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 263

Humaneer and Ground Effect

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.

Open in the interactive guide

Summary

Read the original submission for the full text.

Authors of a human-AI compatibility instrument argue that prohibiting AI access for young and vulnerable people moves use out of sight, and that Australian law has no category for persistent, supervised AI support tools; they say they do not claim such a tool substitutes for a registered practitioner. They argue workforce AI competence should be measured, describe their own early-stage instrument, and make four recommendations.

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 the policy question is not whether teenagers use AI but whether the adults around them can see it and shape it.

    The policy question is not whether teenagers use AI. It is whether the adults around them can see it and shape it.

  2. Describes a persistent tool that holds a memory of the person and a supervised tool with a defined human, escalation triggers and reporting, in contrast to a stateless tool.

    A stateless tool begins every conversation from nothing. A persistent tool holds a memory of the person across months, and because it has a baseline, it can notice change. A supervised tool has a defined human on the other end, defined triggers on which it escalates, and defined reporting.

  3. The submission states that developers of persistent, supervised AI support tools face unsettled questions on medical device status, health service status, duty of care and Privacy Act health information provisions.

    A developer building one today faces questions with no settled answer: whether it is software as a medical device under the Therapeutic Goods Act, whether it is providing a health service, what duty of care attaches to an escalation that is missed, and how the Privacy Act's health information provisions apply to a companion's memory.

  4. The submission states that, in its authors' own experience adopting AI in their businesses, adoption fails at the person rather than the tool.

    Our experience adopting AI in our own businesses is that adoption fails at the person, not the tool. The barrier term of reference (d) asks about is rarely cost or access.

  5. Says the standard’s Clause 7 requires competence and awareness of AI risks and organisational policy roles.

    Clause 7 of the standard requires that the people working with an organisation's AI systems be demonstrably competent (7.2) and aware of the risks and of their role in the organisation's AI policy (7.3).

  6. The submission's position is that sensitive companion data should be stored and processed in Australia, with contribution to shared learning off by default.

    Our position is that data of this kind should be stored and processed in Australia, on infrastructure the operator controls, with the person's own record separable from any shared learning and with contribution to shared learning off by default and opt-in by the individual.

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 older people’s consent, disclosure of what is reported to whom, and a plan for changes in consent capacity; exclude dependency mechanics from companion tools.

    The older person must consent, must know what is reported to whom, and there must be a plan for what happens if their capacity to consent changes. And the companion must not have needs of its own that a vulnerable person can be made to feel responsible for; the research in section 2 is explicit that dependency mechanics are the harm vector.

  2. Have the Commonwealth, through the AI Safety Institute with clinical governance from relevant professional colleges, establish an Australian-participant sandbox for persistent supervised AI tools in aged care, mental-health continuity and education.

    That the Commonwealth, through the Australian AI Safety Institute with clinical governance from the relevant professional colleges, establish a regulatory sandbox for persistent, supervised AI support tools in aged care, mental health continuity of care and education, so that Australian rules for this category are built on observed behaviour with Australian participants rather than on imported assumptions or on nothing.

  3. Include a relational competency stream in any national AI skills or literacy framework, prioritised in professional development for teachers, clinicians, aged care and disability workers and others who work alongside people using AI for support.

    That any national AI skills or literacy framework include a relational competency stream, addressing the capacity to reason about human-AI relationships, and that it be prioritised in professional development for teachers, clinicians, aged care and disability workers and others whose work brings them alongside people who use AI for support.

  4. Require Commonwealth AI assurance and government-funded SME adoption programs to measure workforce competence and awareness before and after intervention, consistent with ISO/IEC 42001 Clause 7.

    That Commonwealth AI assurance, and any government-funded SME adoption program, require workforce competence and awareness to be measured before and after intervention, consistent with Clause 7 of ISO/IEC 42001, rather than counting tools deployed or courses completed.

  5. Accompany any recommendation to restrict, ban or decline a pathway for AI use by young or vulnerable people with an assessment of the harm of the unsupervised alternative, and frame school AI policy as a development task (teaching the distinction between fraud, consultation and research) rather than a detection task.

    That any recommendation to restrict, ban or decline a pathway for AI use by young or vulnerable people be accompanied by an assessment of the harm of the unsupervised alternative those people will use instead, and that school AI policy be framed as a development task, teaching the distinction between fraud, consultation and research, rather than a detection task.

  6. Draw a distinction between AI that is stateless and unsupervised and AI that is persistent and supervised.

    The distinction we ask the committee to draw is between AI that is stateless and unsupervised, and AI that is persistent and supervised.

  7. Store/process sensitive companion data in Australia on operator-controlled infrastructure; separate personal records from shared learning, with contributions off by default and individually opt-in.

    Our position is that data of this kind should be stored and processed in Australia, on infrastructure the operator controls, with the person's own record separable from any shared learning and with contribution to shared learning off by default and opt-in by the individual.

Read the original

Open in the interactive guide