AI mathematics is ushering in the same 'slap-in-the-face moment' that the Go world experienced a decade ago.
Holding a similar view is none other than Demis Hassabis, co-founder of DeepMind, who spearheaded the AlphaGo project a decade ago.
Although Hassabis has recently undergone a series of identity changes—from DeepMind CEO to Google Chief Scientist and Chairman of DeepMind—
when it comes to how AI is transforming human cognition, he remains one of the most authoritative voices.
Ten years ago, he witnessed AlphaGo defeat legendary Go player Lee Sedol with the astonishing move 37.
Ten years later, standing at the forefront of AI's incursion into the frontiers of mathematics and science, he has offered a new judgment:
If AI can solve a Millennium Prize Problem, that would be a breakthrough of 'move 37' magnitude. As it stands, it seems only a matter of time before AI reaches that level; I see no reason why it wouldn't.
Move 37, a moment representing the first time humanity was slapped in the face by AI, has thus been etched into memory.
But few remember that move 37 was not the only answer left by that man-machine battle.
Two games later, Lee Sedol, who had already lost three in a row, played move 78.
This move also had only a one-in-ten-thousand probability of being played by a human, yet it successfully disrupted AlphaGo and helped Lee Sedol secure humanity's only victory in the five-game match.
Looking back at this man-machine battle a decade later, one cannot help but marvel:
The relationship between humans and AI is just like these two numbers.
One made humanity re-understand AI, the other made humanity re-understand itself.
The move 37 that humiliated humanity
Time back to March 2016, the five-game match between AlphaGo and Lee Sedol was unfolding in Seoul, South Korea.
In the first game, Lee Sedol lost to AlphaGo, a bad start.
In the second game, the whole world was waiting for this legendary player to counterattack.
Mid-game, Lee Sedol briefly left the table, smoked a cigarette to calm his nerves, but when he returned to the board, a black stone had already been placed.
Move 37.
Lee Sedol did not sit down immediately; he stood still staring at the board, almost unable to believe his eyes.
The reason was simple: this move was too unlike a human move.
AlphaGo placed the stone on the fifth line, deviating from the positional intuition that professional players had followed for centuries. In conventional human Go understanding, stones should occupy high positions to control the board, but AlphaGo placed it in an area where no direct battle had yet occurred.
The on-site commentators initially even suspected that AlphaGo had made a mistake.
Professional 9-dan player Michael Redmond, who was commentating, was unable to judge for a long time, only saying:
I don't know if it's a good move or a bad move; it's very strange.
His co-commentator was more direct:
I thought it was a mistake.
The DeepMind team later calculated that the probability of a human player choosing this move was only one in ten thousand.
But as the game progressed, people gradually realized that this was not a bad move; it was a brilliant move that changed the course of the battle on a larger scale.
About 100 moves later, this seemingly misplaced black stone turned out to be the key to AlphaGo winning the second game.
Later, this moment of collective human misjudgment came to be jokingly called the 'slap-in-the-face moment'—
Originally thought AI was wrong, but in the end, it was humans who were wrong.
But beyond the slap in the face, this move also made the world realize for the first time that AI not only imitates human play but can also make moves that humans have never thought of.
And a decade later, 'move 37' has indeed reappeared, and this time it is associated with an ancient discipline:
Mathematics.
For AI, Go and mathematics share a major commonality: verifiability.
Because the logic and rules in these fields are sufficiently clear, AI can repeatedly execute a simple process:
Try an idea, test it, learn from the results, and keep trying until success (that reinforcement learning stuff).
This mechanism also explains why mathematics has become the first discipline to be intensively bombarded by AI's 'move 37'.
In fields such as drug development and biology, whether an idea is valid often requires lengthy experiments to verify.
Mathematics is different; the proof process can be checked step by step. Right is right, wrong is wrong, so AI can search tirelessly.
Thus, as AI capabilities continue to strengthen, changes have begun to accelerate on a yearly basis.
Starting with the birth of ChatGPT, in 2023 large models could still be tripped up by basic arithmetic problems; the incident of "which is bigger, 9.11 or 9.9" was once laughable.
But by 2024, DeepMind's AlphaProof and AlphaGeometry 2 had already entered the International Mathematical Olympiad.
The two systems solved 4 out of 6 problems, scoring 28 points, reaching the silver medal level (just 1 point short of that year's gold medal threshold).
Just one year later, Gemini Deep Think solved 5 problems within the official 4.5-hour limit, scoring 35 points, reaching the IMO gold medal level.
In two years, AI went from a novice to an IMO gold medalist, at an astonishing speed.
But this is still just the examination hall.
What truly made the mathematics community feel the tide coming is that AI began to move from "solving problems" to "conducting research".
In 2025, GPT-5 participated in solving Erdős problem #848, proposing the key estimate needed for the proof.
Mathematicians corrected and tightened it, finally completing the full proof.
In the same year, UCLA mathematician Ernest Ryu, with the help of GPT-5, found a breakthrough for an open problem that had troubled optimization theory for 40 years.
