From Guardrails to National Resilience

Submission response

Dear Ms Chaney,

Thank you for publishing Shaping the Future of AI for Australia and inviting feedback. Your paper is an important attempt to seize the reins of a debate that Australia has too often left to overseas technology companies, and its 18 recommendations address genuine problems: AI safety, children, privacy, automated government decisions, workforce disruption, data centres, sovereign capability and the distribution of benefits.

My concern is not that these questions are unimportant. It is that the paper starts from the same place as most AI policy:

• How should government regulate AI?
• How should government manage the risks?
• Which opportunities should government support?
• Which institutions, programs and missions should government establish?

All of those questions are important, but they are not enough. They begin with AI as something being done to Australians and with government as the institution that will decide how the country responds. Even the opportunity side of the paper has a heavy government-led flavour: expand the National AI Centre, establish National AI Missions, fund AI for Science, build the national talent base and develop national literacy programs.

I want to start from first principles. What outcome are we trying to achieve, who should set the priorities, and how could AI itself give Australians more power to understand the choices and shape the decisions that affect them?

My proposed outcome remains a more resilient Australia by 2035: a country in which people, communities and institutions can live well through disruption while continuing to shape a better future. Independent representatives should begin using AI with their communities now, openly and transparently, to help people investigate problems, test claims, understand trade-offs and decide what their priorities should be.

You are well placed to demonstrate this in Curtin. Rather than asking constituents only for their views on 18 priorities already framed for them, invite them to help frame the questions, build and challenge the evidence, model the trade-offs and tell you which priorities they want you to take into Parliament. The process would demonstrate both what AI can do and where it can fail, because the prompts, evidence, assumptions, corrections, minority views and human decisions would all be visible.

Controls remain necessary. AI can distort evidence, manufacture apparent agreement, exclude people without access, and make political manipulation much more powerful. That is precisely why elected representatives should demonstrate the opposite use in public: AI extending people's capacity to participate, while people retain responsibility for the goals, values, trade-offs and final choices.

Summary

Shaping the Future of AI for Australia contains many recommendations worth pursuing, and the current policy context makes its concerns more urgent rather than less. Since the paper appeared, the Commonwealth has announced an Office of AI and proposed Australian Standards for AI, while stronger expectations for data-centre energy, water and network costs are also being developed.[2] These are significant steps, but announcements, institutions and standards do not answer the first-principles question: what is AI policy for, and who decides the outcome?

On my reading, 11 of the paper's 18 recommendations are predominantly about controlling harms, five are predominantly about capturing opportunities and two balance both purposes. That is a judgement about their centre of gravity, not an argument against the controls. The deeper problem is that both sides retain Business as Usual (BAU) politics: experts and institutions frame the options, government leads, citizens are surveyed or consulted, and the final priorities are handed back to the same system that framed them.

The missing option is to use AI to change who can do the work of politics. Independents can help their communities become citizen scientists—people able to investigate public problems with AI assistance, publish testable claims, challenge evidence and make trade-offs visible—then carry the priorities their electorates set into Parliament. Ms Chaney could begin directly with Curtin residents and judge the value of the approach by what they are able to learn and decide together.

1. What the paper gets right—and where it starts

The paper is strongest where it refuses passive optimism. It recognises that markets will not automatically produce child safety, privacy, trusted public decisions, workforce security, environmental additionality or a fair distribution of AI's value. Policies dealing with the AI Safety Institute, privacy, automated decision-making, incident reporting, crisis planning and binding data-centre obligations all respond to real gaps, and the proposals for missions, science, talent and literacy recognise that Australia must build capability rather than merely buy services from overseas firms.[1]

The balance of the 18 proposals nevertheless matters. I read Policies 1, 3, 8–15 and 17 as mainly control-oriented; Policies 2, 4, 6, 7 and 16 as mainly opportunity-oriented; and Policies 5 and 18 as balancing control and opportunity. Reasonable people may classify one or two differently, but the pattern is clear enough. When the paper becomes operational it mostly regulates, restricts, reports, prepares and protects. When it turns to opportunity it mostly asks government to expand a centre, create a mission, establish a program, fund research, attract talent or provide literacy.

