Opening The Rift
© 2026 The Rift. All Rights Reserved.

India can train millions of people to use Artificial IntelligenceArtificial Intelligence (AI)A broad field of computer science focused on creating machines that can perform tasks typically requiring human intelligence, such as learning, problem-solving, and decision-making. and still fail the test that matters most: whether a worker knows when an AI answer is wrong. A certificate might show that someone completed a course or learned to write a prompt. It says far less about the moment a plausible-looking answer could mislead a customer, deny a benefit, or misdirect public money. India’s next step in AI skilling should therefore be a practical test of human judgment.
The country has good reasons to celebrate the scale of its effort. In August, the Ministry of Skill Development and Entrepreneurship reported that its Skilling for AI Readiness programme had expanded into 50 AI and AI-application qualifications, spanning workplace productivity and sector-specific use. The Ministry of Electronics and Information Technology also reported more than 34 lakh registrations and more than 13 lakh certifications on FutureSkills PRIME. Those figures describe different measures and initiatives, but together they show a substantial push to expand access to emerging-technology training.
Yet counting learners and certificates alone can give policymakers an incomplete picture. A learner can pass a test about an AI tool while remaining unable to recognize the circumstances in which its output should be challenged. Employers then discover the weakness after a mistake has reached the public. The cost of that discovery falls on customers, workers and institutions that believed the credential meant more than exposure to a technology.
India’s own Economic SurveyEconomic Survey of IndiaAn annual document presented by the Government of India to both Houses of Parliament, detailing the state of the Indian economy, reviewing developments, and outlining the outlook for the coming year. 2025-26 points toward a stronger standard. Its chapter on the AI ecosystem argues that domain knowledge, analytical reasoning, and the ability to evaluate outputs critically will distinguish capable workers. It explicitly places structured thinking ahead of an obsession with mere prompt engineering. The logic is sound. A fluent answer from a machine becomes valuable only when somebody can judge whether it fits the situation correctly.
Imagine a recruitment trainee using AI to compare applications for a public-facing role. The tool creates an impressive shortlist but quietly treats a career break as a sign of weak commitment. A superficial skills test might ask the trainee to improve the prompt or format the ranking. A meaningful test would ask them to identify the unsupported assumption, examine the actual selection criteria, correct the output, and explain the decision to a supervisor. This is an illustrative exercise, but the capability it measures is very real.
Training providers should build such exercises into their assessments. Give candidates a realistic task, a relevant policy or source document, and an AI output with one or two deliberately planted errors. Ask them to identify which statements the evidence supports, where confidence is unjustified, and when the task requires a human decision. Make them demonstrate what they would do with the answer, rather than merely describing how they would prompt the system.
The same approach works beyond white-collar offices. A small manufacturer might ask trainees to check a machine-maintenance recommendation against a safety manual. A hospital administrator might ask them to reconcile a scheduling suggestion with a staffing rule. A local government office might test whether a draft response correctly applies an eligibility rule. The exercises must reflect local language, occupational practice, and actual authority limits. One generic online quiz cannot establish competence across those settings.
Employers have a role too. An AI credential should be complemented by a short work-based demonstration: one task completed faster, one error caught before use, one decision properly referred to a person with authority. Managers can document these results without building a vast surveillance apparatus. They can compare the quality of a workflow before and after training, including rework and correction time. That produces evidence of useful skills rather than a glossy count of certificates issued.
Such a standard would make the training system fairer. People with strong practical judgment may lack polished English, expensive credentials or familiarity with the newest software interface. A bilingual or locally adapted assessment can show whether they understand a real work problem. If India judges everybody mainly by speed at typing prompts in a particular language, it may overlook precisely the workers who know how a local service actually functions.
There is also a case for public accountability. When taxpayer-supported training programmes report their results, they should distinguish registrations, completions, certifications and demonstrated workplace competence. They should report a small set of practical outcomes by occupation and region. A certification rate is worth knowing. So is the share of graduates who can notice a consequential error and correct it. These measures answer different questions, and India needs both.
The urgency is growing as AI tools gain the ability to take actions. In September, NIELIT and Intel India announced new agentic-AIAgentic AIRefers to AI systems designed to act autonomously, make decisions, and execute tasks or transactions based on their understanding of goals, often involving multiple steps and interactions with other systems. skilling programmes. Systems that can retrieve information are one challenge; systems that can also trigger transactions or send instructions create a greater need to understand authority, exceptions and stopping rules. Training for those jobs should include an exercise in refusing an unsafe automated action and explaining the reason.
This would be a demanding reform, though a workable one. Training providers can start with a small bank of job-specific cases developed with employers. Supervisors can score the trainee’s judgment against a transparent rubric: source checking, error detection, escalation and explanation. Independent samples of those assessments can reveal where courses need improvement. The aim is to reward the skill of making an informed decision under imperfect conditions.
India has already invested in bringing AI education to a wide range of people. It should now ask a harder question of every credential: can its holder use the technology responsibly when the answer looks convincing and the stakes are real? An economy that measures that capability will have a better chance of turning a mass training drive into dependable productivity and public trust.
Disclaimer:The views and opinions expressed in this article are those of the author(s) and do not necessarily reflect the official policy or position of The Rift.



