Education
Testing Thinking, Not Answers: What Assessment Must Become in the AI Age
When I was asked, at a faculty interview many years ago, what the toughest thing about teaching was, I answered without hesitation: grading. I did not imagine then how much harder it would become. Generative artificial intelligence can now produce coherent essays, solve complex equations and write clean code in seconds. The traditional measures we have used to judge student learning are crumbling under that pressure.
The temptation is to fight back with better detection software. I believe that is the wrong battle. The answer lies in a fundamental rethinking of what we assess and how — moving from judging finished answers to judging the thinking that produces them.
The question every institution must now face
When information is everywhere, the professor is no longer the sole gatekeeper of knowledge. That shift raises uncomfortable structural questions for higher education. What should we teach when any fact is a few keystrokes away? And, more pointedly, what are our grades really measuring?
If a student can deploy an algorithm to secure top marks, what do our grading systems actually measure?
In India, the policy intent to respond already exists. The University Grants Commission’s 2023 advisory asked universities to frame comprehensive AI guidelines, and the Central Board of Secondary Education has urged schools to move from rote learning towards critical thinking. But classroom practice lags behind. Assessment methods remain archaic, and the AI detectors on which many are pinning their hopes are fundamentally unreliable. Guarding legacy metrics with unreliable tools is not a strategy.
Four assessment designs that measure real thinking
What we need instead are tech-immune, classroom-centred assessments that replace the transactional, point-scoring model. Four approaches stand out.
1. Bring back the viva voce
A meaningful share of grading weight should move to in-person oral defences. Rather than judging a file uploaded minutes before a midnight deadline, we should judge whether a student can stand before faculty or peers and defend, dissect and unpack an argument under live questioning. Doctoral programmes have long relied on this rigour; undergraduates deserve the same standard.
2. Use live-patch diagnostics
Assess execution, not just the end product. Give students deliberately broken code, a flawed mathematical proof or a weak historical argument, and ask them to troubleshoot it in real time without digital help. This is how problems arrive in the workplace — imperfect, incomplete and urgent.
3. Make AI auditing an assignment
Banning these tools is fruitless. A better approach asks students to evaluate AI-generated output critically: spotting factual hallucinations, verifying sources and identifying where the machine’s reasoning runs out. That builds genuine digital literacy rather than passive dependence.
4. Grade for mastery, not compliance
Partial-credit systems often reward compliance more than competence. An “edit-to-mastery” contract grading model works differently. Major assignments follow a Complete/Revise loop: work is marked Complete only when it meets strict, pre-published professional standards; otherwise it is returned for further rewriting. A case analysis in Strategic Management, for instance, would pass only if it applied the required analytical frameworks and offered data-backed recommendations. Under such a system, an ‘A’ signals demonstrated mastery, not accumulated points, and top marks regain their meaning.
From an administrative problem to a philosophical one
Grading was once difficult because educators had to separate conceptual understanding from ordinary human error. Today the task is to separate authentic human thought from machine output. That makes our central challenge philosophical rather than administrative. We must redefine the purpose of the physical classroom: less automated assessment, more cultivation of deep, focused human intellect in an era designed to distract it.
The irreducibly human role of the teacher
AI has stripped education down to its core. It can synthesise data, but it cannot mentor. The educator’s lasting value lies in work no algorithm can replicate: noticing each student’s trajectory, challenging their assumptions, and holding them to a standard higher than a machine can meet. Assessment redesigned around thinking is how that value becomes visible — to students, to employers and to institutions themselves.
Conclusion
Generative AI has exposed how much of our assessment rewarded the production of answers rather than the quality of reasoning. Oral defences, live diagnostics, AI auditing and mastery-based grading offer a practical way forward. With the National Education Policy 2020 and forthcoming UGC directions providing a framework, Indian universities have a genuine opportunity to turn this technological shock into a lasting improvement in how we teach and how we judge learning.
