South Australia opened an AI royal commission with work, schools and services in scope

THE SIGNAL IN ONE SENTENCE

South Australia has decided that artificial intelligence is too tangled for one department, one industry panel or one neat policy memo. On October 1, the state began an independent Royal Commission into Artificial Intelligence. Its job is to investigate the economic and social opportunities and consequences of AI in South Australia, then recommend practical reforms. The panel is chaired by Dr Iain Ross AO, an economist and former Federal Court judge. Kate Pounder and Stanford computer scientist Christopher Manning are the other commissioners. The final report is due by July 1, 2027. That is the administrative version. The plain signal is more interesting: South Australia has opened one public room where workers, schools, universities, businesses, unions, creative industries, technology developers, researchers and public servants can argue about what AI should actually do to the state. The terms of reference are broad enough to make that argument useful. They ask about productivity and economic growth, but also employment, workplace practices, education, safeguards, democratic trust, data sovereignty, regulation and public services including health. This is not a commission whose only question is how to attract more software companies. It is supposed to ask who benefits, who carries the risk, what rules are needed and which practical uses are worth testing. A royal commission is one of the strongest forms of public inquiry available to an Australian government. It can gather evidence, hear from affected people, examine institutions and make recommendations with considerably more weight than the average consultation. It is not a new regulator, a legislature or a guarantee that government will implement what it recommends. That distinction should sit beside every headline about it. The commission has begun. It has not found anything yet. It has not approved an AI system for hospitals, rewritten a school curriculum, protected a job or simplified a permit. The work between the launch and the report is the story now. The official terms give the panel several useful levers. It must examine the policy, legislative and regulatory frameworks needed for safe and responsible development, deployment, adoption and use across industry, government, research and the community. That means the inquiry can look beyond whether a model performs well in a demo and ask who remains accountable when it is used in a real decision. It must consider labour markets and workplace practices, including ways to mitigate employment impacts. That opens the door to evidence about job redesign, consultation, training, monitoring, workload, bargaining and the difference between removing drudgery and quietly removing people. It must ask how schools, higher education and vocational education should adapt. That question is bigger than whether students may use a chatbot. It includes what teachers need, how assessment changes, which skills remain worth measuring, how vocational courses match changing jobs and whether every school has comparable access to safe tools and support. It must consider public services, including health. Here the standard cannot simply be faster. A useful pilot should identify the specific task, baseline performance, affected people, error costs, human review, data handling, accessibility and the conditions that would stop the system. The commission is also directed to consider human-centric technology, legal safeguards and accountability. It must examine effects on systemic trust, civics and democratic processes. It must consider sovereignty over data relating to South Australian communities, industry and government, along with how much strategic autonomy the state should seek over AI systems, data and critical capabilities. Those words create a better frame than the familiar choice between enthusiasm and panic. The question is not whether AI is good or bad in the abstract. The question is which system, doing which job, with whose data, under whose authority, measured against what outcome and with what route to challenge a mistake. The terms also tell the commission to look for safe practical pilots and to consider the costs and benefits of its recommendations. That matters. A public inquiry can become a handsome shelf of principles if it never connects a safeguard to a budget, owner, deadline and measurement plan. Suppose a public hospital wants software to help summarize referrals. The inquiry could recommend a standard pilot record: what the tool sees, who reviews the summary, how errors are sampled, whether performance differs across patient groups, how clinicians report incidents and how the hospital exits the contract. The point is not to write one universal hospital algorithm. It is to make responsible experimentation legible. The same structure could help a small manufacturer considering computer vision, a council testing document triage or a school designing an assessment policy. Each needs a defined problem, a baseline, affected voices, a responsible person, a review path and evidence about whether the promised benefit appeared. South Australia can also use the inquiry to connect programs that otherwise live in separate administrative boxes. Workforce policy affects vocational courses. Procurement affects local companies. Data rules affect public-service pilots. Energy and infrastructure affect what can be operated locally. Copyright affects creators whose work may be used by AI systems. Then the boundary line arrives. The commission's website says data centres are outside its terms. Matters primarily under Commonwealth responsibility are also excluded, including national and cyber security, financial-market stability, critical national infrastructure, intellectual property and copyright. Those exclusions are legally and institutionally understandable. A state commission cannot resolve every national policy problem. They also remove several issues that will shape the practical answer to questions still inside the room. An education system can ask how students should use generative tools, but copyright rules affect the material those tools learned from and the work students create. A state agency can discuss data sovereignty while the cloud infrastructure and security regime sit partly outside the inquiry. A creative worker can describe an employment impact while the legal treatment of training data remains a federal matter. A plan for local AI capacity can run into the economics and environmental footprint of data centres that the commission is not examining. The commission will therefore need interfaces, not just boundaries. It can document where a South Australian recommendation depends on Commonwealth law or another process. It can specify the evidence the state needs from infrastructure providers even when the larger data-centre policy question is elsewhere. It can recommend who should negotiate, what should be requested and what happens if the dependency remains unresolved. The public participation record will matter just as much as the final prose. The terms direct the panel to consider views from community representatives, businesses, industry bodies, unions, technology developers, academics, creative industries and governments. A serious process would make that breadth visible. It would show whose evidence was received, which communities were actively reached, where testimony conflicted and why the panel preferred one path. That does not require publishing personal or commercially sensitive information. It does require more than a parade of polished submissions from organizations already equipped to write them. Workers in small firms, teachers outside well-resourced schools, public-service users, disability advocates, Aboriginal communities, regional businesses, artists and families affected by automated decisions may need accessible formats, travel support, local sessions or targeted outreach to participate on equal terms. Good consultation is not a microphone placed in Adelaide and a link posted online. It is a designed route into the evidence. The commission can make its work unusually useful before the final report by publishing a living issue map, an evidence register, plain-language hearing summaries and a list of unanswered questions. It can separate verified findings from stakeholder claims. It can record important disagreements instead of sanding them into consensus. When it evaluates an AI use, it can ask for a small public ledger: the task, owner, data, affected groups, baseline, expected benefit, safety checks, incident path, cost, vendor dependence and decision to continue, change or stop. When it recommends a reform, it can attach the same discipline: responsible agency, legal authority, implementation cost, deadline, success measure and public review date. That would turn the inquiry from a nine-month conversation into reusable civic infrastructure. There is a temptation to judge a commission by the size of its report. The better measure is whether South Australians can trace a line from lived experience to evidence, from evidence to recommendation, from recommendation to a government decision and from that decision to a result. The inquiry has time to build that line, but not enough time to wander. AI products, workplace practices and public expectations will continue changing before July 2027. The commission's own terms recognize that pace and allow flexible procedures. Flexibility should mean updating questions when the evidence changes, not lowering the standard of evidence. Speed should mean short feedback loops, not a sprint past people who bear the consequences. South Australia has chosen a powerful instrument and given it a genuinely broad score. Now the panel has to do the less glamorous work: collect evidence, expose tradeoffs, design practical tests, name the limits of state power and leave the public a record that can be checked after the commissioners go home. The inquiry is open. The signal will be whether the evidence stays open too.