AI Inquiry Atlas

An unofficial, independent guide to the public records of the Joint Select Committee on Artificial Intelligence. Not connected with the Parliament of Australia or the committee. Its makers also made a submission (no. 481; see About). Summaries and scores are written by AI and can be wrong: check the original before relying on them.

Submission 466

Mr Craig Thomler

Submitter group: Individuals.

Position labels are not shown for individual submitters.

An AI model identified the submitter group and wrote the summary, key points and recommendation summaries. Counts do not measure importance or merit.

Open in the interactive guide

Summary

Read the original submission for the full text.

Argues that Australia should build sovereign AI capability rather than remain a customer of overseas systems. It proposes a ten-year sovereign AI mission, domain-specific models, adoption support for businesses, accountable government use, clearer copyright rules, function-based regulation and continuity planning for foreign services. The author co-founded an Australian generative AI company.

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.

  1. Argues that Australia should treat AI as national capability and retain enough domestic capability to understand, evaluate, adapt and, where necessary, operate AI independently.

    Alternatively, Australia can treat AI as a component of national capability and sovereignty. This does not require us to reproduce every global model. It requires enough domestic capability to understand, evaluate, adapt and, where necessary, independently operate AI for Australian purposes.

    A program corrected obvious text-recognition errors in this quote for display.

  2. Argues that sovereignty is more than hosting in Australia, and that the test is whether Australia can operate and govern a system if the foreign developer withdraws cooperation.

    A system can’t be described as sovereign merely because it is hosted in Australia. The test is whether Australia can continue to operate, inspect, adapt and govern it if the foreign developer withdraws cooperation.

  3. Argues that Australia should build narrow, domain-specific systems where its knowledge and data give it an advantage rather than the largest general-purpose model.

    Australia does not need to build the world’s largest general-purpose AI model. That would be an expensive attempt to follow nations with greater capital, computing capacity and market scale. We should build where Australian knowledge, data and operating conditions create an advantage.

  4. Argues that government should establish practical adoption centres with universities, vocational education providers, industry associations and regional organisations, running implementation programs rather than awareness seminars.

    Government should establish practical adoption centres with universities, vocational education providers, industry associations and regional organisations. These centres should run implementation programs, not awareness seminars.

  5. Argues that the Commonwealth should require accountability assessments where AI materially influences rights, eligibility or access to services.

    The Commonwealth should require accountability assessments where AI materially influences rights, eligibility, employment, grants, procurement, compliance, enforcement or access to services.

  6. Argues that reform should first clarify accountability, evidence and enforcement, with new prohibitions addressing defined harms that existing law cannot adequately manage.

    Reform should first clarify accountability, evidence and enforcement. New prohibitions should address defined harms that existing law cannot adequately manage.

Recommendations

Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.

  1. Establish a ten-year Australian Sovereign AI Mission linking productivity, research, commercialisation, security and public-interest outcomes.

    Establish a ten-year Australian Sovereign AI Mission linking productivity, research, commercialisation, security and public-interest outcomes.

  2. Define sovereign AI as a layered capability covering computing resources, models, data, skills, operations and institutions.

    Define sovereign AI as a layered capability covering computing resources, models, data, skills, operations and institutions.

  3. Select a small portfolio of national AI and robotics missions where Australia has distinctive expertise, legitimate data access and export potential.

    Select a small portfolio of national AI and robotics missions where Australia has distinctive expertise, legitimate data access and export potential.

  4. Focus Australian AI development on agriculture; water, energy and critical infrastructure; mining, resources and remote operations; disaster resilience; environment and biodiversity; medical and rural health; public administration; and defence, cyber security and intelligence.

    We should build where Australian knowledge, data and operating conditions create an advantage. • agriculture under variable and extreme conditions; • water, energy and critical-infrastructure management; • mining, resources and remote operations; • bushfire, flood and disaster resilience; • environmental and biodiversity management; • medical research and rural healthcare; • regulation, legislation and public administration; and • defence, cyber security and intelligence support.

  5. Give each publicly funded national mission a path from research through operational trials, procurement and domestic scale to export.

    Public funding shouldn’t end with research demonstrations. Each national mission should include a path from research to operational trial, procurement, domestic scale and export.

  6. Create shared national AI infrastructure for Australian researchers, start-ups, public institutions and strategically important industries.

    Create shared national AI infrastructure for Australian researchers, start-ups, public institutions and strategically important industries.

  7. Use Commonwealth procurement to build domestic capability with knowledge transfer, Australian participation, portability and tested exit arrangements.

    Use Commonwealth procurement to build domestic capability with knowledge transfer, Australian participation, portability and tested exit arrangements.

