Having realistic expectations means that you get disappointed half of the time (assuming well-calibrated realism).
We in Australia seem to have habit of getting the first half of the decision right, and that is probably already optimistic. We are reasonably good at recognising when an old industry or technology has reached its end. We are much less good deciding what should replace it, or at the transition to what comes next.
Some decade ago, we accepted that local car manufacturing had become uneconomic: support was cut, Holden closed in South Australia. Maybe that was unavoidable. But we did very little with the people, facilities and engineering capability that were left behind. Besides factories, there were networks of engineers, suppliers, technicians and people who knew how to build complicated things.
I remember saying we could have started building electrical cars then, but we waited to import them. And maybe it would not have worked anyway. But at least it would have been an attempt to turn an existing capability into something new, rather than writing off the capability and then congratulating ourselves for managing the decline.
AI seems to become the same story. Looking at my LinkedIn, Albos speech on AI seems to be taken mildy optimistic, but I can’t help to find “our” take on AI disappointing.
We may accept that foreign AI systems will transform our economy. We may host the data centres, regulate their water and energy use, and retrain some of the people affected. We may become good customers, careful regulators and reliable providers of land, electricity and raw data. Is that really the ambition?
Unless we also invest in national research compute, technical research, procurement pathways, talent and a few serious national missions, the valuable models, platforms and intellectual property will remain overseas. Australian researchers and companies will continue to build on systems whose capabilities, prices and access conditions are decided elsewhere.
This does not mean that Australia has to reproduce everything done by the United States or China. It probably cannot. But there should be areas where we decide to build real technical depth, own parts of the stack, and use Australian needs as the reason to create new capability here. That requires more than speeches about adoption and productivity. It requires choosing areas where Australia intends to be a producer, rather than merely a customer.
“AI made in Australia” will simply mean that the servers are here. Maybe we should try being right half of the time, instead of being half-right all the time.