That is an understandable response to AI risk, but it is still Business as Usual (BAU) thinking. It assumes that the main political task is to improve what government does to or for citizens, rather than asking what citizens can now do for themselves and with one another using tools that were previously available only to governments, universities, corporations and well-funded institutions.

2. Start with the outcome—and with who decides

Before deciding what AI policy should contain, we need to decide what it is for. Economic growth, productivity, competitiveness, jobs and national security all matter, but none is sufficient, because a country can become more productive while becoming less cohesive, wealthier while housing becomes less affordable, more technologically capable while becoming more dependent, and better at building digital infrastructure while the communities providing the land, electricity, water and workers receive little lasting value.

That is why I use resilience as the goal. A more resilient Australia would be better able to anticipate change, protect what matters, recover from shocks and adapt as circumstances change; more capable, not just richer; more cohesive, not just more efficient; and more able to make difficult choices openly before each problem becomes expensive enough to be unavoidable.[3]

But naming resilience is not enough. The prior question is who gets to decide what resilience means in a particular community and which trade-offs should come first. A government can define a national mission for disaster readiness, health, energy or productivity, but people living with bushfire, flood, housing, aged care, transport, small-business viability and energy costs experience these as one connected problem, while councils, state agencies and federal programs continue to manage them as separate portfolios, budgets, timetables and plans.

We must give people the capacity to describe those interactions. We must give them access to the evidence and to AI compute that can help them test competing explanations. We must give them a way to see the costs, benefits and uncertainties rather than being asked whether they support an option after the option has already been designed. People remain responsible for the values and the decisions; AI makes more of the evidence and more of the trade-offs visible enough to argue about honestly.

3. The missing opportunity: change who sets the priorities

The paper asks Australians what kind of AI future they want, and the Curtin survey and invitation for feedback are genuine attempts to listen. Consultation, however, still asks people to enter a process after the questions and proposed answers have been framed. The institution decides what is on the agenda, experts assemble the evidence, citizens express preferences, and the representative or government decides what to do.

For most of modern government that sequence was difficult to avoid. Research was expensive, models required specialist teams, large volumes of public input were hard to examine, and showing the interactions between policy choices took time and institutional capacity. AI changes those constraints. It does not remove the need for experts or make every citizen's claim correct, but it can help many more people search a body of evidence, compare sources, identify assumptions, translate technical material, construct scenarios, expose disagreement and test what would have to be true for a preferred policy to work.

That creates the possibility of a new type of politics in Australia, particularly for independents whose authority depends directly on their relationship with their electorates. An independent can use AI not merely to process correspondence faster or summarise a survey, but to open part of the work of representation to the community: residents help frame the question, publish claims and sources, challenge one another's assumptions, see how competing priorities interact, and decide what they want their representative to pursue.

This is not crowdsourcing a slogan, replacing Parliament with an online poll or allowing an AI system to manufacture consensus. AI is not neutral, and a summary can hide a minority view as easily as it can reveal a shared concern. A credible process would therefore publish the prompts, models, source rules, evidence gaps, corrections and competing outputs; use independent human and AI checks; protect privacy; make non-digital participation available; preserve disagreement rather than average it away; and show exactly where the representative made a judgement that the evidence or community did not settle.

The democratic benefit comes from changing the direction of accountability. Instead of government setting priorities and consulting people about them, people develop enough shared evidence to set priorities for government. Instead of AI being used mainly by political parties, departments and campaigns to predict or influence public opinion, communities use it to build their own capacity to reason and to hold their representatives to the priorities they chose.

This is also a direct way to demonstrate AI's benefits and risks. The benefits would be visible if people can investigate a complex issue more deeply, understand another group's concerns, identify missing evidence or develop a better option. The risks would be visible when an AI invents a source, amplifies a confident voice, embeds a hidden assumption, mishandles private information or produces different conclusions from the same evidence. Publishing both is more valuable than another abstract assurance that AI will be safe, fair or empowering.

4. A practical Curtin demonstration

You could test this approach without waiting for a national program and without pretending that one electorate can settle every design question.

