Fields Medalist Terence Tao Warns of AI “Draining” Mathematics; Scholars Counter: Humans Haven’t Lost Jobs, They’ve Just Changed Roles
Hello everyone, I’m your financial journalist and economist. Today, we’re not talking about stock market fluctuations or company listings, but about a topic that may seem profound yet actually affects everyone’s future career: What value does humanity retain when AI begins to produce answers at an unprecedented rate, like a money-printing machine?
Recently, Terence Tao, the Fields Medalist and a leading figure in mathematics, posted a series of tweets on social media. He then co-signed a statement with 24 other top mathematicians, warning that AI is over-exploiting mathematical problems and could potentially undermine the foundation of human mathematical research.
The author of this article, Zhao Bin from Fudan University, provided a sharp and insightful rebuttal to Tao’s concerns. He argues that Tao is worried about the “water” in the river (the ability to solve problems) being drained, but what’s truly valuable is not the water itself, but the person holding the faucet—the one who decides which water to draw, for whom to draw it, and for what purpose.
Below, I’ll break down this debate into five key points to help you fully understand the struggle between AI and human intelligence.
---
What Does Tao Tao Fear? – “Junk Answers” Drowning Out “Good Questions”
First, we need to clarify what Tao Tao is actually worried about. Many people think he’s against AI, but that’s not the case. What he opposes is the negative impact of low-quality AI outputs on the scientific research ecosystem.
Tao used a very vivid analogy: think of a food bank. In the past, food banks accepted everything donated, but now it’s different. If a box of expired, moldy food is donated, the bank can’t just distribute it; it has to spend resources sorting, storing, and even discarding it. This “junk” not only doesn’t help but also fills up the storage, creating an illusion of meeting demand and excluding those who really need fresh food.
In the context of mathematics:
1. Too many problems, too few good ones: There are countless mathematical problems, but the truly valuable, knowledge-enhancing “good problems” are as scarce as fresh water.
2. AI is “mining”: Modern AI models, like those from OpenAI, can generate countless mathematical proofs and answers. However, many of these answers are “raw” – they have no context or verifiable process.
3. Ecosystem disruption: If laboratories only publish successful AI cases and not the failures, researchers will focus on the easy ones, neglecting the truly challenging and valuable problems.
In simple terms: Tao Tao is concerned that AI, like an endless printer, produces a pile of incomprehensible “books” that fill libraries, leaving scholars with no time for the truly meaningful research.
---
The Author’s Rebuttal: You’re Using the Wrong Measure – The “Difficulty Map” Doesn’t Apply to AI
The author believes that while Tao’s concerns are valid, he’s making a fundamental mistake: **he’s using the standards of the “human era” to measure the changes of the “AI era”.
Tao Tao relies on two “old measures”:
1. The difficulty map: This shows where problems are difficult for humans to solve. For example, a proof that takes three years to solve is considered valuable.
- Author’s critique: This is meaningless to AI. AI doesn’t get tired or slowed down by difficulty; it doesn’t care how hard a problem is for humans.
2. The criteria for good problems: In the human mathematical community, what counts as a good problem is often determined by small circles of experts, traditions, and power.
- Author’s critique: This is like in Go. Humans once thought certain moves were “orthodox,” but AlphaGo Zero ignored all human strategies and developed better ones. This shows that human-established “correct moves” might just be local optimizations. If mathematicians’ criteria for good problems are confined to their circles, AI breaking these boundaries might not be a bad thing; it could lead to new discoveries.
In simple terms: Tao Tao thinks AI flattens the difficulty levels of mathematics, but the author argues that difficulty is a product of human limitations, not the essence of mathematics. AI doesn’t care about difficulty; it only cares about logic and results. Therefore, using human-defined difficulty as a measure of value is outdated in the AI era.
---
The Core Dispute: Is What Remains of Human Ability Still “Exploration” or “Curation”?
This is the most interesting and profound part of the article. The authors and Tao Tao fundamentally disagree on whether the remaining human ability can still be called “exploration”:
- Tao Tao’s view: Yes. He believes that pure exploration for the sake of mathematics is sacred and at the core of human intelligence. Even though AI solves problems, humans’ ability to identify good problems remains valuable.
- The author’s view: No. It’s more like “curation”.
The author introduces the concept of interest alignment:
- AI has no interests: AI doesn’t have its own goals; it just follows instructions.
- Humans do: Humans research for purposes—understanding the universe, developing cryptography, winning Nobel Prizes, or out of pure curiosity. The purpose always comes from humans.
Therefore, while AI handles “finding directions” and solving problems, humans must retain the power to determine the purpose of the information generated.
In simple terms: In the past, humans had to find directions, solve problems, and judge their value. Now, AI generates a lot of data, and humans need to filter out what’s useful, aesthetically pleasing, or practically valuable.
The author argues: Humans used to be like miners, digging, refining, and judging steel. Now, AI is like a giant excavator that clears the land, and humans are like jewel appraisers or curators. We don’t need to dig; we need to identify what’s valuable and decide its use. Calling this “exploration” is an overstatement; it’s more like “curation.” But this doesn’t mean it’s less valuable; on the contrary, with AI’s endless options, choice has become more important.
---
A Specific Problem: Laboratories Only Show Successes
Although the author rebuts Tao Tao on a philosophical level, he acknowledges a valid and urgent warning from a practical research perspective:
Tao Tao points out that AI laboratories only publish successful cases, not failures. This creates a problem: we can’t see the true limits of AI’s capabilities.
- What is the “frontline”? In mathematics, the frontline is the boundary between what AI can and can’t solve.
- Why is it unclear? If labs only announce successes, researchers can’t assess AI’s true capabilities.
- Consequences: They might try problems that AI can’t solve or overlook problems that AI can solve but are unknown. This is like in warfare: if you only see successful attacks, you can’t assess the enemy’s strength.
In simple terms: This is a problem of information asymmetry. AI companies only show their successes, not their failures, leading to a distorted understanding of AI’s capabilities. The author believes this is the most actionable aspect of Tao’s warning and can be easily addressed.
---
Conclusion: Don’t Worry About the River Drying Up; Worry About Who Holds the Faucet
The author summarizes the article with a brilliant metaphor:
- Tao Tao worries about the river: He’s afraid AI will drain the good problems and solutions.
- The author sees the faucet: AI can generate endless answers, just as water can be pumped out or synthesized. Water is no longer scarce.
- The faucet (the power to choose): What’s truly valuable is the person who decides which answers to use, for whom, and for what purpose.
Key Insights:
AI handles the production of information, while humans control its purpose and meaning.
Implications for everyone:
1. Don’t worry about being replaced: If your job involves execution, computation, or repetitive tasks, you might be replaced by AI.
2. Develop curation skills: The future’s core competitiveness lies in your ability to identify valuable information from AI and determine its purpose.
3. Interest alignment is key: No matter technology develops, the purpose and audience of your work remain in human hands. AI is a tool; humans are the masters. As long as you can define meaning, you’re not unemployed; you’re just shifting from a “bricklayer” to a “builder”.
In one sentence: Tao Tao fears the river drying up, but I see that the person holding the faucet is not the river itself. When AI handles production, the remaining human role is more like “curation” than traditional exploration. This doesn’t mean it’s less valuable; on the contrary, it’s more crucial.