Why Do a PhD in the Age of AI
Developing an intellectual direction of your own when AI can do more of the work
Why should someone spend several years doing a PhD when AI can already help generate ideas, write code and carry out parts of an investigation? As a supervisor, I feel responsible for giving that question a serious answer. I am asking a student to commit years of their life to research. The prospect of a few papers and a qualification is an incomplete account of what those years should offer.
My strongest reason for supporting doctoral training is the opportunity it gives someone to develop an intellectual direction of their own. I want a student to become sufficiently knowledgeable about a problem to question how it has been framed, including by me. I want them to discover questions they care about and acquire the ability to investigate them. AI changes how we can support that development. It also makes it easier to confuse a successful project with a successful education.
I have a particular stake in this distinction. My research involves making robotic systems more capable, and I am interested in automating scientific experimentation. I welcome tools that help us do difficult research. At the same time, I teach deep learning and supervise students who are learning to become researchers. I need to think about what they gain when a tool takes over part of the work, and what experience they still need for themselves.
I have discussed the practice of supervision and the development of research taste in earlier essays. The question here is a prior one: what makes a doctorate a worthwhile commitment for a particular person, especially when other research environments may offer excellent opportunities?
A direction that lasts beyond one project
My own research gives me a way to explain the kind of development I value. In our work on optical microrobots, we investigated how a robot's structure and the distribution of optical forces could enable movement beyond the plane of a conventional multi-spot optical-tweezers arrangement. The study connected physical modelling with fabrication and experiments. [1] The aspect that matters to me here is the relationship between the design of a robot and what control can achieve.
What stays with me from that work is not only the technique. It is the continuing interest in how physical design shapes a robot’s capabilities, and where learning and control can extend them. The experimental methods can change while the underlying question continues to develop. A doctorate can offer the time and depth to begin finding such a direction, including one that leads beyond the project originally offered.
That is an educational opportunity, not an automatic consequence of spending several years in a laboratory. A student needs some influence over what is investigated and enough sustained guidance to make informed use of that freedom.
The case for a PhD should survive better AI
I am less persuaded by a defence of the PhD that depends on finding something AI cannot yet do. A student beginning now may see those limitations change considerably before graduating. Their education needs a more durable purpose.
That purpose includes learning to direct work whose next step is uncertain. A tool may produce several plausible research plans. Choosing between them requires some understanding of the field, the available evidence and the consequences of each choice. Even when AI helps with that assessment, I want the student to participate in the reasoning. I want them to be able to explain why one uncertainty deserves a month of work and another can wait.
An idea need not originate in a student’s mind without assistance to become part of their own research. A suggestion from AI, a colleague or a supervisor may reveal a useful connection. What matters is whether the student investigates it carefully enough to understand the implications and develop a reasoned position of their own.
There is a difficulty here that I do not think we should gloss over. The ability to judge an analysis usually develops partly through doing analyses. Understanding an experimental failure requires some familiarity with how experiments fail. If students delegate most of this experience before acquiring it, asking them to check the output places a demanding responsibility on a foundation they may not yet have.
For doctoral training, I would preserve selected opportunities to work through a problem directly: implementing a small model, tracing a calculation or designing an experiment whose assumptions the student can inspect. The scope can be modest. Once the reasoning is understood, AI can help extend the work. I see little value in making students repeat hours of routine implementation simply because that was once unavoidable. The experience should earn its place through what it teaches.
Compare the actual opportunities
None of this gives universities an exclusive claim to good research training. An industrial team may offer outstanding colleagues, substantial computing resources and problems that become visible only when a system is used outside the laboratory. For someone interested in building and deploying technology, that could be a better environment. Fundamental research also takes place in companies. I would not ask a student to dismiss those opportunities in order to defend the route I work within.
The comparison needs to be specific. A company with impressive technology may or may not give a new researcher the chance to influence its direction. A university project may or may not provide the resources and sustained guidance it needs. I would want to know what the person would actually do in either role, who they could learn from and how much scope they would have to pursue a question beyond the original brief.
In robotics, access to a working experimental system can make an enormous difference. An attractive academic question does not remove the need for equipment, technical support or a feasible plan. Equally, the newest model or largest computing allocation does not by itself determine which questions deserve investigation. I would compare the opportunities to learn and contribute, alongside the resources that make those opportunities possible.
I also think it is reasonable to take salary, funding security and personal circumstances seriously. Several years is a substantial commitment. Curiosity does not make those considerations disappear, and they should not be treated as evidence that someone lacks commitment to research. Going into industry can be an affirmative choice about the work and life a person wants.
Questions I would ask before accepting a place
Before committing to a programme, I would want to discuss the experience it can realistically provide. An impressive project description is a starting point, but it cannot answer these questions on its own.
What is already fixed, and where could the student influence the question or the approach?
Who will provide regular scientific guidance and practical help when the work becomes difficult?
Are the equipment, data, computing and technical support needed for the initial investigation actually available?
How would the project respond if an important assumption proved wrong or the original plan became infeasible?
Do the funding arrangements and expected working conditions make the commitment viable for this person?
I would ask comparable questions about an industrial research role. A university title does not guarantee intellectual freedom, and a company setting does not rule it out. The relevant comparison is between the opportunities someone could actually take, with the people and resources they would actually work with.
It would also be useful to hear from current researchers about an ordinary week: how discussions happen, how decisions are made and how much time goes into developing expertise. A formal description of a programme can leave those parts of the experience unclear.
When I would encourage a different route
I would encourage a PhD when someone wants the depth of a sustained investigation and can find a programme that gives them the support and room to develop it. They do not need to arrive with a perfect question. Part of the education is learning enough to formulate one. But there should be something about the subject they want to understand beyond completing a prescribed set of tasks. A desire for the title alone seems a fragile basis for the years involved.
If someone is mainly excited by building and deploying a particular technology, and a research or engineering role offers strong colleagues, practical experience and room to grow, I would take that option seriously. If the available PhD project lacks the resources or guidance it needs, enthusiasm for the subject does not resolve those shortcomings. It may be better to seek a different programme or gain experience elsewhere before deciding.
Choosing another route need not settle the question for the rest of a career. What matters now is whether the next few years offer a good environment for the work the person wants to learn to do.
What I would hope the student takes away
This also asks something of my own research ambitions. A student may become interested in a direction that I would not initially have prioritised. If they can make a convincing case and the work is feasible, I want to give that interest serious consideration. The doctorate should allow their intellectual identity to emerge. Its success cannot depend entirely on how closely their thinking comes to resemble mine.
The experience should leave a graduate with a clearer sense of the questions they want to pursue, the conditions in which they work well and the people from whom they can learn. Those gains should be useful whether their next step is in a university, a company or another field.
By the end of a doctorate, I would hope a student could explain which problem now matters to them, how their understanding of it has changed and what they want to investigate next. Their answer might take them into a company, another laboratory or a field I know little about. I would regard that as a good outcome. The years we have spent working together should give them a stronger basis for choosing their own direction.
Source and related reading
Continue reading
AI assistance was used in developing and editing this article.
Illustrations were generated with AI. People, settings and research sketches are illustrative; they do not depict actual individuals or experimental results.
Join the discussion
Which features of a research environment would matter most in your decision to do a PhD? Leave a comment below, or use the Like button if you found the essay useful.
Conceptual illustrations
The illustrations below were generated with AI. They are conceptual and do not document actual people, events or experimental results.
Figure 1. A worthwhile doctorate needs more than an interesting title: a question worth sustained attention, access to research resources, thoughtful supervision, time and viable funding. Conceptual illustration generated with AI.
Comments