An era has begun in which generative artificial intelligence (AI) writes research papers and even helps review them. In research settings, AI-generated references are now being cited in published papers, and researchers are receiving review comments written by AI. Neither is unfamiliar anymore.


Many people ask whether AI can replace scientists. But the most important issue right now is not AI's capabilities. Before discussing AI, we need to look back at how science has developed.


The philosopher of science Karl Popper said, "Science advances not through 'proof' but through 'falsification.'" No matter how many experiments support a theory, they cannot prove it is eternally true. Conversely, a single reproducible counterexample can overturn an established theory.

A scene in a laboratory where a human scientist and AI collaboratively review a research paper. Image generated by ChatGPT.

A scene in a laboratory where a human scientist and AI collaboratively review a research paper. Image generated by ChatGPT.

View original image

Science, then, is not a discipline devoted to producing a large number of correct answers. It is a process of constantly questioning, testing the possibility of being wrong, and provisionally accepting explanations that withstand scrutiny. For this reason, science is often called the discipline of data, but more precisely, it is closer to a "discipline of trust."


Science can move forward only when researchers are trusted to record experimental results as they are, reviewers to assess papers without conflicts of interest, and other researchers to repeat experiments and verify the results.


Research papers are the most important means of making research findings known to the world. But a single paper does not, by itself, complete the work of science. Only after other researchers verify and reproduce its findings can it become established as new knowledge.


AI is changing this structure. Generative AI has dramatically reduced the time needed to find literature, write abstracts, and draft papers. In fact, a study published in 2025 analyzed more than 15 million biomedical paper abstracts from 2010 to 2024 and found that expressions frequently used by generative AI surged in papers after ChatGPT emerged. The researchers also suggested that at least 13.5% of scientific papers published in 2024 may show signs of AI-assisted writing.


But as the number of papers grows, verification becomes more important than production. In 2023, more than 10,000 papers were retracted worldwide, a record high, and major international academic publishers have also reported that the burden of verifying papers is growing rapidly in the AI era. AI has made it possible to write papers faster, but it has not increased the speed at which those papers can be read and verified. The greatest concern in the global academic community right now is not AI itself, but the verification system's inability to keep up with it.


Fabricated references generated by AI are appearing in published papers, while the number of researchers available to verify the rapidly growing volume of manuscripts is dwindling. AI can ease researchers' workloads, but at the same time, it is placing new strain on the verification system that science has maintained for centuries.


So, will AI eventually replace scientists?


Michael Polanyi, a British scientist and philosopher, said, "We know more than we can tell." His point was that the knowledge scientists accumulate through intuition, experience, and failure cannot be fully explained through papers or formulas alone. A good researcher is not someone who knows a lot of data. It is someone who can intuit what to question, which results seem unusual, and when to abandon conventional wisdom.


A famous scene in the science fiction novel The Hitchhiker's Guide to the Galaxy symbolizes this idea. After calculating for 7.5 million years, a supercomputer gives the number "42" as the ultimate answer to life, the universe, and everything. Author Douglas Adams said he chose it as an ordinary, meaningless number, aiming for absurd humor, but later generations of scientists and programmers came up with ingenious interpretations. They noted that the number 42 corresponds to the asterisk (*) in ASCII code and that the asterisk is used in programming as a wildcard that can stand for "anything," adding the interpretation that "the answer is for us to fill in ourselves."


But the characters in the novel who asked the question did not know what the answer meant. To understand the answer, they first needed to know what the "ultimate question" was. This biting satire speaks directly to the AI era. No matter how sophisticated and fast generative AI is at producing answers (data), it is ultimately up to humans to read the context behind that data and ask about new directions: "Is this really the best we can do?" AI can read tens of millions of research papers and combine their answers, but it is difficult to say that it has learned how to ask new questions.


Newton wondered, as he watched an apple fall, "Why does it fall downward?" Einstein imagined, "What would the world look like if I traveled alongside a beam of light?" Great discoveries have always begun with questions, not answers. Science has advanced not through those who write papers the fastest, but through people who hold on to the most important questions and verify the answers to them.



AI learns from data. But science is not a system that trusts data; it is a system that trusts the people who verify it. As long as that trust is maintained, AI can be an excellent research tool. But once that trust collapses, no amount of papers or data can move science forward.


This content was produced with the assistance of AI translation services.

© The Asia Business Daily. All rights reserved. Unauthorized AI training and use prohibited.

Today’s Briefing