How do we ensure AI strengthens Australia’s economy while reflecting the values we want to build it on?

That was the question at the centre of the AI and the Future of Labor Fringe Event at the ALP National Conference in Adelaide on 24 July. Rozetta Institute CEO Dr Paul Dalby was invited to join Assistant Minister for Science, Technology and the Digital Economy, the Hon Dr Andrew Charlton MP, Chief Scientist at the Australian Institute for Machine Learning, Professor Anton van den Hengel, and Executive Director of the Chifley Research Centre, Emma Dawson for a timely conversation on sovereign AI capability, productivity, and supporting workers through technological change.

The ALP National Conference brings together some 2000 thousand delegates in a major policy forum. AI was a major theme of the conference, particularly in the wake of the Prime Minister’s recent announcements.

Paul’s opening remarks to the Conference are shared below. He explores the tensions at the heart of Australia’s AI debate — a high level of distrust that sits alongside a high level of adoption, real job anxieties that coexists with a looming skills shortfall, and mounting pressure to build sovereign capability as global access to frontier models grows more conditional.

Australia’s Complicated Relationship With AI

AI has been part of our lives for over a decade. It reliably flies our planes, sets the price of an Uber trip, sorts our mail, and recommends the next movie to watch. More recently, and more troublingly, we’ve watched recommender engines change social media from wonderful community message boards into human attention engineers — constantly searching for ways to keep us engaged so we see more advertising.

The public release of large language models has shown a new capability altogether: AI that uses language prompts to predict what we want next, whether that’s code, answers to questions, policy documents, or images.

The relationship status between Australians and AI is, in a word, complicated:

• 44% of Australians say they have not much or no trust that AI will be used responsibly — yet a separate report finds 62% of Australians trust AI more than their best friend when seeking advice on purchases.
• There is high anxiety that we won’t have enough jobs in the future because of AI, even as the Department of Employment and Workplace Relations forecasts we’ll be 1.2 million skilled workers short across industrial and social services in coming years.
• 30% of workers fear AI will replace them at work — but also 30% are using AI without authorisation at work to help get the job done.
• Australians are the highest per-capita users of Claude in the world, yet our companies are laggards in working out how to use AI successfully in the workplace. Most workers are anxious about falling behind and feel their employers aren’t providing adequate, formal training on AI tools.
• Data centres threaten energy security and prices for Australian consumers, but AI is also a potential solution — improving grid efficiency in ways that could ultimately lower energy prices.
• We need our own AI capability in a world where trade relationships have become more transactional, and access to the most advanced models is increasingly at risk or conditional.

Policy responses need to balance all of these competing desires, interests and uncertainties, and land on an outcome that serves the longer-term national interest.

Three no-regrets moves

Despite the uncertainty, there are things we know now and can act on without waiting for the picture to fully clear:

1. A human must remain in control.

 We should require, by law, that a human is always responsible for the actions of an AI system — just as we do for any other automated system. No AI system should be permitted to become a legal cul-de-sac in which responsibility disappears. The machine may recommend, and may even execute defined processes, but a human who can feel the consequences of a governance failure must remain accountable.

2. We need to explore and negotiate how to use AI safely in the workplace.

When any other powerful new tool enters a work environment — vehicles, metal presses, pumps, chemicals — companies are required to train workers properly, for their own safety, for their colleagues’ and customers’ safety, and to get real value from the tool. The AI that flies planes or sorts mail comes with exactly these safeguards. LLMs, by contrast, were released with almost no guidance on how to use them safely or check their performance. The result: many workers are using AI because they know it makes them more productive, but they’re doing it quietly, without their manager’s knowledge, unsure whether it counts as “cheating.” That strongly suggests workers are willing and ready to embrace AI if they feel safe and are clear about how it can be used — but management has real catching up to do on training and governance.

3. We must build national agency.

The short-term suspension of Anthropic’s Mythos model by the US government was a warning shot: as AI becomes more powerful, access to it will become more conditional. At the same time, China’s ability to stay close to the performance of US models — without the advanced chips or cost of the US approach — shows that brute-force scale (massive, high-performance data centres) is not the only path to high-performance AI capability. As part of our broader industrial and technology strategy, Australia can and should build its own AI capability, rather than being a price-taker in the global AI marketplace. We have a long history of doing a lot with less, and should be confident in our ability to build our own agency and capability. While plenty of economists remain sceptical, Australia’s actual technical AI experts are confident we can build world-class, saleable AI for specific applications — not by wrapping a skin around a US LLM, but by building from scratch.

Toward a new settlement

History offers a useful model here. Major technology transitions have, at their best, been managed through negotiated settlements rather than left to work themselves out — arrangements that bring productivity gains and shared responsibility together in one institutional package, rather than pitting one side against the other.

We find ourselves again facing a major technology transition, a productivity lag in the workplace, and a fast-changing international environment — compounded, this time, by a demographic shift that has left Australia more than a million skilled workers short.

AI sits at the centre of all of this, and it may need its own governing bargain. This isn’t a contest between labour and management, or between technologists and everyone else. It’s about Australia collectively working out how to extract the full value of this technology for current and future generations — which means everyone involved needs to understand what’s at stake, what’s still genuinely uncertain, and what the real options and trade-offs are.

There’s still plenty to be concerned about. But at its best, AI can lift worker productivity and wages, sharpen company competitiveness, improve public service delivery, retrain workers into skills gaps faster and more cheaply, and widen opportunities for social and economic mobility — all to Australia’s advantage. A collective effort by our leaders and institutions can help ensure we minimise the downsides and maximise the upside.