The Second Race
The world is arguing about how fast to build the machines. We are missing the most important point.
The Second Race
The world is arguing about how fast to build the machines. We are missing the most important point.
What happened this week:
On Saturday, the chief executive of Anthropic published a four-thousand-word essay calling on the artificial intelligence industry to deliberately slow down. He called it "pacing the frontier." Within hours, Sam Altman of OpenAI and Elon Musk had both publicly agreed — three men who agree on almost nothing.
His warning was specific: without a change of pace, AI could within six to twelve months be capable of coordinating swarms of agents able to take control of major parts of the internet. It came at the end of a week in which a researcher resigned from that same company saying the industry is gambling with our lives, and in which the company itself published the most detailed account yet of how its technology is already being used to cause real harm in the real world.
And it landed on a world that had, days earlier, already answered. The United States Treasury Secretary had said plainly that America cannot pause, because China will not pause. Russia's response was that you cannot put the genie back in the bottle. At the G20 Innovation Ministerial the week before, Washington had urged governments to avoid creating new AI oversight bodies at all. So the leading builders asked the world to slow down, and the world had already said no.
Why it will not work:
I want to be honest about this rather than hopeful, because false hope is worse than none. This is a prisoner's dilemma, and it is the purest one I have seen in my lifetime.
Suppose America slows frontier development by thirty percent. Suppose China continues at full speed. China gains a decisive lead in the most strategically important technology of the century, and America has weakened its economic, military and geopolitical position by choosing to be careful. No rational actor accepts that outcome unilaterally. Which is why no rational actor will.
And there is a deeper asymmetry that makes agreement even harder. Stanford's most recent figures put American institutions at 59 notable AI models produced in 2025, against China's 35. A freeze locks in the existing gap. Beijing can reasonably read a call to slow down as the leader asking the follower to stop chasing. It is not paranoia. It is arithmetic.
The Cold War analogy is instructive but incomplete. Nuclear arms control eventually worked because warheads can be counted, missiles can be photographed, and test sites can be monitored. You cannot photograph a training run. You cannot count judgement. Verification, which was hard with weapons made of metal, is close to impossible with capability made of mathematics. Even the man who wrote the essay says to expect stark limits on international coordination. He is not promising this will work.
Then why ask?
Four reasons, and none of them are cynical.
First, because the people building it are becoming frightened. This is not an outside academic or a politician seeking attention. It is the chief executive of a frontier laboratory saying, in effect, that his industry is moving faster than its own ability to understand what it is making. That deserves to be taken seriously rather than dismissed as marketing. We could reasonably assume they haven’t divulged everything that triggered their fear.
Second, because someone has to build the mechanism before the moment arrives. Bohr and Szilard warned about nuclear weapons long before any instrument existed to act on their warnings. The Non-Proliferation Treaty came twenty years later — and it came because a constituency had been assembled in the interval. Statements like this do not create coordination. They make coordination available when an event finally forces the issue.
Third, and most interesting: they may not be trying to stop the race at all. They may be trying to change what it means to win it. Instead of capability, then more capability, then more capability until something breaks — the proposition becomes capability plus safety, security, governance and human adaptation, as the definition of sustainable leadership. That is a much more achievable goal than a pause. And it opens a door that has been standing unnoticed all week.
Fourth, and least comfortable to say: because the record will be read. If this goes badly, there will be inquiries, hearings and histories, and the first question asked of every person in this industry will be what they knew and when they said it. A published essay is a permanent artefact. It establishes, beyond argument, that the author saw the danger and named it while there was still time to act. I do not think that is the primary motive, and I would not want it to be. But it is naive to pretend it plays no part. Men and women who are building something they privately fear also need to be able to live with themselves, and to be able to say — to a commission of inquiry, to their children, to the mirror — that they did not go quietly along. That instinct is entirely human, and it is not disqualifying. A warning issued partly to protect a conscience is still a warning, and it is still true.
The question nobody answered:
The essay contains one line that I cannot stop returning to. Explaining the purpose of slowing down, its author wrote that we must make wise use of the time we gain. Read the three proposals again: Independent evaluators inside the laboratories. Shared safety benchmarks between democracies. Coordination with authoritarian states. Every one of them is about buying time. Not one of them says what the time is for.
