is correct. A task-based framework — the analytical lens used in responsible-AI labour research — disaggregates jobs into component tasks. Because AI may automate some tasks within a role while augmenting or creating others, the same job can simultaneously experience displacement (of automatable tasks) and augmentation (where AI enhances human output). This is the foundational insight that distinguishes task-level analysis from job-level analysis. Statement 2 is correct. Responsible-AI governance frameworks increasingly treat algorithmic monitoring — AI scoring of call-centre interactions, productivity dashboards, real-time exception tracking — as a distinct dimension of job quality, separate from whether someone is employed and at what wage. Even workers who keep their jobs may face intensified surveillance, elevated stress, and reduced autonomy. This is why governance discussions separate 'employment effects' from 'job-quality effects.' Statement 3 is false. The concern about hollowed-out entry-level career pipelines is explicitly directed at India's formal IT, BPO, finance, and professional-services sectors, which have historically absorbed large cohorts of entry-level graduates. Informal workers face AI's effects through different indirect channels — prices, platform management, credit access — not primarily through the disappearance of junior apprenticeship roles. Attributing the entry-level pipeline risk to informal workers inverts the actual structural concern.

Explanation

All three pairs are correctly matched. Model drift refers to the well-documented phenomenon in machine learning whereby a model's predictive performance deteriorates over time because real-world input data changes while the underlying model remains untrained on new patterns — exactly the failure identified in the Tinder case, where scam-detection models had not been updated for years. Human-in-the-loop (HITL) moderation is a standard AI safety architecture where human moderators handle cases that automated systems cannot resolve confidently; the failure to scale HITL capacity alongside a degrading AI layer is a central concern in platform safety literature. Platform accountability is the governance principle — embodied in instruments such as the EU Digital Services Act and India's IT (Intermediary Guidelines) Rules, 2021 — that platforms cannot disclaim responsibility for systematic harms arising from negligently designed or unmaintained algorithmic systems. Understanding these concepts together is essential for analysing regulatory responses to AI-driven harms on online intermediaries.

Explanation

All three pairs are correctly matched conceptual principles in AI governance for education. First, the opt-in consent model holds that AI tools should be disabled by default, with institutions actively choosing to enable them — this preserves institutional and administrative autonomy, as opposed to opt-out models where tools are on by default and must be actively switched off. Second, age-appropriate safeguards are a recognised principle in AI and digital governance frameworks, acknowledging that generative AI access must be calibrated to users' developmental stage — for instance, restricting chatbots for children below a certain age is a standard protective measure. Third, platform dependency risk (sometimes called 'lock-in coercion') is a concept in technology regulation and antitrust policy: when a dominant provider bundles an optional product with an essential service, institutions face structural coercion to adopt the optional product — a concern central to debates on regulating AI deployment through monopolistic digital platforms. All three concepts are foundational to contemporary AI governance discourse in education.