Can AI-driven growth be made responsible? An Expert Explains how

Many countries, including India, are embracing AI as a pathway to growth, competitiveness and national power. How did AI adoption become such a widely shared political goal?

The story of AI as a national imperative is, at its core, a story about the seductiveness of technological shortcuts. For governments navigating post-colonial legacies, widening inequality, and the anxieties of geopolitical irrelevance, AI offers a way to skip the messy, slow work of structural transformation and arrive at a future that looks like growth through algorithmic efficiency. The US and China modelled this playbook aggressively, and the pressure to follow has been enormous in the so-called AI Arms Race, with countries jostling for AI supremacy.

What the AI Resist List reveals is that this “AI as development” framing is itself a form of narrative power. It is one of the four pillars we document: Resist, Refuse, Reclaim, and Reimagine. The narrative infrastructure of AI — the stories told about what AI is for, who it serves, and what progress looks like — is as material as the data centres and chip supply chains. Countries like India have absorbed not just the technology but the idea that power in the 21st century requires AI dominance, and that resistance stands in the way of progress.

This framing obscures what actually drives these decisions: defence procurement interests, surveillance state consolidation, investor capture of public infrastructure, and the very concrete political utility of appearing modern and as an “AI Leader”.

The list breaks AI power into interconnected pillars: labour, infrastructure, surveillance, data, and narrative. India appears across all of them. What does India’s position tell us?

India’s appearance across all the pillars is no coincidence. AI power is assembled through extraction at multiple levels, and no country is simply a victim or a beneficiary. India occupies a revealing position: it is simultaneously a site of exploitation and a site of aspiration.

On labour, India supplies much of the invisible workforce that makes AI systems legible. On infrastructure, data centres are being pitched as development anchors while drawing on scarce water and electricity. On surveillance and data, facial recognition systems, networked CCTV, and the vast reserves generated through Digital India raise significant questions about oversight and consent.

Thus, AI power is not assembled in Silicon Valley and then shipped outward. It is assembled through relationships that are extractive and unequal. AI’s global supply chains depend on hierarchies that have structured capital for generations.

AI companies often present data centres and computer infrastructure as engines of development. What questions should communities and policymakers ask before accepting those promises, especially in countries facing water stress, electricity pressures and uneven development?

In my mind, the single most important question is: who bears the costs, and who receives the benefits? For example, data centres consume vast quantities of water for cooling and electricity that, in water-stressed and power-insecure regions, comes directly at the expense of households, farmers, and small businesses. The promised benefits of jobs, tax revenue and a stronger position in the digital economy are unevenly distributed. The engineers who staff these facilities are not from the communities that lose access to water.

View of the Yotta D1 building’s rooftop chiller units at the complex in Greater Noida on March 2, 2026. Photo: Tashi Tobgyal

Policymakers should ask whether commitments are binding, whether communities can refuse or renegotiate projects, whether environmental and social impacts have been independently assessed, and whether data governance and technology transfer arrangements genuinely benefit local communities.

India wants strategic autonomy in AI while remaining dependent on American chips, cloud providers and frontier models. Are genuinely sovereign AI ecosystems realistically possible for countries outside the US and China?

In the short and medium term, probably not in any comprehensive sense. The infrastructure that underpins AI, from frontier models to semiconductor supply chains, is so deeply concentrated that formal sovereignty coexists with profound dependence. A country can localise data and build its own large language model while still relying on foreign cloud infrastructure and Nvidia chips subject to US export controls.

This is not a reason to abandon sovereignty as a goal, but it is a reason to be precise about what sovereignty would actually mean. The most achievable forms of AI sovereignty are about the capacity to regulate, audit, and refuse.

India has become central to AI data labelling, moderation and annotation work. Why does the AI industry erase the labour that makes these systems possible?

Because erasure is structural, not incidental. The magic trick of AI, and the reason it commands the valuations it does and the political imagination it captures, depends on the appearance of automation. If you could see the worker in Hyderabad who spent 12 hours today labelling images of car accidents for a self-driving system, or the moderator in Bengaluru who reviewed thousands of pieces of graphic content this week to keep a platform clean, the automation thesis collapses. The labour has to be made invisible for the product to function as promised.

This is data colonialism in its most concrete form: the extraction of cognitive labour from the Global Majority to build systems whose profits accumulate in the Global North, under conditions that echo colonial relationships through low wages, weak labour protections, and little control over how the systems are used. The industry’s silence about tech workers is a deliberate choice, not an oversight.

Do you see AI labour in countries like India as a new version of digital outsourcing, or something structurally different?

Structurally different, and I think more extractive in specific ways. Earlier outsourcing waves, from call centres to coding, involved visible functions that appeared in corporate accounting.

However, AI data labour is more insidious because it is designed to disappear into the product. The worker’s contribution is absorbed into a model, flattened into a training dataset, laundered into a benchmark score. There is no moment at which the system announces that a capability exists because of work performed by a specific human being under specific conditions.

There is also a deeper question of power asymmetry. Earlier outsourcing relationships, however unequal, at least positioned the outsourcing company as a visible counterparty with some accountability to contract terms. Much AI annotation work today is organised through platform-mediated gig structures designed to avoid employment relationships, leaving workers with fewer protections, less collective power, and little visibility into how their work is used.

What forms of accountability efforts indicate that the current trajectory of AI development is not inevitable?

The database (of the AI Resist List) documents efforts ranging from legal challenges in European courts to community refusals of facial recognition in public housing and worker organising across annotation supply chains.

What gives me hope is the diversity of these efforts and the degree to which they are connected, even when the people involved do not know each other. A community in the US refusing a facial recognition contract and an activist in India challenging the Digital Personal Data Protection Act are both insisting on the same principle: that the deployment of technology requires democratic consent and a genuine possibility of refusal.

Whether AI becomes democratic infrastructure will depend on whether institutions can provide meaningful accountability and whether communities retain the capacity to challenge technologies that affect their rights.