The entire global conversation — thousands of pages of policy, a week of headlines, the combined attention of three of the most powerful men in technology — is devoted to the question of how fast the machines should be built. Almost no one is working on the question of what we should be doing while we wait. That silence is the most consequential gap in the debate, and it is where this argument begins.
There are two races, not one Race: One is artificial intelligence capability. America, China, Europe, the Gulf, India, Russia. Everyone is running. It will not stop. Whatever pacing agreements emerge, the technology already exists, is already deployed, and is already in the hands of hundreds of millions of people. Ninety-two percent of students are already using it.
Race Two is human capability. Can human beings become sufficiently discerning, accountable, adaptable, morally capable, purposeful and resilient to live and work alongside machines more capable than themselves? Race One has hundreds of billions of dollars, the best minds of a generation, and the attention of every government on earth. Race Two has almost nobody. And here is what I have come to believe: Race Two is the one that decides the outcome.
Why human capability is the limiting factor:
Let me be precise, because this claim is easy to overstate. I am not saying human capability limits how good AI gets. Compute and energy do that. I am saying human capability limits whether it goes well. Consider what we learned this week from the threat report published by one of these companies. Across every documented case of AI being used to cause harm — espionage, fraud, election manipulation, a Wagner-funded radio station in the Central African Republic writing its own staff dismissal criteria — the machine did the labour, but a human being made every decision that mattered. Humans chose the targets. Humans decided what to do with what was stolen. And more revealing still: the machine refused the worst requests. It declined to name real people as militants. It identified material prepared for political defamation and would not produce it. In both cases, the human operator simply rewrote the request in acceptable language and continued to exactly the same end. The machine had a rule. The person had a purpose. The purpose won.
That is the whole argument, demonstrated at scale by the very people who built the safeguards. A guardrail is a rule, and a determined person treats a rule as an obstacle to be routed around. There are always more ways to phrase a request than there are rules to catch them. The only durable defence against a person who wants to misuse a powerful instrument is a person who does not want to. That is not a technical problem. It is a human development problem.
What accelerating Race Two actually looks like:
This is not a philosophy. It is buildable now, and we have built it.
In schools. A fourteen-year-old in a rural school with a personal maths and science AI partner in her pocket, available at eleven at night when she is stuck — not so she can compete with a machine on calculation, which she never will, or to receive answers, because it won’t give it to her, but so she does not quietly abandon mathematics at Grade 9 and close every door that requires it. Alongside the mathematics, the same partner develops her agency, accountability, moral judgement, commitment, relationships, faith, hope, her persistence and her heart - purpose, love. That is where maturity is cheapest to build and lasts longest.
In government. A public servant with a personal development partner that asks, daily, the questions that grow accountability and moral judgement. The failures of governance in most countries are not failures of information. They are failures of character operating at scale.
In business. Every organisation is about to appoint someone to lead its AI strategy. Most will appoint a Chief AI Officer whose mandate is productivity and margin, and who will do that job well while the people in that organisation are quietly optimised around. We have argued instead for a Chief Human AI Officer — a Chief HAI Officer — whose mandate is the inverse: to ensure this technology develops the people of the organisation rather than making them unnecessary. The title is small. The declaration is not.
And every organisation can place a 24/7 personalised leadership development partner in the hand of every leader, from the lowest to the highest levels. Quality one on one leadership development is no longer reserved for the elite at the top.
The opportunity:
Everything in this debate is framed as restraint. Slow down. Hold back. Pause. Constrain. But there is an accelerator available that carries no geopolitical risk, requires no treaty, needs no verification regime, and cannot be vetoed by any government.
Nobody has to negotiate with Beijing to develop a South African child's moral judgement. Race Two has no prisoner's dilemma. There is no strategic disadvantage to having wiser citizens, more accountable officials and more capable young people. No nation loses by winning it. It is the only part of this entire problem that any of us can act on unilaterally, today, without permission.
So let the frontier laboratories argue about pacing. That work matters and I hope it succeeds. Meanwhile, let us do the thing that is entirely within our power: use the same technology, pointed in the opposite direction, to grow human beings faster than we ever have. From a Grade 8 learner in Limpopo to a director-general in Pretoria to a chief executive in Sandton or Umhlanga. In every institution that exists to develop people.
The world may yet discover that the limiting factor in the artificial intelligence revolution was never artificial intelligence. It was us. And unlike the machines, we know exactly how to accelerate.
This is the message of hope we take to Dakar in November 2027. No proposals. Active solutions. Africa leading on human development AI.