Thoughtlas.Our Story

Why Indian University Students Are Submitting AI-Written Assignments — And What Faculty Can Do About It

By Thoughtlas Team

Why Indian University Students Are Submitting AI-Written Assignments — And What Faculty Can Do About It

Walk into any faculty lounge at a private university in India today — Galgotias, GLA, Amity, Sharda, Bennett — and the conversation is the same: "The assignments all look the same," "The language is too perfect," "They clearly didn't write this." AI-generated assignment submission has become the most pressing academic integrity challenge Indian higher education has faced in a generation. But the real problem isn't the tool — it's the design. Here's what the research says, and what faculty can realistically do right now.

The Scale of the Problem in Indian Higher Education

India has one of the largest higher education systems in the world — over 1,000 universities and 40,000+ colleges enrolling more than 40 million students. The proliferation of ChatGPT, Gemini, Claude, and dozens of Indian-language AI tools has put powerful content generation capabilities into the hands of every student with a smartphone.

The specific pressures of Indian university culture make AI misuse particularly attractive:

Assignment volume is extremely high. Engineering and management programs, in particular, assign continuous internal assessments, lab reports, seminar papers, and project write-ups throughout the semester. Students managing 5–6 subjects simultaneously find AI an obvious pressure valve.

Marks allocation creates perverse incentives. When internal assessment carries 30–40% of total marks and is graded primarily on submission quality rather than demonstrated reasoning, the rational calculus for students is clear: a polished AI submission outperforms an honest but imperfect personal effort.

English-medium instruction creates an additional barrier. For students whose first language is Hindi, Bhojpuri, Marathi, or any of India's 22 official languages, writing fluently in English is genuinely effortful. AI eliminates this barrier entirely — which means the attraction isn't always intellectual laziness; sometimes it's language anxiety.

Why Detection Alone Is Not the Answer

Many institutions have responded to AI-generated submissions by deploying AI detection tools — Turnitin's AI detector, GPTZero, Copyleaks. This is understandable, but deeply insufficient for several reasons.

Detection tools are unreliable. They produce significant false positives (flagging legitimately written student work) and false negatives (missing sophisticated AI-generated text). A student who instructs ChatGPT to "write this as a non-native English speaker" or who runs text through paraphrasing tools will often defeat detection easily.

Detection is adversarial. Framing the faculty-student relationship as a detection-and-evasion game creates exactly the wrong culture: students who are motivated primarily by avoiding detection rather than by actually learning. This is the wrong optimization target.

Detection doesn't address the root cause. Students turn to AI when the assignment can be completed by AI and the incentive structure rewards completion over understanding. Detection addresses the symptom. Only redesigned assessment addresses the disease.

The faculty who are most effectively responding to AI submission are not the ones running the most aggressive detection — they're the ones designing assignments that AI cannot authentically complete.

The Root Cause: Assignments That AI Can Do Better Than Students

The uncomfortable truth that every educator needs to internalize: if an AI can complete your assignment in thirty seconds and the result would receive a good grade, the assignment was not actually measuring what you think it was measuring.

AI can:

  • Write a literature review on any topic
  • Summarize any paper or chapter
  • Generate a lab report following a standard template
  • Produce a case study analysis in the format you specified
  • Answer standard textbook questions comprehensively
  • Create a project report on virtually any engineering or management topic

AI cannot:

  • Participate in a live viva voce and answer follow-up questions
  • Demonstrate understanding in real-time oral examination
  • Produce a genuine reflection on what they personally observed or experienced
  • Argue a position convincingly when challenged by a faculty member who knows the domain
  • Apply a concept to a novel situation they've never seen before — in real time, without preparation

The gap between these two lists is exactly where redesigned assessment should live.

Five Assessment Redesigns That Work in Indian University Contexts

1. The Mandatory Viva for Every Submitted Assignment

This is the single most effective immediate intervention. Implement a short (5–10 minute) oral viva component for every written submission. Students who genuinely wrote their assignment can explain their reasoning, their sources, their conclusions. Students who submitted AI-generated work cannot.

The viva need not be elaborate:

  • "Walk me through your main argument."
  • "What was the most surprising thing you found in your research?"
  • "If I challenged point 3, how would you defend it?"
  • "What would have changed your conclusion?"

A student who cannot answer these questions about their own submission has confirmed, definitively and without any detection tool, that the work is not theirs.

Practical implementation: Build 10 minutes of viva time per student into your internal assessment calendar. For larger batches, assign students to viva slots in groups of 5, each student interviewed briefly while others observe. The transparency is itself pedagogically valuable.

2. Process Documentation Requirements

Require students to submit their thinking process alongside the final product. This means:

  • Initial brainstorm or rough draft (handwritten is strongly recommended)
  • A "revision log" noting what they changed and why
  • Screenshots or timestamps showing iterative development
  • A brief "what I found hard about this" reflection in their own voice

AI generates polished finals, not authentic processes. A student who submits a polished report alongside handwritten notes, a rough draft, and genuine reflection on the difficulty is demonstrably engaging.

3. Locally Specific, Experience-Based Assignments

AI's weakness is specificity. Replace generic assignment prompts with locally grounded ones:

Instead of: "Discuss the challenges of rural healthcare in India." Try: "Interview someone in your family or neighbourhood who has navigated rural healthcare in the last two years. Analyse their experience using the frameworks from Unit 4."