  8. Establish practical adoption centres for small, medium and family businesses, including regional delivery.

    Establish practical adoption centres including regional delivery and programs for small, medium and family businesses.

  9. Build adoption centres with universities, vocational providers, industry associations and regional organisations; run implementation programs delivering assessed use cases, data/risk plans, controlled trials, workforce plans and measurable success criteria.

    Government should establish practical adoption centres with universities, vocational education providers, industry associations and regional organisations. These centres should run implementation programs, not awareness seminars. Each business should leave with an assessed use case, data and risk plan, controlled trial, workforce plan and measurable success criteria.

  10. Require accountable human decision-making where Commonwealth AI materially affects rights, services, eligibility, employment, procurement, compliance or enforcement.

    Require accountable human decision-making where Commonwealth AI materially affects rights, services, eligibility, employment, procurement, compliance or enforcement.

  11. Require Commonwealth assessments where AI materially influences rights, eligibility, employment, grants, procurement, compliance, enforcement or services, identifying accountable officials, AI role, model/data provenance, limitations, validation evidence, review rights, monitoring, recordkeeping and suspension.

    The Commonwealth should require accountability assessments where AI materially influences rights, eligibility, employment, grants, procurement, compliance, enforcement or access to services. It should identify the accountable official, the role of the AI, model and data provenance, limitations, validation evidence, review rights, monitoring, recordkeeping and suspension arrangements.

  12. Disclose government AI use where a system materially influences a person's rights, obligations, access to services or treatment by government.

    It should disclose AI use where a system materially influences a person’s rights, obligations, access to services or treatment by government.

  13. Create a public register of significant Commonwealth AI systems subject to limited and justified national-security exemptions.

    Create a public register of significant Commonwealth AI systems subject to limited and justified national-security exemptions.

  14. Develop machine-usable government knowledge assets with validated links to authoritative sources, version control and clear ownership.

    Develop machine-usable government knowledge assets with validated links to authoritative sources, version control and clear ownership.

  15. For machine-usable government rules, retain the original law or approved guidance as authority; preserve source traceability, independent validation, version and approval records, realistic testing and updates when sources change.

    But an AI-generated interpretation must never quietly become the authoritative rule. Authority remains with the legislation, policy, standard or approved guidance from which it was derived. Each machine-usable rule should identify its source, preserve traceability, be independently validated, record its version and approval, be tested against realistic cases and be updated when the source changes.

  16. Invest in shared synthetic datasets and controlled evaluation environments for grants, health, infrastructure, regulation and service delivery; test synthetic data for unintended disclosure, distorted patterns, missing edge cases and bias.

    Government should invest in shared synthetic datasets and controlled evaluation environments for areas such as grants, health, infrastructure, regulation and service delivery. Synthetic data is not automatically safe or representative. It must be tested for unintended disclosure, distorted patterns, missing edge cases and bias.

  17. Clarify significant training-material category disclosure, licensing and payment, unlicensed online material, AI-assisted work’s status, confidential information protection, attribution and provenance.

    Government should clarify disclosure of significant training-material categories, licensing and payment mechanisms, treatment of online material not licensed for training, the status of AI-assisted work, protection of confidential information, and attribution and provenance standards.

  18. Give Aboriginal and Torres Strait Islander cultural material special care; consider Indigenous data sovereignty and cultural intellectual property separately from ordinary commercial copyright, recognising legal availability does not establish cultural authority or informed consent.

    Special care is required for Aboriginal and Torres Strait Islander cultural material. Legal availability does not necessarily amount to cultural authority or informed consent. Indigenous data sovereignty and cultural intellectual property require separate consideration from ordinary commercial copyright.

  19. Embed critical AI literacy across education and workplaces, including model limitations and worker participation in job redesign.

    Embed critical AI literacy across education and workplaces including verification, bias detection, model limitations and worker participation in job redesign.

  20. Include authoritative-source checks, separating option generation from decisions, testing competing cases, recording important outputs and model versions, bias identification and knowing when to avoid AI in AI literacy.

    AI literacy must include disciplined scepticism Relevant term: 7. Australians need access to AI tools. They also need the judgement to challenge them. Fluent output is not proof. AI literacy must include checking authoritative sources, separating option generation from decision- making, testing competing cases, recording important outputs and model versions, identifying bias, and knowing when not to use AI.

  21. Use workforce policy to examine who receives productivity benefits; involve workers in role redesign and use benefits for service quality, capability and sustainable work.

    Workforce policy must also ask who receives the productivity benefit. Adoption that merely increases monitoring, workload or employment insecurity will not improve living standards. Workers should participate in redesigning affected roles, with benefits used to improve service quality, capability and the sustainability of work.

