I have spent much of my working life around mathematics.
I trained as a mathematician and computer scientist and completed a PhD in mathematics. Since then, I have worked in investment and retail banking, funds management, insurance, federal and state government, consulting, startups, and academia. In each setting, I have seen that mathematical training offers much more than the ability to calculate an answer.
It teaches people to define a problem, make assumptions explicit, distinguish evidence from intuition and recognise when a result cannot be right. Those habits become more valuable, not less, when people work with complex systems and imperfect information.
That is why I paid attention when Jensen Huang, the chief executive of Nvidia, was asked whether it mattered if children forgot basic mathematics because of AI.
His answer was blunt: not really. In a recent conversation with New York Times columnist Ezra Klein, Huang acknowledged that AI could cause students to lose basic mathematics skills. But he questioned whether those skills would remain important when machines could perform them instantly.
There is a reasonable point here — education should change as technology changes. Few people now memorise telephone numbers, and most professionals rely on software for calculations that were once performed manually.
But Huang’s conclusion is too casual. Basic mathematics is not simply a collection of procedures that machines can take over. It provides a way of thinking. It gives people enough fluency to estimate, test assumptions and recognise when an apparently authoritative answer is nonsense. That distinction matters because AI systems are not consistently reliable. The 2026 Stanford AI Index describes a “jagged frontier” in which AI systems can perform impressively on difficult tasks while failing at apparently simple ones.
The more knowledge people outsource, the more important it becomes to retain enough understanding to evaluate the systems doing the work.
The timing of Huang’s remarks is also significant. Nvidia is one of the companies whose chips and systems underpin much of the current AI industry.
And the technology industry is now beginning to build educational institutions of its own. Andreessen Horowitz has launched the Horowitz Andreessen Academy, a private, full-time school in San Francisco for young people coming out of high school. Its founding partners include Anduril, Anthropic, Coinbase, Google, Meta, Nvidia, OpenAI, Palantir, Replit and Stripe.
The academy is not currently a university in the conventional sense. It is being presented as a college alternative, with a focus on project-based learning, company-building and direct pathways into the technology industry.
The academy may offer valuable experimentation, as many students would benefit from more practical experience, stronger industry connections and opportunities to work on real problems.
But it also raises a question Australia should take seriously: What happens when the companies building AI begin designing the institutions that train the people who will work with it?
This is not simply a debate about whether children should learn long division. It is a debate about who decides which forms of knowledge matter, which capabilities can be outsourced and what education is ultimately for.
A corporate education venture will inevitably reflect the priorities of the companies and investors behind it. It is likely to value speed, experimentation, entrepreneurship and commercial opportunity. Those qualities are useful, but they are not the whole of education.
Universities also preserve knowledge whose value is not immediately commercial. They investigate questions that may not yet have a market. They teach students to challenge powerful institutions, understand social context and think about consequences that cannot be captured in a business plan.
A company may want graduates who can build the next product. A democratic society also needs people who can ask whether that product should exist, who may be harmed by it, how it should be regulated and who should be accountable when it fails.
Those roles are not interchangeable.
There is already increasing evidence that using AI to complete academic work can create a gap between apparent performance and genuine learning. The important distinction is whether it supports thinking or replaces it.
In mathematics, this distinction is familiar. A proof is not valuable merely because it reaches the correct conclusion. The reasoning, assumptions, and route from one step to the next matters. Without them, it is often impossible to determine whether the conclusion is sound.
The same issue is emerging at the highest levels of mathematics. In his essay Mathematics in the Age of AI, mathematician Terence Tao considers what happens if AI tools become capable of carrying out significant research-level mathematical tasks.
His focus is not simply on whether machines can solve problems. It is on what mathematics should value when they can: the choice of worthwhile questions, the verification of results, the explanation of ideas and the development of knowledge that other mathematicians can understand and use.
That is a useful distinction. The purpose of mathematics — and education more broadly — is not exhausted by producing answers.
A system may produce a fluent answer, sophisticated model or impressive prediction. That does not remove the need for people who understand how to interrogate it.
This is particularly important for Australia. The Australian Government’s National AI Plan identifies domestic AI capability, infrastructure, workforce training, responsible adoption and regulation as connected national priorities.
Australia’s challenge is not merely to produce more people who can use AI. It is to develop people who can build, evaluate, govern and, when necessary, reject AI systems.
That requires more than coding skills. It requires mathematics, statistics, engineering, computer science, law, public policy, ethics and the social sciences. It requires researchers who can understand both the technical operation of AI systems and their institutional consequences.
Australia is already developing stronger links between universities and industry. The National Industry PhD Program supports doctoral projects designed with industry participation and aims to strengthen the movement of researchers between academia and industry.
That is a constructive direction. But industry-linked education should complement independent university research, not replace it. Companies can help identify practical problems and provide valuable experience. Universities must retain the independence to investigate questions whose value is not yet obvious to employers or investors.
If universities weaken while private companies expand into education, Australia risks becoming dependent not only on overseas AI models and computing infrastructure, but also on overseas institutions to educate the people capable of understanding them.
That would be a serious loss of national capability.
The decision about what children and students can afford to forget should not be left solely to the companies that profit from making those capabilities unnecessary.
AI may make some forms of manual calculation less important, but that does not mean the underlying knowledge is dispensable.
A valuable lesson that my mathematics training taught me is that there is a difference between not needing to perform a calculation every day and not needing to understand what the calculation means.
Australia needs universities that are more connected to industry, more responsive to technological change and better at preparing students for an AI-enabled economy. But it also needs universities capable of producing independent knowledge, challenging commercial assumptions and training people to govern powerful technologies.
The question is not whether AI will change education, it is whether Australia will shape that change deliberately — or allow the companies building AI to decide what the next generation needs to know.
Dr Alex Antic is the faculty head of AI Strategy at UNSW Canberra, and deputy director of the UNSW AI Institute
Do you know more? Contact James Riley via Email.