正文 · AI 翻译
On November 4, Beijing time, Fields Medal winner and renowned mathematician Terence Tao recently warned about humanity’s growing reliance on artificial intelligence. He worries that as people increasingly hand over the processes of thinking and solving problems to AI, humans are losing their understanding of the problems themselves and their solutions, a trend that could soon have serious consequences in the real world.
Tao made these remarks in a speech at the Science x AI Summit held at the University of California, Los Angeles, on October 16. The Science and AI Research Foundation (SAIR) recently posted a video of the full talk, titled “Math 2.0,” on YouTube, and clips from it subsequently drew attention on the overseas social media platform X.
“We are outsourcing our thinking to AI and having it solve problems for us. But at the same time, we are losing a safeguard — we cannot be sure whether the solutions AI gives us are actually what we want,” Tao said. “I am really very worried that we will soon see the consequences of this in the real world. This is actually a concern I have outside mathematics.”
Tao pointed out that advances in artificial intelligence in mathematics are changing the traditional way humans solve problems. In the past, mathematicians tackling complex problems had to spend a long time studying and understanding relevant theories, accumulating new knowledge in the process. Now, however, AI can provide answers quickly, while humans may not understand the reasoning behind them.
He believes the risks of this change are relatively manageable in pure mathematics because mathematical proofs can usually be checked using formal tools. But when the same approach is applied to the real world, matters become much more complicated.
Tao specifically mentioned fields such as power grids, air traffic control, drug development, and chip design. He worries that in the future, people may hand complex problems directly to AI without sufficiently understanding the underlying models and real-world constraints.
In such cases, a solution produced by AI might be entirely valid within a mathematical model yet fail to meet real-world needs. AI might even find loopholes or boundary conditions in the model, satisfying superficial metrics while straying from the goals humans originally wanted to achieve.
Tao emphasized that the problem lies not with AI technology itself but with how humans use it. He advocates moving mathematical research into a so-called “Math 2.0” era, in which AI does more than help humans solve problems: it also deepens their understanding of those problems and helps researchers identify the limitations and potential risks of models.
On October 22, Tao also published an article in Nature calling on the mathematics community to reassess its research goals in the AI era. He noted that as AI generates mathematical proofs faster than humans can understand and absorb the results, mathematical research faces new challenges.
Tao believes AI tools could indeed strengthen human understanding, but this would require changes in research methods and incentives. What truly needs to be preserved in the AI era is not only the ability to obtain correct answers, but also the human ability to understand, verify, and judge them.