Summary of Key Points
This article discusses the impact of AI on scientific research and knowledge innovation, focusing on the following core arguments:
- AI increases the efficiency of individual scientists (leading to more published papers and citations), but it may narrow the scope of collective research and reduce interaction.
- As AI makes it easier to find answers, the main challenge for humans shifts from acquiring information to asking meaningful questions and redefining problems.
- The essence of scientific breakthroughs is not about filling in existing gaps but about creating new frameworks for understanding the world.
- While AI excels at accelerating processes within existing frameworks, it cannot yet redefine them on its own. The most valuable skill in the future will be the ability to identify and correct flawed problem definitions.
The Double-Edged Sword of AI in Research
A study published in Nature found that scientists using AI produce better individual results (more papers and citations), but the overall scope of scientific research has narrowed by 4.63%, and interaction has decreased by 22%. This is similar to navigation software guiding all cars to the fastest route: while each car travels faster, the entire road becomes congested.
Why does this happen? AI is particularly useful in fields with large amounts of data, well-defined problems, and clear evaluation criteria (e.g., analyzing experimental data or drafting papers). As a result, scientists tend to concentrate on these areas where success is more likely, leaving less exploration of uncharted territories and thus narrowing the collective perspective.
Answers Are Getting Cheaper, but Anxiety Is Rising?
In the past, finding answers was difficult; to understand quantum entanglement, one had to consult experts and read papers. Now, AI can provide explanations in seconds. However, this may lead to increased anxiety because if the question is wrong, the answer is useless. For example, if a company wants to improve user retention, AI might suggest 100 methods, but if the real issue is that users have not developed a habit of using the product, those methods will be ineffective. The challenge has shifted from finding answers to questioning whether the question is even relevant or correctly framed. It’s like trying to go to Beijing by asking how to get to Shanghai—the most advanced AI cannot help with that.
Scientific Breakthroughs Are About Creating New Maps, Not Filling in Gaps
Many people think of knowledge as a large table with empty cells to be filled in. However, scientific revolutions involve creating new frameworks, not just filling in existing ones. Newton unified the motion of planets and the falling of apples with a new theory of mechanics; Einstein combined time and space with relativity. AI currently accelerates existing processes but cannot create new frameworks on its own. For instance, it can solve complex equations but cannot spontaneously introduce concepts like entropy or genes that redefine our understanding of the world.
Can AI Help Us Discover Hidden Similarities Across Disciplines?
Interdisciplinary innovation does not simply combine unrelated fields; it involves identifying underlying structural similarities. For example, electricity and water flow share similar principles (described by concepts like flow rate, resistance, and potential difference). AI can help in this process by quickly analyzing data and combining information, but it still requires human guidance to identify relevant similarities.
The Most Valuable Skill in the Future: Knowing When to Change the Question
When a problem remains unsolvable, true experts do not try harder to find the answer but pause and ask if the question is fundamentally wrong. For example, scientists once believed that light required an “ether” to propagate, but when that theory failed, Einstein redefined the question, leading to the theory of relativity. This ability to reframe problems is the next major challenge for AGI (Artificial General Intelligence). While AI cannot yet do this, future learning will focus on learning to ask the right questions and restructure problems. After all, only those who change direction can discover new possibilities.
In Conclusion
AI makes us work faster, but we must occasionally stop and ask: Are we going in the right direction? The essence of science and innovation is about creating new frameworks, not just fixing old ones. (Note: The data comes from a 2026 Nature study, indicating trends rather than absolute causality. AI’s capabilities are currently limited to well-defined tasks and cannot replace human creativity in redefining knowledge.)