The $2,000 Proof…

…What Happens To Human Genius When The Machine Gets There First, And Why Nigeria Should Be Paying Very Close Attention

By Folu Adebayo

“A mathematical question had stood unanswered since 1999. Some of the finest minds of a generation had circled it and failed. On the first day of this month, a machine answered it , along with nine other problems that had defeated human experts for at least a decade each. The total cost of the computing power was about two thousand dollars. Less than a decent laptop. That number should reorganise how every Nigerian business leader, parent and educator thinks about what human talent is actually for.”

Let me tell you what happened on the first day of August and then let me tell you why I think it matters more to Lagos, Abuja and Port Harcourt than almost anything else that occurred in technology this year.

OpenAI announced that an internal, still-unreleased version of its next major model,  a system it calls Astra had produced solutions to ten longstanding open problems in mathematics and theoretical computer science. Not exam questions. Not textbook exercises with answers in the back. Genuine open problems, each of which had resisted human solution for at least ten years.

The headline result was the first explicit construction of what mathematicians call a non-sofic group. If that phrase means nothing to you, it means nothing to most people. What matters is this: the question had been open since 1999, when the concept was introduced by one of the most celebrated mathematicians alive. Twenty-seven years of some of the best-trained minds in the world, and no answer.

The machine answered it. And then it did nine more.

Why This One Is Different

I want to be careful here, because I have written in this column before about the difference between genuine technological progress and vendor theatre. The technology industry has an unfortunate habit of announcing miracles that dissolve on inspection. Only last October, the same company made claims about mathematical results that were publicly dismantled by researchers who checked the work.

So why do I treat this differently?

Because of how it was published. Alongside the announcement, the company released a 249-page technical manuscript and critically machine-checkable proof certificates for every single result, posted publicly under an open licence.

Let me explain why that detail changes everything, because I think it is the most underappreciated part of this story.

Mathematical proofs have traditionally been verified by other mathematicians. A person writes a proof, submits it to a journal, and specialists spend weeks or months reading it line by line before it is accepted. That process is slow, human, and built entirely on trust and expertise.

A machine-checkable certificate works differently. The proof is written in a formal language whose compiler will simply refuse to accept a step that does not logically follow. Run the file. The computer either accepts every step or it does not. There is no interpretation, no reputation, no benefit of the doubt. Any person with a laptop can now verify a result that would previously have required a doctorate and a peer review committee.

And the files were published with no gaps left unproven. Every step, in all ten proofs, verified.

“Any person with a laptop can now verify a result that would previously have required a doctorate and a peer review committee.”

So we are not being asked to believe a press release. We are being invited to check. That is an enormous difference, and it is the reason serious mathematicians, including a Fields Medallist, the closest thing mathematics has to a Nobel laureate, have responded to this work with cautious respect rather than dismissal.

The Caveats An Honest Columnist Must State

Three of them, and I will not bury them.

First, this is a vendor announcement about a model the public cannot yet use. We are told what it did. We cannot yet examine how reliably it does it, or how many attempts failed for every success. The two-thousand-dollar figure covers the successful runs not the wasted ones.

Second, none of the results has yet completed formal journal peer review. The proofs have been machine-verified and informally examined by experts, which is meaningful, but it is not the same as publication.

Third, a compiler can confirm that an argument is internally sound. It cannot confirm that the question was framed to mean what mathematicians actually intended it to mean. That judgement remains stubbornly human.

Those caveats matter. They do not, in my assessment, undo the significance of what happened.

The Quiet Revolt

There is a second story running underneath this one, and it is the more human of the two.

In June, the international mathematics community issued a formal declaration endorsed by its global governing body objecting to how artificial intelligence companies are operating in their field. The complaints were specific: that these companies build on published research without consent, that they bypass peer review, and that they are eroding established standards around proof and credit.

And there is the unresolved question of authorship. The company has said it takes responsibility for the written manuscripts, while attributing the mathematical reasoning itself to the model.

Sit with that sentence for a moment. Who solved the problem?

Not a person. Not exactly a company. A model, which cannot hold a grant, receive a prize, be appointed to a professorship, or be held accountable for an error.

This is not a technicality for academics to squabble over. Every system of human recognition we have built, degrees, professional certifications, promotions, prizes, tenure, seniority  rests on a single unspoken assumption:

That difficulty is evidence of ability.

We reward people for solving hard problems because, historically, only capable people could. Difficulty was our proxy for merit. It was a good proxy for a very long time.

That proxy is now failing in front of us.

