PhD, quo vadis?
The Navier Stokes result is overflowing in all of our social media feeds. There are posts that proclaim the arrival of AGI, with others being more measured in their response. Some contest the nature of the discovery, while there are reports of how unpublished results were allegedly used by the OpenAI models for arriving at theirs. Along another dimension are the more passionate/philosophical calls to the spirit of human endeavour by the likes of Terrence Tao, followed by collective declarations, and now a statement from the Clay Institute that has instituted the Millenium Prizes.
These are very interesting conversations to have, especially in the light of my colleague’s Siddharth Panwar’s recent comments in our Institute Colloquium (which I paraphrase) that the fundamental way we relate to work, and the way it relates to our identity and notion of self-worth is going to drastically change. And with the recent OpenAI results, I don’t think that time is in the distant future.
I am coming at this from a different direction.
In a week’s time, I am going to coordinate and also deliver a series of lectures to a batch of about 40 students on the topic of Research Methodology. This is an Institute-mandated compulsory 1-credit course that has to be completed by all research students for the continuation of their research activities. Conventionally, courses like these consist of lectures on the principles of research methodology (Pasteur’s quadrant, qualitative vs. quantitative research, research workflow, etc.), ethics, how to write for publications, prepare presentations, and the like.
More recently, the courses have been updated in piecemeal by its instructors to include components of AI usage. For instance, I have been delivering a lecture on Literature Reviews for the past 4 years. In the first year, I was focusing more on bibliometric tools. With time, I graduated to AI-guided reviews, the problems of hallucinations, etc., and in the last iteration I also introduced them to literature review agents. The common thread that guided my delivery was that the use of tools that accelerate one’s research workflow is good, but the ‘human-in-the-loop’ is important to carry out impactful research. In my opinion, an element of the personal is crucial not only for ownership, but also for arriving at truly impactful innovation.
But I cannot also limit to my worldview as a tenured academic. The stakeholders on the other side of the lectern are 25-something olds, waiting to make a way for themselves in the world. They comprise of both intrinsically motivated individuals, and those who see the degree as a rung on their career ladder. For both, the changing landscape of AI-enabled research and writing tools holds the promise of fast-tracking this journey.
There is now a collision of worlds, with the old-school believing that research is an individualistic expression that demands your all – and the new school seeing papers and the thesis as an MVP for getting the degree and moving on.
The dearth of Institutional guidelines on AI use makes the situation nebulous, as there will be no scaffolding available for conflict resolution between the supervisor and student.
If not from within, then from without?
Publication houses now have sufficiently detailed guidelines on AI use. With them being the penultimate gatekeepers (prior to our peers), the guardrails put by them can potentially serve as guidelines for students on what-, and what-not-to-do when it comes to research.
Springer Nature has a sufficiently detailed set of guidelines, which does not prohibit the use of AI tools for grammar and polishing prose, but starts drawing the line at methodology formulation and even “comparing results to existing literature”.
Be it publication houses, conferences, or institutions, there is no question about ownership – if you write it on the paper, you have to own it. In general, the point of no return is a full delegation of critical thought to the AI, where it is AI that identifies the research gap, frames the questions, identifies the methodology, carries out the experiment, generates the data (by way of coding perhaps), summarizes with critique, and essentially also writes the paper.
But I think a time will soon come when these guardrails/guidelines will also need a rethink. We now have agents that do the research and help discover new knowledge. Further, the OpenAI results clearly signal the changing face (and scale) of research discovery.
Do we need to rethink assessment?
This brings us back full circle to Siddharth’s point that probes the very existential nature of researchers, which is very much in the vein of Tao’s position on the matter.
Based on my experience within the UK Higher Educational (UKHE) system, I typically lay out the following to incoming research students as the primary evaluation criteria for the award of a research degree –
- The work done as part of your research is novel.
- The work is done by you.
- The work done is peer reviewed and accepted by the community (i.e. you have one or more publications).
It is interesting to note that the contribution of the student is implicit in the above. That is, it is assumed if the work done is novel, and done by the student, it automatically is representative of the critical thinking skills and capability of the student in carrying out independent research.
I believe these criteria were put in place to indicate whether the student/research candidate is capable of carrying out research independently. In my opinion, these still remain essential, especially if publishing houses assign ownership of the paper and its contents upon the author/s – warts and all.
However, AI use has now completely penetrated all three of the above evaluative frameworks.
For instance, a student can use AI
- for identifying a critical knowledge gap – thanks to agents.
- for identifying and constructing process workflows – including generation of code and the like) for carrying out experiments.
- for analysing the data and write the paper – either partially or in its entirety.
- to respond to peer reviews with AI – a Pandora’s box on both sides of the table.
- to write the thesis.
So, how do we go about assessing the student and recommend award of the degree? And should the criteria/benchmarks for assessment change?
The Academic Pincer movement
I feel we have to come at this from two directions.
1
Levelling the playing field is one possibility. There are several Indian institutes that have laid out AI-usage policies, either at Institutional or a departmental level based on the tenets of equity of access, transparent disclosure, and use with adequate deliberation, whilst underscoring the pitfalls of AI use. The move from a blanket ban to reflective use is in keeping with global competition. This will also lower several barriers to entry – be it language, coding, or even representational skills.
2
Given this, the viva-voce becomes the ultimate trial-by-fire of research acumen. PhD examiners will then have the unenviable role of assessing the capability of the researchers on all levels of cognition/meta-cognition (e.g. using frameworks like the Anderson Kratwohl/SOLO taxonomy). Beyond creation of new knowledge, the capability to evaluate information critically would need a more eagle-eyed assessment. Original approaches to problem solving, and ingenuity would come a close second. These skills will remain relevant even with the availability and advancement of AI tools.
The above would indubitably be obvious to the more experienced academics amongst us. But the dynamically transforming landscape of research that we are seeing today requires stewardship for students. A stronger emphasis on the viva-voce would send a clear signal to the research students on the grounds for award of their degree, and the expectations from a person holding one. Working backwards from the point of assessment, the students will then see the need to know their basics without overly relying on AI tools, and on the importance of critical assessment and deliberation at each step of their research journey. The production of original independent work will still be a necessary criterion for graduation, but not a sufficient one. The shift is significant, as the award of the research degree will require a systematic assessment of the intellectual and research capabilities of the student – beyond their immediate research contributions.
Science does not, and will not need saving. Market forces, and our innate curiosity are stubborn beasts.
However, the onus of pressing the button still lies upon us – and when it comes to tipping points and singularity events which requires us to think before we leap, we cannot be lax.
This work is licensed under Creative Commons Attribution-ShareAlike 4.0 International
AI declaration statement
AI was used to collect AI Policy resources across HE institutes – prompt used – “I am currently researching the status of AI usage policies within higher education institutions in India. Can you give me an overview of this landscape?” Response – https://chatgpt.com/s/t_6aa4fce238b881919633dba9bafafec3
No AI tools were used to write the above article. All opinions are my own.
Resources
- Artificial Intelligence (AI) | Nature Portfolio
- CVPR 2026 Author Guidelines
- IIT Delhi AI policy
- Central University of Punjab AI Policy
- IISER Pune Guidelines to Generative AI usage
- IIT Bombay ET – Guidelines on AI Usage
- AI Policy for teaching, research and administration, Department of HSS, IIT Bombay