Why a PhD Still Matters in the Age of AI
Source: Phillip Isola, “On the Value of Doing a PhD in the Age of AI”
With frontier AI models improving quickly, it is reasonable to ask whether spending four or five years doing a PhD still makes sense. A research topic that looks important today may look different a few years from now, and AI is already taking over parts of research that once required substantial human effort.
Phillip Isola’s article offers a useful way to think about this. One point that particularly resonates with me is that becoming an expert still takes time. If it once took roughly N hours of serious study, experimentation, failure, and reflection to understand a field deeply, AI is unlikely to reduce that to a tiny fraction of N. It can help us learn faster, find information, write code, analyse data, and test ideas, but we still need to build our own mental models of the subject.
That expertise becomes more important, not less, as AI becomes more capable. If we use AI to conduct research, design systems, analyse evidence, or generate solutions, someone still needs to understand the domain well enough to ask:
- Does this result make sense?
- What assumptions is the model making?
- What has it missed?
- Is this explanation consistent with what we already know?
- Most importantly, what should we investigate next?
Without human expertise, it is difficult to distinguish a convincing AI-generated answer from a genuinely useful contribution.
The second point concerns the nature of a PhD. A PhD is not simply several years spent mastering today’s techniques. It is training in how to explore something that is not yet well understood. Research begins where existing knowledge becomes incomplete.
The rapid development of frontier models does not necessarily make a PhD topic obsolete. The frontier moves, and researchers move with it. Methods may change, questions may need to be reformulated, and tools unavailable at the beginning of a PhD may become central halfway through it. That is part of research. Learning to revise our understanding as technology changes may become one of the most important research skills in the AI era.
My takeaway is not that AI leaves PhD research unchanged. AI will change how we search the literature, build prototypes, run experiments, analyse results, and test ideas. But faster tools do not remove the need for deep understanding.
As long as important questions remain unanswered, we will need people willing to spend the time required to understand them deeply and explore beyond the current frontier. That, ultimately, is what doing a PhD is about.