Submission 60
DataMPowered Pty Ltd
Submitter group: Companies.
An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.
Summary
Read the original submission for the full text.
Argues Australia should govern AI-enabled institutional decision-making and workflows, not only external models. Proposes keeping AI inference separate from established fact, separating capability, access, authority and judgement, treating sovereignty as institutional control, and testing the approach through a Commonwealth demonstration program.
Key points
Quotes are the submitter’s own words. Text taken from documents, including scanned pages read by text-recognition software, can differ from the PDF. Check the original. Listed in the same order as on the submission’s page in the interactive guide.
Argues that Australia must govern high-stakes AI-enabled decision-making and administrative workflows rather than solely regulating external AI models.
Australia must govern high-stakes, AI-enabled decision-making and administrative workflows, rather than solely regulating external AI models.
Argues that AI inference should be distinguished from facts established through trusted records or accountable processes.
Inference is not establishment. AI models make inferences. Institutions often need facts to be confirmed through trusted records, evidence or accountable processes.
Argues that sovereignty includes institutional control over knowledge, rules, evidence, access and decisions, and the practical ability to substitute models where possible.
Sovereignty should include control over authoritative knowledge, rules, context, evidence, access decisions, decision rights and audit records, as well as the practical ability to substitute models where possible.
Argues that controls should be proportionate, with stronger requirements for AI uses that affect rights, payments or access to services.
This should be proportionate. Everyday drafting, summarising and productivity tools should not carry the same level of control as systems that affect rights, obligations, payments, access to services or significant institutional decisions. Stronger requirements for provenance, authority and reconstructability should apply where AI materially influences consequential work.
Argues that productivity gains depend on reducing institutional friction in finding and using existing knowledge, not only on better models.
But productivity will not come from better models alone. It will also depend on whether institutions can reduce the everyday friction that stops people from finding, trusting and using the knowledge, systems, rules and expertise the organisation already holds.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Commonwealth AI policy and guidance should look at the full chain of work in which AI is used (intent, context, evidence, rules, access, authority, judgement, execution, outcomes, review), not only models.
Govern consequential work, not only models. Commonwealth AI policy and guidance should look at the full chain of work in which AI is used: the person’s intent, the surrounding context, the evidence, the rules that apply, access, authority, human judgement, execution, outcomes and review.
Keep AI-generated interpretation, prediction or synthesis separate from authoritative institutional knowledge, with systems showing where information came from and how key facts were established.
AI-generated interpretation, prediction or synthesis should not quietly become authoritative institutional knowledge. Systems should show where information came from and how important facts have been properly established.
Encourage standards and reference patterns that make existing policies, services, APIs, data, workflows and accountable human roles discoverable through governed metadata, without moving those assets into foundation models.
Government should encourage standards and reference patterns that make existing policies, services, APIs, data, workflows and accountable human roles visible through governed metadata and discovery mechanisms, without having to move those assets into foundation models.
Treat capability, access, authority, execution and judgement as separate control questions, not bundled into one agent permission, broad access setting or simple human-approval step.
Separate capability, access, authority, execution and judgement. These are different control questions. They should not be bundled into a single agent permission, broad system access setting or simple human-approval step.
Treat sovereign AI capability as institutional control over knowledge, rules, context, evidence, access decisions, decision rights and audit records, including the ability to substitute models where possible, not just compute and models.
Treat sovereign AI capability as institutional control, not just compute and models. Sovereignty should include control over authoritative knowledge, rules, context, evidence, access decisions, decision rights and audit records, as well as the practical ability to substitute models where possible.
For important decisions and actions, show knowledge, inference, applicable rules, authorisation, AI contributions, human judgement and pathways to review or challenge.
For important decisions and actions, institutions should be able to show what was known, what was inferred, which rules applied, who or what was authorised to act, where AI contributed, what judgement was applied and how the outcome can be reviewed or challenged.
Choose real service journeys where knowledge is spread across systems and teams and test whether governed discovery and context assembly improve service while preserving existing legal and accountability structures.
The Commonwealth should choose real service journeys where knowledge is spread across systems and teams, then test whether governed discovery and context assembly can improve service while preserving the legal and accountability structures already in place.
Continue building domestic research, infrastructure, investment and sector-specific AI capability.
Australia should continue building domestic research, infrastructure, investment and sector-specific capability.
Apply stronger provenance, authority and reconstructability requirements where AI materially influences consequential work, with lighter controls for everyday productivity tools.
This should be proportionate. Everyday drafting, summarising and productivity tools should not carry the same level of control as systems that affect rights, obligations, payments, access to services or significant institutional decisions. Stronger requirements for provenance, authority and reconstructability should apply where AI materially influences consequential work.
Use standards and guidance to require provenance, separation of inference and fact, decision authority, execution controls, meaningful human judgement, explainable outcomes and opportunities for review or challenge.
Whichever legislative path Parliament chooses, standards and guidance can still set clear expectations for important AI-enabled work. They can require institutions to show where information came from, keep AI inference separate from established facts, define who has authority to decide or act, control when systems can execute actions, use human judgement where it genuinely matters, keep outcomes explainable and make review or challenge possible.
Run a small Commonwealth demonstration program across a few agencies.
To test these ideas in practice, the Commonwealth could run a small demonstration program across a few agencies.
Measure discovery speed, hand-offs, search effort, context completeness and traceability, access and authority compliance, unresolved gaps, human judgement, reconstructability and users’ ability to understand or challenge results.
The program should test more than speed. It should also ask whether AI helps the institution work better and more safely. Useful measures would include how quickly people can find the right information or service, whether hand-offs and search effort are reduced, whether the assembled context is complete and traceable, whether access and authority limits are respected, how unresolved gaps are handled, whether human effort is focused on real judgement, whether outcomes can be reconstructed, and whether users can understand or challenge a result.
Turn demonstration lessons into practical patterns reusable by Commonwealth agencies and adaptable by industry where appropriate.
The lessons from these demonstrations could then be turned into practical patterns that Commonwealth agencies can reuse, and that industry could adapt where appropriate.