For the past two years the Australian conversation about artificial intelligence has fixated on the hardware and the headline numbers: the billion-dollar data centres rising on the fringes of Sydney and Melbourne, the sovereign compute deals, the venture rounds and the eye-watering valuations. A new analysis from the Australian Strategic Policy Institute wants to shift the frame. Its argument, in short, is that the country most likely to prosper as AI matures will not be the one with the biggest cluster of chips, but the one with the deepest bench of genuinely expert people who know how to point the technology at the right problems.
The piece, published on the think tank’s ASPI Strategist blog, opens from an uncomfortable premise. Many of the people building frontier AI systems now believe the technology could become capable of improving itself, with little human input, within roughly a year. They are not certain of that timeline, and plenty of researchers think it is wildly optimistic. But the leaders of the largest labs are broadly confident that AI will eventually be able to match almost any human skill. If that is even partly right, the analysis contends, then Australia’s policy settings cannot simply chase the last war of buying infrastructure and licensing overseas models.
The news: expertise as the scarce resource
The core of the ASPI argument is a reversal of the usual scarcity story. In a world where capable models are increasingly commoditised and available through an API, the binding constraint is no longer access to the tool. It is the judgement to use it well. Deploying AI inside a hospital, a defence procurement process, a mining operation or a courtroom requires people who understand both the domain and the failure modes of the technology deeply enough to know when its confident-sounding output is wrong. That combination of technical fluency and hard-won domain knowledge is rare, slow to build, and exactly the thing Australia risks hollowing out if it treats AI purely as something to buy rather than something to master.
The essay frames this as a national capability question rather than a jobs-versus-robots panic. The worry is not only that automation might displace workers, but that a thin layer of local expertise leaves the country dependent on foreign systems it cannot fully evaluate, adapt or hold accountable. For a middle power that has spent the past few years talking up sovereign AI, the analysis suggests, sovereignty means little if the humans who can interrogate these systems all sit offshore.
Two ways to read the argument
One view, and the one ASPI leans toward, is that expert humans are the durable advantage precisely because they are hard to replicate. On this reading, government and industry should be pouring resources into training, retention and the messy work of embedding specialists inside institutions, because a model can be copied overnight while a seasoned radiologist who also understands machine learning cannot. Investment in people, this camp argues, compounds in a way that a depreciating GPU cluster never will.
The counter-view is more sceptical, and worth airing honestly. If the frontier labs are right that AI will match virtually any human skill, then betting the national strategy on human expertise could look like training cavalry officers in 1935. Why invest heavily in scarce specialists if the whole premise of the technology is that it will eventually do what they do? The ASPI piece does not pretend this tension is resolved. Its answer is essentially a hedge grounded in uncertainty: nobody actually knows whether or when self-improving AI arrives, the transition will be long and uneven even in the optimistic scenarios, and in every plausible version of the next decade the countries that adapt fastest will be the ones with people capable of steering the technology. Expertise is the insurance policy that pays out whether the aggressive timelines prove true or not.
What it means for Australia
For Australia the stakes are concrete, and they land on ground FluentSea readers will recognise. The country has been building the physical layer of the AI economy at pace, from the Firmus and Blackstone-backed AI factories to the frontier lab ambitions of the big banks. Yet the same institutions repeatedly report that the hardest part is not acquiring the technology but finding the people who can implement it responsibly. Health, defence, resources and financial services all sit on rich proprietary datasets and thorny regulatory obligations, which is exactly the terrain where generic imported models underperform and local expertise earns its keep.
There is also a distribution problem the analysis implicitly raises. Australia’s AI talent is heavily concentrated in a handful of city precincts and a few large employers, while the industries with the most to gain, including agriculture, mining and regional health, are geographically and culturally distant from those clusters. A strategy built on expert humans has to reckon with where those humans are, how they are trained, and whether the pipeline from universities and TAFEs actually feeds the sectors that need them. It is one thing to say expertise is the advantage; it is another to build the institutions that produce it at scale and keep it onshore when global labs are paying premium salaries to lure it away.
The policy conversation in Canberra has so far tilted toward guardrails and sovereignty, with recent debate over safety-first regulation, an OpenAI arrangement pursued by New South Wales, and the Prime Minister’s insistence that Australia will set its own terms on AI. The ASPI framing adds a workforce dimension that is easy to lose in those bigger geopolitical fights. Rules and compute matter, but they are inert without the people who can wield them.
What’s next
The practical questions now fall to government and industry in roughly equal measure. Do the skills and migration settings prioritise the rare hybrid experts the analysis describes, or the entry-level prompt-writers that vendors are keen to certify? Will the coming rounds of AI funding carve out serious money for training and retention rather than concrete and cooling? And can regional and second-tier institutions get a share of the expertise, or will it pool where it already is? None of these are settled, and the honest uncertainty about AI’s trajectory means Australia is being asked to invest in people without knowing exactly how capable the machines will become. The ASPI case is that this uncertainty is precisely the reason to back the humans: they are the one asset that remains valuable across every version of the future.
Sources: ASPI Strategist.
