Interview with a UCL Professor: AI Is Pulling Out the Bottom Rung of the Career Ladder, and Universities Must Step In

"If we compare career development to a ladder, then what is disappearing is the very bottom rung—and the absence of that bottom rung will inevitably affect the entire career-development system, creating a classic chain reaction. For this reason, higher education will shoulder a more important responsibility in the future than ever before."

Recently, at the 2026 World Artificial Intelligence Conference (WAIC), Mutlu Cukurova, Professor of Learning and Artificial Intelligence at University College London (UCL), told The Paper (www.thepaper.cn) that as the spread of artificial intelligence (AI) reduces entry-level positions, universities need to redefine their role—not only imparting knowledge, but also taking on the capability-building that was once completed on the job.

How education systems should respond to the shock of AI became a hotly debated topic at this year’s WAIC. "New challenges are bearing down on us; we can no longer continue with the old ways and expect top universities to stay ahead, because new winners will emerge in this race," said Andrew Yao (Yao Qizhi), academician of the Chinese Academy of Sciences and professor at Tsinghua University, in his speech. He noted that AI is breaking down the traditional boundaries between disciplines and between universities, and that talent cultivation requires both innovative thinking and execution.

A growing body of research from real educational settings shows that when AI intervenes in learning primarily as a "substitute," it can improve efficiency but does not necessarily promote genuine learning. A randomized controlled trial published last year in the Proceedings of the National Academy of Sciences (PNAS) found that students who used ChatGPT during practice, and then took an exam where AI was not permitted, scored 17% lower than students who had never used AI.

At the same time, teachers are finding it increasingly difficult to tell which assignments were generated by AI, placing higher demands on universities’ existing evaluation systems. The impact is systemic. When students graduate, they may not have truly mastered the competencies their degree is supposed to represent. Employers and higher education institutions trust these credentials, only to quickly discover a vast capability gap. Over time, the damage extends beyond the students who relied on AI to complete their studies—it also means the coherence of the university degree-certification system is unraveling.

"Most universities around the world share a common trait—a strong aversion to risk. These mechanisms played an important role in the past, reducing the risks posed by scientific misjudgment or technological runaway," Cukurova noted. "Yet in an era of rapid technological iteration, some of the very institutions originally designed to reduce risk have begun to constrain—or even obstruct—innovation."

On July 17, Mutlu Cukurova delivered a speech at the World Artificial Intelligence Conference.

Higher education needs a genuine cultural transformation

The Paper: Today, traditional humanities disciplines, business, and even computer science are all being impacted by AI. From the perspective of curriculum systems and program design, how should universities adjust their talent-development approaches?

Cukurova: At the university-presidents’ roundtable during the World Artificial Intelligence Conference, a president also pointed out that in the past, strength in science and engineering was an advantage; but today, it may instead become a new challenge. The reason is that AI is rapidly changing the talent demands in these fields, which means universities need to rethink the importance of different disciplines and how educational resources are allocated.

I also hope that this transformation will have a positive impact. In the future, perhaps more students will return to the humanities—especially philosophy, social sciences, and the arts—because value judgment, creativity, social understanding, and human relationships are precisely the capabilities AI finds hardest to replace.

Yet the greatest contradiction at present is that while society’s demand for these capabilities keeps growing, students’ interest in related fields has not risen in step. This will become a major challenge for higher education in the future.

The Paper: At the end of your speech at the World Artificial Intelligence Conference, you posed a question: "In the face of AI’s impact, can the education system complete its restructuring before the existing model collapses?" Are you optimistic about this? How is the restructuring progressing?

Cukurova: From the perspective of the Western countries I know best—particularly the UK—our overall response to the enormous shock brought by this wave of generative AI, whether in society, the economy, the labor market, or higher education, has been rather slow, and more reactive than proactive.

A very concrete example is the university evaluation system. When large language models such as ChatGPT first became available to the public, many British universities began redesigning how they assess students. In the past, we evaluated primarily the final output students submitted; later, a growing number of schools shifted toward evaluating the learning process. This is because the "proxies" we once relied on to judge students’ abilities suddenly became invalid.

In the humanities and social sciences, we used to ask students to write a paper, submit it, and the school would grade it based on quality and award the degree. Engineering was much the same: a student completes a project, and the school evaluates mainly on the final outcome. But after the advent of large language models, these evaluation metrics lost their meaning almost instantly. Today, a student can hand any of these tasks to an advanced large language model to complete.

Therefore, many universities began shifting to another approach to evaluation. Instead of looking only at the final deliverable, we started paying attention to the entire learning process. For example, we ask students to submit multiple versions of a paper draft rather than just the final version, so we can observe how they revise and refine their ideas step by step. We also increasingly encourage group collaboration. Teachers care not only about what was ultimately accomplished, but about how students thought, discussed, and advanced the work throughout the collaboration. In other words, we began searching for new evidence that could prove students’ genuine learning process.

But the problem is that new challenges have emerged again. In recent months, as AI’s reasoning capabilities have developed rapidly, we have come to realize that even these "process-based evaluations" are failing. The new generation of AI can not only generate final answers but also fully demonstrate its reasoning process. AI can help students complete results, and can also help them "generate the entire learning process." And so the evaluation system we had just built is challenged once again.