The recent UGC regulations on AI use in Ph.D. theses impose stringent penalties: those showing 10-40% AI content and plagiarism must be revised and resubmitted within six months; those with 40-60% cent face a one-year bar from submission, and anything above 60% can lead to cancellation of registration. The intent is understandable, as these steps address real fears that excessive AI use might undermine independent human analysis but the approach’s overreliance on detection makes it less reliable. 

False positives

Tools such as ZeroGPT or Turnitin scan for perplexity, the presence of predictable, common and expected words or sentences, and also flag low burstiness, the missing natural swings in sentence length, and rhythm that mark human writing. Hence, texts written well before large language models arrived in 2022 may show AI patterns and trigger false positives. There is a high possibility that multilingual scholars, neurodiverse writers, and anyone using standard academic phrasing may be wrongly accused.

To counter these challenges, learners insert obvious errors that do not impede meaning just to dodge the flag. For example, they write “an mean” instead of “a mean”. Some students feed samples of their old human-written text into AI and instruct it to generate text in that exact pattern. Some researchers are now removing their own punctuation, such as em-dashes and commas, because it looks like AI-generated text. The result is regressive: honest researchers damage the quality of their own prose to appear human. 

Institutionalised detection may encourage hidden use rather than open dialogue and create a defensive, fear-based response when surveillance replaces scholarship and critical inquiry. Honest students get caught while clever ones use hacks and tools more effectively to escape. Detectors may not be able to tell the difference between polishing a sentence and writing the whole paragraph. We may begin to reward clunky writing over clear ideas. If the UGC system functions as an AI-written text detector, it will prove regressive. Even if it tries to catch AI patterns of thought, it still falls short without human insight.

In addition, we must ask: do we have detectors to catch doctoral theses written by ghostwriters who are paid a few lakhs to write the entire thesis? How do we tackle the “professional service” provided by firms to humanise AI-generated content? 

Writing in itself is a cultural technology, a tool created by humans to store, process, and transmit information. Interestingly, human cognition is also not localised (internalised) but has always been distributed and technology-driven. Therefore, the real challenge is not in catching machine-generated sentences but in assessing cognition across hybrid systems that include humans and machines. In the AI age, writing is no longer a simple transmission; rather, it is a conceptual orchestration. The writer selects ideas, uses tools, and synthesises their unique voices into a coherent whole, just as a conductor produces music through an orchestra. 

While humans and AI act as instruments working together, the craft lies in the choices of selection, exclusion, authenticity checks, bias negation, arrangement, and synthesis. Generic literature reviews that skip critical gaps, hedged conclusions that avoid taking a clear stance, data analysis that fails to explain its own methods and citations that look perfect but are partially hallucinated reveal the true absence of original thought or human oversight. Software may struggle to spot them, but a careful examiner will notice those effortlessly.

Way forward

The way ahead should be driven by a shift from detection to effective integration. Scholars could file a mandatory AI disclosure statement listing every tool and the exact prompts used. To assess whether the researcher is the conceptual orchestrator who has shaped the machine to resonate with their ideas, the prompts submitted might reveal their intent, thoughts, and vision. In our opinion, such a prompt to write a 2000-word essay would also be close to 2000 words, which may be written/spoken to the machine, as it is during the prewriting stage of writing, when we use freewriting to pool all our ideas and focal points before organising them. In addition, they could attach a metacognitive reflective note explaining how they integrated the output from the initial prompt and where they synthesised fresh insights. 

The viva must become the focal point of the evaluation and should not be viewed as a mere formality. Every PhD viva must have one guiding purpose: to assess whether the candidate is truly the author of the work. That question cuts to the conceptual clarity of what is produced and the process the researcher has undergone to produce it. There can be regulations to stream viva voce examinations online, with open links for anyone across the globe interested in the domain to join and pose their questions. It would turn the defence into a living scholarly debate rather than a closed-room ritual. 

More importantly, we must invest in empowering research supervisors and external examiners to focus on depth of thought, originality, novelty, social relevance, and potential impact rather than on mere pattern-based detectors. Scholars should not live in eternal fear of AI and become cautious avoiders of machines, but rather become confident orchestrators who use AI to produce knowledge that is ethically transparent, deeply reflective, and genuinely transformative for our society. Such an approach would honour the spirit of human academic rigour even as it uses AI. 

The views expressed are personal.

J. Jehoson Jiresh is Assistant Professor of English and Cultural Studies, Christ Deemed-to-be University, Bengaluru. K. Abarna Sri Preethi is an independent Humanities researcher and visiting faculty of English, Bengaluru.