Instead of: "Analyse a supply chain disruption." Try: "The district of [your city] experienced [specific recent event]. Using the data from the newspaper coverage you collected this week, apply lean supply chain principles to this real case."

AI can discuss generic Indian healthcare challenges fluently. It cannot interview your student's grandmother. Assignments grounded in personal, local, and real-time experience produce student work that is inherently resistant to AI generation.

4. In-Class Reasoning Sessions (Where Thoughtlas Fits)

One of the most effective ways to verify genuine understanding is to require students to reason through related concepts in class — live, collaborative, and visible to the faculty member.

Platforms like Thoughtlas enable this directly. A faculty member posts a seed question related to the assignment topic. Every student in the lecture hall submits their reasoning simultaneously and anonymously. The faculty member can immediately see whose thinking is coherent, whose is vague, whose demonstrates genuine engagement with the material.

This creates a real-time calibration: a student whose assignment demonstrates sophisticated analysis but whose in-class Thoughtlas contribution is shallow and superficial has revealed a gap that warrants a conversation.

5. Comparative and Evaluative Prompts

AI generates arguments well but evaluates them poorly. Design assignments that require genuine evaluation:

Instead of: "Explain the advantages of cloud computing." Try: "Read these two articles presenting opposite views on cloud security in Indian banking. Which argument is more convincing, and why? Where does each author's reasoning break down?"

Instead of: "Describe transformational leadership." Try: "Evaluate whether Narayana Murthy's leadership of Infosys in its early years is better described as transformational or transactional, using specific evidence. Then argue for the opposite position. Which argument did you find more difficult to make?"

Evaluation requires a genuine perspective. Arguing both sides requires genuine understanding of both. AI can produce arguments — but requiring students to identify which argument they found harder to make and why introduces an introspective dimension that AI cannot authentically provide.

💡 Faculty Tip: The "One Question I Can't Answer" Submission Rule

Add this to every assignment rubric: "At the end of your submission, write one question about your topic that you genuinely could not answer after your research. Explain why you couldn't answer it and what kind of evidence or investigation would be needed."

This single addition does several things at once:

  • It makes intellectual honesty a graded virtue rather than a liability
  • It reveals genuine engagement — students who actually researched the topic encounter real gaps; students who used AI often cannot identify authentic gaps
  • It gives you an immediate topic for the viva: "Tell me more about why you couldn't answer that question"
  • It normalises uncertainty as part of scholarship, which is itself an important academic value

The NEP 2020 Connection: This Is Already Policy

Faculty who feel they lack institutional backing for assessment redesign should note that NEP 2020 explicitly mandates a shift toward holistic, competency-based, multi-modal assessment. The National Education Policy calls for reducing dependence on high-stakes summative examinations and increasing the proportion of formative, ongoing, and portfolio-based assessment.

Redesigning assignments to emphasise process, oral defense, local application, and demonstrated reasoning is not a departure from NEP 2020 — it is its implementation. Faculty who frame assessment redesign in these terms gain institutional language that makes pedagogical reform easier to champion in departmental meetings.

Addressing Faculty Concerns Honestly

"I have 120 students in my section. I can't do individual vivas."

A 5-minute viva per student for one section of 120 is 600 minutes — 10 hours. Spread across a semester with 4–5 major assessments and shared with teaching assistants or junior faculty, this is manageable. But it is additional work. The question is whether the work of genuine assessment is worth performing. Most faculty, on reflection, believe it is.

"My department won't allow me to change the assignment format."

You may not be able to change the format of the submission — but you can add the viva as a separate internal marks component, require process documentation as a submission prerequisite, or build in-class Thoughtlas sessions as participation marks. Most departments have enough flexibility for a faculty member who makes the pedagogical case clearly.

"Students will resent it."

Some will, initially. Students who are caught in the AI dependency habit often experience genuine discomfort when asked to demonstrate understanding they don't actually have. But the faculty who have implemented these approaches consistently report that student relationships improve over time — because students feel genuinely seen, genuinely challenged, and genuinely prepared for placements and professional life where the same performance standards apply.

FAQ

Q: Should Indian universities have a formal AI policy? Yes — urgently. Many still don't. A good institutional AI policy addresses: what uses are permitted, what constitutes academic dishonesty, how violations are adjudicated, and how assessment is being redesigned to reflect the AI landscape. Faculty can advocate for this through academic councils, BOG representations, and NAAC documentation.

Q: Are there AI tools that can help faculty detect AI use more reliably? No detection tool is reliable enough to be used as the sole evidence of academic dishonesty. The viva voce and process documentation approaches described above are far more defensible. Detection tools can flag work for closer scrutiny, but should never be the final word.

Q: What about students who have legitimate learning difficulties and benefit from AI assistance? AI as a writing and comprehension aid for students with documented learning difficulties is legitimate and should be accommodated. The assessment designs above — particularly oral viva and process documentation — can be adapted for these students with appropriate adjustments, the same way any other accommodation is handled.

The Bottom Line

The AI assignment crisis in Indian universities is real, significant, and not going away. But it is solvable — not through detection and enforcement, but through assessment design that makes genuine understanding the only path to genuine marks.

The faculty who win this challenge won't be the ones running the most aggressive plagiarism checks. They'll be the ones who build courses where the only way to succeed is to actually know the material. That has always been the goal of education. AI has simply made the gap between design and goal more visible.

Thoughtlas helps university faculty make student reasoning visible in real time — enabling the live calibration and Socratic facilitation that AI-era assessment demands. Explore it at thoughtlas.com.