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 156

Dr Gabrielle Josling

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.

Examines agencies’ AI claims using FOI records and 24 transparency statements, a non-representative sample. Describes inconsistent definitions, narrowed assurances, disclosure gaps and coverage limits, without establishing non-compliance. Recommends verifiable disclosures and risk-based coverage, and evidence, necessity and proportionality before AI-related claims justify restricting public access or participation.

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 where one agency counts a rules-based or automated process as AI and another does not, their transparency statements are not directly comparable.

    Where one agency counts a rules-based or automated process as AI and another does not, their statements are not directly comparable.

  2. Says that in six of the 24 transparency statements it compared, later versions removed or narrowed specific assurances or forward-looking commitments about decision-making, human involvement, monitoring or governance.

    I compared successive archived versions of the same 24 AI transparency statements. In six cases, later versions removed or narrowed specific assurances or forward-looking commitments about decision-making, human involvement, monitoring, or governance.

  3. Says procurement records, vendor publications, speeches and FOI releases can contain more specific information about agency AI use than the designated statement.

    In some cases, procurement records, vendor publications, speeches and FOI releases contain substantially more specific information about an agency’s AI use than its designated AI transparency statement.

  4. States that it is unclear who, if anyone, is responsible for ensuring the policy intent is delivered, and that agencies' public accounts of their AI use and governance remain largely self-attested and unverified.

    It is therefore unclear who, if anyone, is responsible for ensuring that the policy intent is delivered. Agencies’ public accounts of their AI use and governance remain largely self-attested and unverified.

  5. Argues that mandatory coverage depends on institutional form, so entities such as the NDIA, which makes consequential decisions, are not required to comply with the Policy.

    The National Disability Insurance Agency (NDIA), for example, is a corporate Commonwealth entity. Despite making highly consequential decisions about individuals and handling sensitive personal information, it is not required to publish an AI transparency statement or comply with the Policy’s associated use case register, accountable owner, and impact assessment requirements.

  6. Argues that where AI is relied upon to justify changes that make access or participation harder, government should publish the evidence, distinguish harmful conduct from mere AI use, and show any new restriction is necessary and proportionate.

    Where AI is relied upon to justify changes that make access or participation harder, government should publish the evidence for the claimed effect, distinguish harmful conduct from the mere use of AI, establish why existing mechanisms are inadequate, and show that any new restriction is necessary and proportionate.

Recommendations

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

  1. Require agencies to disclose how they apply the AI definition, including any interpretation, inclusion or exclusion that materially affects what counts as AI; DTA to provide guidance where materially different approaches emerge.

    Agencies should be required to disclose how they apply the AI definition. Any interpretation, inclusion or exclusion that materially affects what an agency treats as AI should be disclosed. The DTA should provide guidance where materially different approaches emerge across agencies.

  2. Make preservation of AI transparency statement history mandatory and enforceable, with previous versions publicly accessible and brief explanations of material changes to assurances.

    Preservation of AI transparency statement history should be mandatory and enforceable. Previous versions should remain publicly accessible, with brief explanations of material changes to assurances about AI use, decision-making, human involvement, monitoring, or governance.

  3. Require agencies to publish a standard, non-sensitive extract of their internal AI use case registers (covered use, purpose, risk level, accountability), with sensitive operational information withheld.

    Agencies should be required to publish a standard, non-sensitive extract of their internal AI use case registers. The published information should identify each covered use, its purpose, relevant risk level, and accountability arrangements. Sensitive operational information could be withheld.

  4. Give an independent body responsibility for periodically checking AI transparency statements against agencies' internal use case registers for omissions, inconsistencies, outdated information and differing classification of AI.

    An independent body should be responsible for checking AI transparency statements against agencies’ internal use case registers. It should periodically identify material omissions, inconsistencies, outdated information, and materially different approaches to classifying AI.

  5. Apply mandatory AI transparency and assurance requirements to consequential AI use regardless of legal form, with a mechanism to extend them to corporate Commonwealth entities where AI materially affects individuals, influences decisions or involves sensitive information.

    Mandatory AI transparency and assurance requirements should apply to consequential AI use regardless of an entity’s legal form. The framework should provide a mechanism to extend relevant transparency and assurance requirements to corporate Commonwealth entities where AI use materially affects individuals, influences administrative decision-making, or involves sensitive information.

  6. Require that claims about AI used to justify reforms making public access or participation harder be evidence-based: publish the evidence, distinguish harmful conduct from mere AI use, explain why existing mechanisms are inadequate, and show the response is necessary and proportionate.

    Claims about AI used to justify reforms that make public access or participation harder should be supported by evidence. Government should publish the evidence for the claimed effect, distinguish harmful conduct from the mere use of AI, explain why existing mechanisms are inadequate, and demonstrate that the proposed response is necessary and proportionate.

  7. Public institutions should not use administrative friction to manage workloads or treat increased participation alone as abuse or system failure.

    Increased participation is not, by itself, evidence of abuse or system failure. Public institutions should not rely on administrative friction to keep workloads manageable.

Read the original

Open in the interactive guide