In an AI era, PhDs will be more important than ever | Times Higher Education (THE)

Recently, an Anthropic researcher warned PhD students that their career trajectories will be disrupted by AI. After all, AI has already begun solving age-old mathematics problems and producing scientific papers that can pass peer review. Research has even found that AI-generated article reviews “dominate human peer reviews at their best”.

Yet those forecasting the end of the PhD degree carry an incomplete understanding of its purpose – and the importance of that purpose. AI’s ability to accomplish discrete tasks such as writing code and solving mathematical problems does not replace the role of experts who can produce, apply and curate knowledge for the betterment of human society over a lifetime. These are the people that, at their best, doctoral degrees prepare.

There is a lot of talk about managing the risks of AI by keeping humans “in the loop”. But as time goes on in an AI-dominated, post-PhD society, humans risk falling ever more distant behind the research frontier. So even if they are “in the loop”, they will be unable to exercise their own judgement and correct AI’s mistakes and poor decisions about the trajectory of knowledge.

Unlike with other technological advancements – from the calculator to the internet search engine – human society is thus at risk of ceding its collective agency in determining how it progresses.

In light of these dangers, trustworthy humans who can keep up with and direct the machines will become highly valued by policymakers, corporations and broader society. In particular, research experts with a reputation for deep knowledge and sound judgement will become invaluable.

PhDs were always designed to develop depth and expertise, of course. But trends in higher education have pushed many research-focused institutions towards the production mindset. As tenure-track faculty positions become increasingly competitive, it becomes easier and perhaps more “objective” to select candidates based on productivity, rather than more vague notions of expertise.

This competition, in turn, has had knock-on effects on how PhD students are trained. More and more PhDs are encouraged to spend substantial time producing incremental research, often based on data availability rather than substantive coherence, to ensure that they have sufficient publications before graduation. This strategy might pay off in the short run, but it comes at the expense of developing a deep repertoire of domain knowledge.

Global university rankings also reinforce the tendency to overemphasise output metrics. Most rankings dedicate large portions of their scores to off-the-shelf citation metrics. University leaders, in turn, are often under great pressure to elevate their institution’s rankings.

Accordingly, researchers-as-producers prioritise publication counts, citations, grant awards and other such output metrics. But if the only purpose of the researcher is to be productive in such a myopic way, then researchers (and universities) will be vulnerable to being replaced by AI. Asking researchers to use AI to produce more does not change this fundamental problem.

To be clear, it is not that publication and citation counts have no place in the academy. Publications are indeed the building blocks of the scientific enterprise, but these pieces must connect into something larger. Unlike researchers-as-producers, researchers-as-experts confine themselves to publishing a coherent set of high-quality papers, representing a focused and impactful accumulation of deep domain knowledge.

In an ideal world, of course, citations indicate how one piece of research has meaningfully impacted others, but a variety of unsavoury practices have eroded the reliability of that signal. There are now schemes selling fake references to researchers, and the advent of AI-assisted research has generated thousands of hallucinated citations. Paper mills, citation cartels and other related issues have similarly corrupted bibliometric metrics.

Universities must therefore look beyond the naive counting of citations when they assess research and those who produce it. Outputs must be scrutinised for quality, not only quantity. Beyond that, universities must select, nurture and promote genuine expertise and safeguard their reputations for training and employing trustworthy experts.

That second aim demands a more holistic institutional attitude towards scientific careers that, as well as research papers, emphasises research communication, translational work for societal impact and broader stakeholder network building. The entire pipeline of PhD training, new faculty hiring, and promotions and tenure must reflect these priorities.

These points do not provide a full policy road map, and the conversation among university leadership must continue about how to retain the value of human scientists in the age of AI. Despite the challenges, there can be no ducking that conversation if the university – and human expertise – is to endure.

Jonathan A. Chu is an associate professor and assistant dean of research at the Lee Kuan Yew School of Public Policy, National University of Singapore.