The most consequential economic question raised by artificial intelligence may not be how many jobs it destroys. It may be how many decisions it quietly removes from human hands. For two centuries, technological change largely followed a reassuring pattern. Machines displaced particular tasks, sometimes entire occupations, but usually expanded the frontier of human agency. Tractors reduced farm labour; factories mechanised production; computers automated clerical work. Yet engineers, managers, analysts, doctors and other professionals retained something machines did not possess: judgement.
AI is beginning to disturb that settlement. The public debate still revolves around employment: which jobs will disappear, which skills will become obsolete, and which new occupations will emerge. These questions matter. But they risk missing a subtler transformation. Generative AI does not simply automate labour. It increasingly enters the terrain of structured judgement: comparing alternatives, identifying patterns, ranking risks, drafting recommendations and proposing decisions.
Let us call this distinction decision poverty. Decision poverty is not unemployment or underemployment. It is the gradual erosion of opportunities to exercise consequential judgement. People may remain employed, productive and reasonably paid, yet the space in which their judgement shapes outcomes can shrink. Expertise becomes verification. Choice becomes approval. Authority moves upstream to systems designed, owned or trained elsewhere.
The employee remains; agency recedes. This is no longer hypothetical. The International Labour Organisation reported in 2025 that one in four workers worldwide is in an occupation with some exposure to generative AI, while stressing that transformation, rather than outright replacement, is the more likely outcome. The IMF has estimated that around 26% of employment in India falls into high-exposure categories. These figures should not be read as forecasts of mass redundancy. They point instead to something closer to the core argument: jobs can survive while their internal architecture changes.
Consider the financial analyst reviewing machine-generated investment scenarios rather than constructing them; the software engineer correcting AI-produced code; the consultant beginning with an algorithmically generated strategy; the customer-service executive following predictive prompts; or the doctor confronting a machine-generated diagnostic recommendation before forming an independent view. None is unemployed, but each faces a shift from producing judgement to supervising it. The danger is not that algorithms will always decide better. It is that organisations may gradually redesign work around the assumption that they do. Once recommendations become default settings, disagreeing with the machine becomes costlier than following it. Accountability, however, rarely migrates as neatly. A manager, doctor, civil servant or banker may still answer for an outcome even when the reasoning that shaped it was partly generated by systems they neither designed nor fully understand.
Tamil Nadu offers an unusually revealing test case, not because it is uniquely vulnerable, but because its strengths place it close to the frontier.
AISHE 2023-24 puts Tamil Nadu's higher-education Gross Enrolment Ratio at 52.3%, among the highest of India's large states. Chennai has also become a major centre for IT, engineering, finance and Global Capability Centres. State data report 305 GCCs in Chennai in 2024-25, while Tamil Nadu's electronics exports reached $14.65 billion that year. This is an economy increasingly built around organised expertise, technical competence and knowledge-intensive production.
Precisely for that reason, the State may encounter decision poverty earlier than others. AI's strongest effects are moving beyond routine clerical work into highly digitised professional tasks. The ILO's latest occupational analysis specifically notes growing exposure among financial analysts, application programmers, web developers and investment advisers. These are not peripheral occupations in a modern service economy. They sit close to the professional core that states such as Tamil Nadu, Karnataka and Telangana have spent decades building.
The vulnerability therefore emerges from success. Tamil Nadu made the difficult journey from agriculture to industry and from industry toward a knowledge economy. But what happens when codified knowledge is no longer scarce? If analysis, drafting, prediction and technical synthesis become cheap, the premium may shift from possessing knowledge to retaining authority over how it is used.
This has consequences beyond wages. Modern professional societies rest on an implicit bargain: education produces expertise; expertise earns responsibility; responsibility confers status, voice and participation. Decision poverty can loosen that chain. A society may become more educated and technologically productive while increasing numbers of people discover that they execute frameworks they did not design, validate recommendations they did not generate, and remain accountable for decisions whose logic sits partly inside opaque systems.
The political fault line of the AI era may therefore concern not only ownership of capital, but ownership of judgement. Who can override an algorithm? Who controls the criteria embedded in automated recommendations? Who has access to the data from which decisions are derived? And when machine advice goes wrong, who has both the authority and the confidence to say no?
None of this justifies technological nostalgia. AI can widen access to expertise, improve diagnostics, reduce drudgery and raise productivity. Human judgement may be augmented rather than displaced. The policy challenge is therefore not to preserve every old task. It is to preserve meaningful human agency where consequences are serious, and uncertainty cannot be reduced to probability scores.
The next labour-market debate should ask not only whether people have work, but what kind of authority remains inside that work.
A prosperous society can survive automation. It should be more cautious about automating away the habit of judgement itself. Agency, unlike efficiency, cannot simply be outsourced forever.
Thakur is Professor and Director, Centre of Excellence (CoE) for Public Policy, Sustainability and ESG, Alliance University, Bengaluru