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 353

Ms Michelle Patkowski

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's AI framework should recognise trustworthy evidence of rights and permissions, so that obligations on copyright, licensing and consent can be shown to have been met. Makes five recommendations on evidence, inputs and outputs, Commonwealth assurance and stakeholder participation. Discloses that the author founded SafeDraftIP, and does not ask the Committee to endorse, procure or fund it.

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. The submitter argues that a dataset disclosure may show a work was present without establishing who authorised its use or under what licence conditions.

    A dataset disclosure may show that a work was present. It does not necessarily establish who authorised its use, whether that person held the relevant rights, what the licence permitted, whether the permission applied to training, whether payment was required, or whether those conditions were observed.

  2. The submitter argues that watermarks and output labels do not establish source, ownership, authority, consent or licence scope for input material.

    This is why watermarking should not be treated as the complete answer to provenance. It can provide useful information about an output. It does not, by itself, establish source, ownership, authority, consent or licence scope on the input side.

  3. The submitter does not recommend a single licensing model but argues that whatever model is used, the authority, scope and conditions of permission should be capable of being established.

    I am not recommending one licensing model. Collective licensing, direct licensing and other arrangements may each have a place. My point is narrower: whatever model is used, the authority, scope and material conditions of the permission should be capable of being established.

  4. Recommends appropriate rights assurance where Commonwealth AI procurement or use materially involves third-party IP, proportionate to risk without unnecessary process for low-risk activity.

    Where a Commonwealth agency procures or uses an AI system that materially involves third party intellectual property, the agency should be able to obtain appropriate assurance about the rights position. What is appropriate will depend on the use and level of risk. Low risk activity should not be burdened with unnecessary process.

  5. The submitter says the detailed national AI training conditions are not yet public, and asks what an organisation would need to retain or produce to demonstrate compliance if they involve copyright, licensing, provenance or similar matters.

    The detailed conditions are not yet public, so I do not assume what they will contain. If those conditions ultimately involve copyright, licensing, creator control, payment, provenance, transparency or lawful source, an implementation question follows immediately: what will an organisation need to retain or produce to demonstrate that it complied?

  6. Argues domestic governance capability can help establish whether Australian IP and data are used under valid authority and conditions.

    Australia will be in a stronger position if it can develop and commercialise domestic capability that helps establish whether Australian intellectual property and data are being used under valid authority and conditions, and that supports trustworthy dealings between rightsholders, AI developers, businesses and government.

Recommendations

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

  1. Recognise trustworthy rights and permissions evidence in Australia's AI governance alongside copyright, licensing, transparency and remuneration, proportionate to risk and neutral as to technology.

    My recommendation is that Australia recognise trustworthy evidence of rights and permissions as a functional part of AI governance. The requirement should be proportionate to risk and neutral as to technology. Five recommendations 1. Recognise trustworthy evidence of rights and permissions as part of Australia's AI governance framework, alongside copyright law, licensing, transparency and remuneration.

  2. Where the law requires a fact about rights or permitted use to be true, ensure there is a proportionate way to establish and preserve evidence of it, which may include source, rightsholder authority, consent, licence scope, permitted use and payment.

    Where the law requires a fact about rights or permitted use to be true, ensure there is a proportionate way to establish and preserve evidence of that fact. Depending on context, this may include source and provenance, rightsholder authority, consent, licence scope, permitted use and payment obligations.

  3. Address both inputs and outputs, since watermarking and labelling do not show whether input material was lawfully sourced.

    Address both inputs and outputs. Watermarking and output labelling can be useful transparency measures, but they do not establish whether input material was lawfully sourced or used under valid authority.

  4. Build appropriate rights and provenance assurance into relevant Commonwealth AI development, procurement and use where third-party intellectual property is materially involved.

    Build appropriate rights and provenance assurance into relevant Commonwealth AI development, procurement and use where third party intellectual property is materially involved.

  5. Give creators, rightsholders, collecting bodies, Australian developers and other affected stakeholders meaningful input before arrangements for standards, training conditions, procurement and assurance are settled.

    As Australia moves into standards, training conditions, procurement and assurance, creators, rightsholders, collecting bodies, Australian developers and other affected stakeholders should have a meaningful opportunity to contribute before the practical arrangements are settled.

  6. Distinguish legal rights, licensing or authorisation mechanisms, transparency about use, and evidence of the relevant authority and conditions.

    For that reason, I suggest the Committee distinguish between four functions that are sometimes treated as if they were interchangeable: the legal right itself, the mechanism used to license or authorise use, transparency about that use, and the evidence that demonstrates the relevant authority and conditions.

  7. Preserve proportionate evidence of human and AI contribution where legal or commercial consequences depend on it; preserve relevant facts rather than recording everything.

    Where a legal or commercial consequence depends on human and AI contribution, that contribution may also need proportionate evidence. The objective should not be to record everything simply because technology makes recording possible. It should be to preserve the facts that matter when rights, payment, lawful use or accountability depend on them.

  8. Consider security, privacy, access control, resilience, auditability and data minimisation in any future rights-evidence requirements.

    If evidence about rights and provenance is to support licensing, procurement, assurance or enforcement, the evidence itself must be reliable. Any future requirements should therefore consider security, privacy, access control, resilience, auditability and data minimisation.

  9. Define evidentiary outcomes and remain neutral about which compliant technology or market mechanism delivers them.

    My recommendation is that Australia define the evidentiary outcomes it needs and remain neutral about which compliant technology or market mechanism delivers them.

  10. Consider trustworthy evidence of source, authority, permission and use across AI training, licensing, procurement, assurance and enforcement where those matters are legally relevant.

    I ask the Committee to recommend that Australia's emerging AI framework recognise the need for trustworthy evidence of source, authority, permission and use where those matters are legally relevant, and that government consider this requirement across AI training, licensing, procurement, assurance and enforcement.

Read the original

Open in the interactive guide