Start with one issue nominated and framed by Curtin residents, not by an AI company, government department or parliamentary office. It could be a local resilience question involving housing, energy, care, transport, small business or the costs and benefits of digital infrastructure. The test would then proceed in six bounded steps:

1. Let the community frame the question. Invite residents to identify the outcome they want, the groups affected and the trade-offs that must not be hidden.
2. Provide access, literacy and non-digital alternatives. Work through libraries, schools, community organisations and local institutions so participation does not depend on a paid AI subscription, technical confidence or spare time. Participation is not free, and a process that ignores access and time will reproduce the inequalities it claims to overcome.
3. Check the evidence in public. Use AI to help residents find and compare sources, but require every material claim to carry a source, confidence statement, limitation and correction path. Treat participants as citizen scientists, not as a focus group.
4. Model options and trade-offs. Publish the assumptions and show how different priorities affect cost, fairness, resilience and other community outcomes. Where the evidence cannot answer a question, say so and identify what would need to be measured next.
5. Return the priorities to the community. Ask residents to deliberate and rank what you should take forward, preserving minority findings and stating where no agreement exists.
6. Report what happened in Parliament and back to Curtin. Show what you accepted, what you changed, what other decision-makers rejected and why, then repeat the process with the evidence gained from the first attempt.

The testable hypothesis is that this process will produce priorities that are better informed, more transparent and more legitimate than consultation on options framed in advance. It may not. The process could be captured by organised groups, exclude people with less time, overwhelm participants with information or create an illusion of agreement. Those are reasons to define and measure the test, not reasons to hand the question back to Business as Usual (BAU) before trying.

If the demonstration works, other independents could run their own electorate tests and compare methods openly. That would allow a different form of national capability to grow from communities upward, with elected representatives connecting local priorities across electorates, rather than waiting for a single national institution to design participation from the centre.

5. What remains for government

Less than the current debate assumes, but what remains is important. Government must still police real harms, protect children and privacy, ensure human accountability for consequential automated decisions, require incident reporting, maintain crisis capability, prevent communities carrying the infrastructure costs of private AI investment, and build enough sovereign capability to understand the systems on which Australia depends—informed interdependence rather than technological isolation.[1][4][5]

Government should also help provide the foundations that make community agency possible: access to capable AI, lifelong AI literacy, trusted public data, protection for private information and time and support for people who would otherwise be excluded. But it should not own the priorities simply because it funds the foundations.

So Australians have a choice. We can treat AI mainly as a technology that needs to be controlled, which is the safe and familiar path and which ends in Business as Usual (BAU). Or we can manage the harms while using AI deliberately to give people more capacity to understand their choices, build evidence together and tell governments what their priorities are instead of waiting to be consulted about priorities chosen for them.

To do that we must break our dependency on government to decide for us.

Yours sincerely,

Rick Molony

Our Resilient World

References

1. Kate Chaney MP, AI Discussion Paper: Shaping the Future of AI for Australia, May 2026; official PDF.
2. Prime Minister of Australia, AI in Australia's interests, 15 July 2026; Department of Industry, Science and Resources, Expectations of data centres and AI infrastructure developers, 23 March 2026.
3. Rick Molony, AI for a Resilient Australia: Who Sets the Priorities?, Our Resilient World, 11 August 2026.
4. Department of Industry, Science and Resources, Australia's AI Safety Institute; see also the National AI Plan.
5. Commonwealth Ombudsman, Fairness in the Targeted Compliance Framework, 2025; Office of the Australian Information Commissioner, Automated Decision-Making Transparency Obligation (APP 1): Issues Paper, 2026.

Method and authorship note

Rick Molony set the argument, values, policy direction and voice controls. This revision was developed with Codex using Rick's 13 August 2026 Product Owner feedback, his published Substack article, Kate Chaney MP's discussion paper and the source-controlled evidence from the earlier response. AI was used to analyse, structure and draft the revision. It is not an author, decision-maker or source of authority. This revision has not yet undergone a new independent audit. Rick Molony retains authorship and all Product Owner decisions.

Suggested citation