“Every system of human recognition we have built rests on one unspoken assumption: that difficulty is evidence of ability. That proxy is now failing in front of us.”

Why This Lands Hardest In Nigeria

Now let me bring this home, because I believe this story has sharper implications for Nigeria than for almost any country I could name.

Nigeria is a credentialist society. We are, and we have reason to be. For generations, education was the most reliable ladder out of constrained circumstances, and our families made extraordinary sacrifices to climb it. We revere the certificate. We revere the professional title. We test our children relentlessly on recall and on their capacity to reproduce difficult work under pressure.

That system was rational when difficulty was scarce and expensive. It is becoming irrational as difficulty becomes cheap.

Consider what a Nigerian graduate is currently being trained to be excellent at absorbing a body of established knowledge and reproducing it accurately on demand. Consider what a two-thousand-dollar computing run can now do to a problem that defeated the world for twenty-seven years.

We are training our brightest young people to compete in precisely the category where machines are improving fastest, and neglecting the categories where humans remain irreplaceable.

I say this as someone who has spent over twenty years in enterprise technology, and who has watched extraordinary Nigerian talent enter global firms-talent that consistently outperforms once it is inside the room. The ability is not in question. The direction of training is.

The Skill That Survives

Here is the part of this story I find genuinely hopeful, and it is the reason I chose to write about it this month rather than something more alarming.

If a well-defined problem can be handed to a machine and verified in minutes, then the bottleneck in human work moves. It moves away from solving and towards three things machines cannot yet do:

1. Choosing the right problem. Of the infinite set of questions that could be asked, which ones actually matter? Which are worth the compute, the capital and the attention? That is judgement, and judgement is built from experience, context and knowing your own society.
2. Framing it correctly. A machine will faithfully answer the question you actually asked, which is frequently not the question you meant. Translating a messy human problem into a precise, answerable form is a genuine skill and a scarce one.
3. Judging whether the answer means anything. A technically correct result that solves the wrong problem is worse than useless, because it is persuasive. Someone must decide whether an answer is true, useful and safe to act upon.

Notice something about all three. They are not about processing power. They are about proximity to a real problem and the wisdom to interpret it.

And on that measure, Nigeria is not disadvantaged. It is unusually well positioned.

We are surrounded by unsolved problems of enormous consequence — in power generation, in agricultural yield, in health delivery, in credit access, in logistics across difficult terrain, in financial inclusion for the unbanked. These are hard, specific, poorly formalised problems that global technology companies do not understand and are not incentivised to prioritise.

If the capacity to solve well-posed problems is becoming abundant and cheap, then the ability to correctly pose the problems that matter to four hundred million West Africans becomes extraordinarily valuable.

That is not a consolation prize. That is the premium skill of the next decade, and it cannot be outsourced to a data centre in California, because the people running it do not know which questions to ask.

What I Would Do This Month

For business leaders: examine what you are actually testing for when you hire. If your recruitment process rewards the reproduction of established knowledge, you are selecting for the skill that is depreciating fastest. Start testing for problem framing. Give candidates a messy, ambiguous situation and see whether they can find the real question inside it.

For educators and parents: keep the rigour. Rigour builds the judgement that everything else depends on, and I am not arguing that our children should learn less mathematics. I am arguing they should also be asked, far more often than they currently are, a question our education system almost never poses: which problem here is worth solving, and why?

For policymakers: the strategic asset is no longer the ability to train people who can compute. It is the ability to identify, formalise and prioritise the problems that matter to Nigeria — and to keep the people with that judgement in the country long enough to apply it.

For everyone: understand that the machine has not replaced human intelligence. It has repriced it. The parts of thinking that were expensive are becoming cheap, and the parts we treated as soft and unmeasurable judgement, framing, discernment, knowing what matters are becoming the scarce and valuable ones.

The Question Behind the Question

A problem stood unsolved for twenty-seven years. A machine solved it for the price of a laptop and published a proof that any of us can verify.

The instinctive human response to that news is anxiety. I understand it. I felt it.

But I think the anxiety asks the wrong question. It asks: what is left for us to do?

The better question is the one the mathematicians themselves have arrived at. When solving becomes cheap, the value moves to asking. And nobody has yet built a machine that knows which of humanity’s problems deserve to be solved first.

That decision is still ours. It has always been the most consequential one we make.

The only thing that has changed is that we can no longer hide behind the difficulty of the answer.

Folu is a Solution Architect and AI Governance Director with over 20 years of experience across  financial services, insurance and enterprise technology in the UK and internationally.