At this stage, AI is no longer just computing answers; it begins to contribute key ideas.
Then, time came to 2026, and a series of major mathematical problems began to achieve breakthroughs.
The first to drop the bombshell was still OpenAI.
An internal general reasoning model targeted the planar unit distance problem. This problem was proposed by mathematician Paul Erdős in 1946, studying how many pairs of points at distance exactly 1 can be maximized when placing n points in the plane.
For nearly 80 years, the mathematical community generally believed that constructions like square grids were close to optimal.
But AI directly brought in a set of tools from seemingly unrelated algebraic number theory, constructing a completely new set of points, overturning this long-standing conjecture.
The entire proof was subsequently verified by external mathematicians.
OpenAI stated that this is the first time a general AI has autonomously solved an open problem of significant importance in a branch of mathematics.
Immediately after, Anthropic also made a move.
In July 2026, Anthropic mathematician Levent Alpöge, with the help of Claude Fable 5, found an explicit counterexample to the Jacobian conjecture.
A conjecture that had troubled the mathematical community for 87 years was thus disproven by AI overnight.
There were even joint efforts by the two companies (OpenAI and Anthropic).
Just a couple of days ago, GPT-5.6 and Fable 5 teamed up to solve a mathematical problem that had been open for 25 years.
(There are too many similar stories...)
But the most exaggerated is Google; DeepMind is turning this breakthrough into an assembly line.
Its mathematical agent Aletheia no longer just generates an answer once, but repeatedly proposes proofs, checks for loopholes, overturns itself, and then searches for new paths.
DeepMind used it to scan 700 open problems in the Erdős conjecture database.
Among them, AI independently solved 4 unsolved problems and also contributed to multiple mathematical results of paper quality.
This attempt shows that AI has begun to search the blank areas of the mathematical world on a large scale.
So it is not hard to imagine:
It is only a matter of time before AI solves more major mathematical problems.
From solving problems to conducting research, from reciting known knowledge to proposing unknown answers, the 37th move in the field of mathematics is being played one after another.
Finding Humanity's 78th Move
And in the face of this moment that is likely to arrive soon, humanity will undoubtedly face a soul-searching question:
If AI continues to make new moves that humans have never thought of, what should humans do?
The 78th move that appeared in that match ten years ago might be the answer.
In the fourth game of the five-game match, Lee Sedol, who had already lost three consecutive games, wedged a white stone into the center of AlphaGo's formation.
This move also deviated from the norm, and AlphaGo estimated the probability of a human playing it to be only one in ten thousand.
Subsequently, the machine lost its composure.
Lee Sedol ultimately secured humanity's only victory, and the 78th move has since been called the "Move from God."
This victory did not mean that humans had defeated AI again.
But it at least showed that humans can, after understanding the machine, find the next move that the machine did not anticipate.
This is precisely the lesson that the 78th move leaves for today.
As AI becomes increasingly adept at providing answers, human value will shift more toward asking questions, determining direction, and deciding which answers are truly worth seeking.
Borrowing the concept of "centaur chess" from chess, it means having human players and computer engines work together.
p.s. After IBM's "Deep Blue" defeated world champion Garry Kasparov in 1997, Kasparov advocated and hosted the first "Centaur Chess Tournament" in 1998, exploring the possibilities of human-AI collaboration.
Hassabis believes that science is entering a similar "centaur era":
We are entering an era of human-computer collaboration, and I don't know how long this cycle will last. But for very complex fields, it could be a long time. For example, drug discovery, biology, chemistry, etc. These fields are highly complex, with changing conditions, and not everything can be verified. This is where human intuition and insight are needed to decide the direction forward.
The subsequent evolution of Go has already proven this point.
After AlphaGo appeared, AI quickly became a review tool for professional players. Players could use AI to re-examine past joseki, discover which experiences were actually unreliable, and also study new moves that were once considered unreasonable.
A study analyzing over 5.8 million moves by professional players found:
After the emergence of super AI, the quality of human players' moves improved significantly, and novel moves also increased.
However, there is another side to the matter, which is:
People are beginning to rely on AI for answers, gradually losing the ability to judge answers.
Just as Lee Sedol said upon retirement, facing an opponent he could not defeat, he found it difficult to continue enjoying the Go he once loved.
This might be the most concerning issue in the AI era.
As more fields beyond mathematics are influenced by AI, will more powerful systems make us more creative, more efficient, and more human?
Or will they cause people to give up trying?
In the future, AI will continue to play the 37th move.
The remaining question is whether humans can find their own 78th move.
Reference links:
[1]https://www.wsj.com/tech/ai/move-37-ai-demis-hassabis-google-deepmind-alphago-ec832a41?st=9XMAjn&reflink=desktopwebshare_permalink
[2]https://x.com/demishassabis/status/2085914742414061886
This article is from the WeChat public account "Quantum Bit", author: Focus on Frontier Technology
