(Photo: Bozhin Karaivanov / Unsplash) Lying smouldering beneath the kindling of AI fear lies this basic question – will this thing one day displace us completely? Will it be able to scale the heights of human achievement? Gauss? Newton? Einstein? Mozart, Miles, Shakespeare or Caravaggio?
Perhaps the opinion of someone who embodies that sort of creative genius might provide a compass.
In the high priesthood of mathematics, Terence Tao occupies that rare space. Born in Adelaide, he was doing university-level mathematics as a child, had his bachelor’s and master’s degrees by 16, a Princeton PhD by 20, and was a full professor at UCLA by 24 – still the youngest in that university’s history.
In 2006, he won the Fields Medal, mathematics’ highest honour, for work spanning partial differential equations, combinatorics, harmonic analysis and additive number theory.
His peer group of Fields medallists and tenured professors at the world’s elite institutions treat him not merely as an equal, but as a benchmark. When Tao says something about mathematics, the field listens. Which is why, in recent months, what he has had to say about artificial intelligence has reverberated well beyond the seminar room.
For most of AI’s history, the answer would have been that AI would never be admitted to the same club that has Tao and his ilk as members. Machines have been trying to do mathematics almost since AI was born.
As far back as 1956, Newell, Shaw and Simon’s Logic Theorist proved dozens of theorems from Principia Mathematica (the three-volume work by Alfred North Whitehead and Bertrand Russell that attempted to derive all mathematical truths from purely logical axioms). It was impressive, but also telling. But all it could really do was replay and optimise known logic. It was not inventing modern mathematics. It was just carefully stepping through theorems and cross-checking them against the known canon.
Chatbots: ‘Like a mediocre, but not
completely incompetent grad student’
But advanced research mathematics is different. As Tao has pointed out, sometimes it involves not knowing what you are looking for, or whether what you are looking for exists. As late as 2024, Tao himself described chatbots as feeling like advising “a mediocre, but not completely incompetent, graduate student”. Useful for drafting code or checking a calculation, but not remotely ready to sit at the frontier.
I know a couple of professional mathematicians. They do indeed periodically frolic in a qualitatively different sector of the universe from the rest of us. I once asked someone to explain one of mathematics’ great unsolved problems – the Riemann Hypothesis. The statement of the problem is short, but don’t let that deceive you. I was lost within seconds, notwithstanding his brave attempts to break it down for me. This was the sort of fortress that AI was never expected to invade.
And so into this story drifts the remarkable ghost of Paul Erdős. Born in Budapest in 1913, Erdős was the most prolific mathematician of the 20th century, co-authoring papers with some 500 collaborators across six decades. He owned almost nothing, lived out of a single suitcase, and moved between universities like a mathematical mendicant, knocking on colleagues’ doors and announcing: “My brain is open.” He died in 1996 at a conference in Warsaw, mid-problem. He is survived by his theorems – and by his questions.
Erdős left behind more than 1,000 unsolved conjectures, many of them deceptively simple to state and ferociously hard to prove. Some of the most famous have been solved in the decades since his death. But hundreds of Erdős’ problems remain open. They have become, in the community, something of a holy grail – a curated list of genuine research-level problems, accessible in formulation to any educated specialist or non-specialist, requiring original insight rather than mere computation to crack.
It is against this backdrop – the Erdős problems as a litmus test for mathematical intelligence – that AI has recently staged something extraordinary.
On 24 April, Scientific American reported the story of Liam Price, a 23-year-old with no advanced mathematical training, who gave an Erdős problem to GPT-5.4 Pro. The model produced a solution to a 60-year-old problem concerning “primitive sets” – collections of whole numbers in which no member divides another (yeah, I’ll stop there).
This was not merely a case of an AI rediscovering something on page 47 of an obscure journal. Tao’s verdict was more interesting. It arrived at the end of what he has called a genuine inflection point.
“2025 was the year when AI really started being useful for many different tasks,” he told Quanta Magazine. The shift has been cumulative and, for Tao, personal. He now uses AI to search literature, write and debug code, make plots, run calculations and test whether a line of attack is worth pursuing.
‘Ready for prime time’
He reaches for it, in his own phrase, to try “crazier things” – ideas that would have cost days to explore manually can now be stress-tested in hours. In March 2026, at a conference at UCLA’s Institute for Pure and Applied Mathematics titled “Accelerating Math and Theoretical Physics with AI,” Tao declared that current models are now “ready for prime time,” because in his field, AI “saves more time than it wastes.” This is from a man who, less than two years earlier, was comparing the technology to a middling postgraduate.
That is the important bit. Not “AI replaces mathematicians”. Not yet anyway, and perhaps not in the way people imagine. Rather, AI can now sometimes produce a genuinely new mathematical manoeuvre – perhaps messy, badly expressed, half-buried in sludge – which a human expert can recognise as gold.
Tao described the new human-machine workflow with a characteristically unfussy image – one person has a shovel, another has a pickaxe; together they can bore a tunnel. He also suggested that mathematics may shift from solving one problem at a time to exploring thousands, even statistically studying whole landscapes of problems.
“Mathematics is entering an era of scale,” he has said, “where you can actually crowdsource big research projects that you would not think of doing if it were just you and your graduate student.”
He imagines new kinds of mathematicians – people skilled not at solitary proof-craft but at orchestrating large collaborative efforts between humans and machines. The field, he notes with evident satisfaction, has always been “a bunch of cats” – famously individualistic, resistant to coordination.
AI may be what finally changes that.
The deeper irony is that the figure adjudicating AI’s entry into mathematics – the man whose stamp of approval the field waits for – is himself an almost supernatural example of what unaided human mathematical intelligence can accomplish. Tao does not need AI. He chose, gradually and with rigour, to embrace it anyway.
That choice, made by the best mind in the room, may say more about where mathematics is headed than any individual proof, however clever. While it does not directly answer our question about our collective displacement in the future, it quietly makes its point. DM
Steven Boykey Sidley is a professor of practice at JBS, University of Johannesburg, a partner at Bridge Capital and a columnist-at-large at Daily Maverick. His new book, It’s Mine: How the Crypto Industry is Redefining Ownership, is published by Maverick451 in South Africa and Legend Times Group in the UK/EU, available now.