  22. Regulate AI by function and consequence.

    Regulate by function and consequence clarifying responsibility under existing law before creating broad new prohibitions.

  23. Focus regulation on system function, who may be harmed, harm scale and reversibility, whether operation can be examined, whether decisions can be challenged, and accountability.

    Regulation should focus on what a system does, who may be harmed, the scale and reversibility of harm, whether its operation can be examined, whether a decision can be challenged and who remains accountable.

  24. First clarify accountability, evidence and enforcement; confine new prohibitions to defined harms that existing law cannot adequately manage.

    Reform should first clarify accountability, evidence and enforcement. New prohibitions should address defined harms that existing law cannot adequately manage.

  25. For increasingly capable systems, address objectives, incentives and context rather than relying entirely on containment.

    For increasingly capable systems, static restrictions may not be enough. Strengthening a barrier does not necessarily alter an intelligent system’s goal. It may only create a more difficult obstacle for it to overcome. Governance must address objectives, incentives and context, not rely entirely on containment.

  26. Strengthen prevention and response for AI-enabled scams, impersonation and deepfakes with particular attention to vulnerable Australians.

    Strengthen prevention and response for AI-enabled scams, impersonation and deepfakes with particular attention to vulnerable Australians.

  27. Provide stronger identity/transaction safeguards, rapid reporting, clear platform/telecommunications/financial-institution responsibilities, accessible education, high-risk synthetic-media provenance standards and specialist support for vulnerable groups.

    Australia needs stronger identity and transaction safeguards, rapid reporting processes, clear responsibilities across platforms, telecommunications and financial institutions, accessible education, provenance standards for high-risk synthetic media, and specialist support for vulnerable groups.

  28. Combine technical controls, institutional responsibility and public capability in responses to AI harm.

    No single watermark or detection tool will solve the problem. Malicious actors will adapt. The response must combine technical controls, institutional responsibility and public capability.

  29. Include service-withdrawal risk in the Australian AI Safety Institute’s technical-evaluation remit.

    Give the Australian AI Safety Institute a strategic technical-evaluation role including foreign dependency, supply-chain concentration and service-withdrawal risks.

  30. Give the AI Safety Institute independent technical evaluation and government/research/industry evidence coordination, without duplicating intelligence, cyber security or law enforcement; cover dependency, supply concentration, continuity and critical-function model changes.

    The Australian AI Safety Institute should not duplicate intelligence, cyber security or law-enforcement bodies. It should provide independent technical evaluation and connect evidence across government, research and industry. Its remit should include foreign dependency, supply-chain concentration, operational continuity and material changes to models used in critical functions.

  31. Provide secure evaluation facilities, computing access, multidisciplinary teams, structured information sharing, protected vulnerability-reporting channels, foreign-service loss/compromise exercises and domestic alternatives for essential national-continuity functions.

    Australia needs secure evaluation facilities, appropriate computing access, multidisciplinary teams, structured information-sharing, protected channels for reporting vulnerabilities, exercises testing the loss or compromise of foreign AI services, and domestic alternatives for functions essential to national continuity.

  32. Require tested continuity plans for critical AI-supported functions including alternatives where foreign services become unavailable.

    Require tested continuity plans for critical AI-supported functions including alternatives where foreign services become unavailable.

    A program corrected obvious text-recognition errors in this quote for display.

  33. Report annually to Parliament on Australian AI capability including regional adoption, skills, computing access, research commercialisation, public-sector capability and strategic dependencies.

    Report annually to Parliament on Australian AI capability including regional adoption, skills, computing access, research commercialisation, public-sector capability and strategic dependencies.

    A program corrected obvious text-recognition errors in this quote for display.

  34. Measure adoption programs by benefits for workers, customers and communities, rather than licence deployment counts.

    Adoption programs should measure benefits to workers, customers and communities, not the number of licences deployed.

  35. Adopt a regional strategy providing reliable connectivity, practical training, shared capability for smaller councils and businesses, and participation by regional universities and vocational education providers.

    A regional strategy should provide reliable connectivity, practical training, shared capability for smaller councils and businesses, and direct participation by regional universities and vocational education providers.

  36. Continue using international models, attracting investment and participating in global research while preserving alternatives if commercial terms, export controls, geopolitics or corporate priorities change.

    Australia should continue to use international models, attract investment and participate in global research. But we should be able to benefit from global AI without becoming helpless if commercial terms, export controls, geopolitical conditions or corporate priorities change.

Read the original

In some quotes a program corrected obvious text-recognition errors for display.

Open in the interactive guide