In the first half of this decade, the global AI conversation was dominated by leaderboards. Models were ranked by their scores on benchmarks: a model is state-of-the-art because it scored 87.3 on MMLU, or 79.1 on HumanEval, or 92 on some other three-letter test that was not designed with any Indian use case in mind.
These benchmarks measure something real. They do not measure most of what matters. A model that scores top on a multiple-choice medical exam can still produce confidently wrong advice to a duty resident in Vijayawada. A model that aces a coding benchmark can still write the kind of brittle, unmaintainable code that a senior engineer would politely send back for rewrite.
What the benchmarks miss is wisdom, the quality of an answer judged not by whether it is technically correct, but by whether it is the kind of answer a thoughtful senior practitioner would give. We think a community-maintained Wisdom Index, focused on Indian professional contexts, is one of the most leveraged pieces of infrastructure the next three years can produce.
What wisdom adds beyond correctness
Correctness is binary or near-binary. The answer either matches the gold answer or it does not. Wisdom is graded and multi-dimensional. A wise answer is correct, but it is also calibrated in its confidence, appropriate to the user's apparent context, scoped to what the user actually asked, transparent about its reasoning, and useful in a way that fits the user's actual workflow.
A correct-but-unwise answer to a tenancy question correctly cites the relevant section of the state Rent Control Act and provides no warning that the user's described situation has a procedural prerequisite they have not mentioned. A wise answer cites the section, notes the prerequisite, and asks whether it has been met.
A correct-but-unwise answer to a fever case lists the differential diagnosis. A wise answer asks the three questions whose answers would actually narrow the differential, and surfaces the one possibility the user, given their stated context, is most likely to under-weight.
A correct-but-unwise answer to a vendor selection question recommends the vendor with the highest stated rating. A wise answer notes the missing information, payment terms, dispute history, local presence, that would change the recommendation if known.
The gap between correctness and wisdom is the gap between passing the exam and being good at the work. The first is testable in an afternoon. The second is what users actually need.
Why nobody has built a wisdom benchmark yet
For three reasons, mostly tractable.
Grading wisdom is harder than grading correctness. It requires senior practitioners, not undergraduates, to score responses. Senior practitioners are expensive and have less time than undergraduates. So far, almost no benchmark project has been willing to pay the cost of doing this seriously.
The right scoring rubrics do not exist. We do not have a settled vocabulary for what makes a medical answer wise versus merely correct, what makes a legal answer wise versus merely cited, what makes an administrative answer wise versus merely procedure-compliant. The rubrics have to be built sector by sector, with the senior practitioners of that sector at the table.
There is no obvious commercial owner. A wisdom benchmark is a public good. The model labs have no incentive to fund a benchmark that might rank their flagship below a smaller, more careful competitor. The natural funders are professional bodies, public-interest foundations, and the user communities themselves.
What a working Wisdom Index would look like
A serious Indian Wisdom Index has four parts.
A taxonomy of professional contexts, built sector by sector with practitioner participation. Indian primary care, district administration, civil litigation, small-business compliance, agricultural extension. Each is a category with its own rubric, raters, and canonical evaluation cases drawn from anonymized real practice.
A panel of senior practitioner raters, paid for their time. The panel is publicly named at the institutional level, AIIMS participated, the Bar Council of Maharashtra participated, ICAR participated. Individual raters are protected from vendor pressure.
A continuously updated leaderboard, published openly. Each model gets a sector-specific wisdom score, with sub-scores for calibration, contextual appropriateness, reasoning transparency, and language fidelity. Indian-language performance is reported separately, not averaged into an English-dominated score.
A governance body that owns the benchmark long-term. It needs to be Indian, credible, and funded independently of the labs it ranks.
What this would change
Procurement. An Indian institution buying AI capability today is choosing on the basis of marketing claims and foreign benchmarks. A credible wisdom index gives them a sector-relevant Indian basis for comparison.
Model improvement. Labs respond to what is measured. When wisdom in Indian primary care is measured publicly, model improvements flow toward that target. The capital allocation shifts away from longer context windows and toward calibrated uncertainty, contextual fit, and Indian-language reasoning.
User trust. A regulator, a hospital administrator, a senior IAS officer asked to approve AI deployment will trust a system that scores well on a credible Indian wisdom benchmark in a way they cannot trust foreign correctness benchmarks alone.
What you can do
If you are a senior practitioner in any Indian professional field, your time on a wisdom evaluation panel is among the highest-leverage uses of professional time available in 2026.
If you are an institution, a hospital, a court, a department, your participation in defining the rubrics for your sector is the way you get AI procurement that actually serves your users.
If you are a builder, do not wait for a public Wisdom Index to exist. Build the internal one for your domain, with senior practitioners in the loop, and let it shape your roadmap.
Correctness benchmarks got us through the last decade. Wisdom benchmarks will get us through the next one. Start building yours.
Join the conversation
This essay is part of an ongoing community. If it resonated, the next step is to be in the room.
Join Bharath.club → Read more essays