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 481

The AI Accelerator

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.

Argues Australia's AI response should pair protection with substantially greater investment in capability; regulation is necessary but not sufficient. Draws on the author's operating experience to address productivity, SME adoption, sovereign compute and models, workforce, and cyber and biological risk. Declares commercial interests, including that the proposed AI workforce incentive would benefit the author's businesses. Makes thirteen numbered 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. Policy should target the capability expected when settings take effect rather than current capability.

    Policy must be aimed at the capability that will exist, not the capability that exists today.

  2. Describes a compliance system the author built that assesses insurance sales calls, with coverage increased several-fold and unchanged headcount.

    Coverage has increased several-fold and is being expanded toward the majority of calls, with the architecture designed to scale to complete coverage. Compliance headcount has not changed.

  3. Says that in the author's experience the decisive barrier to SME adoption is being stuck between knowing and doing (inertia, not knowing where to start), rather than access or awareness.

    In my experience, the decisive barrier is neither access nor awareness. Working with organisations of various sizes, I find that people are not unaware. They are stuck. They know the change is coming, often say so directly, and still do not act.

  4. Argues that supplying land and power for foreign-owned AI infrastructure, while models and control stay offshore, is not sovereign capability, though the investment is welcome.

    Australia is attracting significant foreign AI infrastructure investment, which is welcome. But supplying land and power while the models, the commercial terms and the control mechanisms remain offshore is a landlord strategy.

  5. Argues that without deliberate action AI gains are likely to flow to owners of capital while disruption falls on automated workers, and says neither outcome is fixed.

    Without deliberate action, the gains are likely to flow disproportionately to those who own capital and capability, while the disruption lands on workers whose tasks are automated. Neither outcome is fixed.

  6. Argues that AI is lowering barriers to serious biological and cyber harm and that both need dedicated national initiatives rather than general AI regulation.

    In both the biological and cyber domains, AI is lowering the barrier to serious harm and proliferating capability to people without rigorous protocols. Australia should expect both deliberate attacks and accidents. Neither will be adequately addressed by general AI regulation.

Recommendations

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

  1. Adopt a two-pillar national AI strategy giving capability development weight and resourcing at least equal to protection and regulation, with explicit targets and accountability for both.

    Adopt a two-pillar national AI strategy. Give capability development weight and resourcing at least equal to protection and regulation, with explicit targets and accountability for both.

  2. Establish, building on the Office of AI, a standing mechanism combining policy and frontier AI expertise, responsible for detecting capability changes that invalidate existing settings, with a fast path to revise guidance, procurement rules and regulation in weeks.

    Building on the Office of AI, establish a standing mechanism combining public policy and frontier AI expertise, with explicit responsibility for detecting capability changes that invalidate existing settings and a defined fast path to revise guidance, procurement rules and regulation in weeks rather than years.

  3. Require the AI Adopt program and similar initiatives to report how many supported deployments are still operating months later, measure awareness among target businesses, and scale funding toward delivery models producing lasting deployments.

    Require the AI Adopt program and similar initiatives to report how many supported deployments are still operating months later, measure awareness among the businesses the programs are meant to serve, and scale funding toward the delivery models that produce lasting deployments.

  4. Establish an incentive modelled on the R&D Tax Incentive for hiring and training staff in AI-specialist roles, with accelerator support for small firms that build AI systems and deliver AI training, kept simple, generous enough to change decisions and visible to the businesses it is meant to reach.

    Establish an incentive modelled on the Research and Development Tax Incentive for hiring and training staff in AI-specialist roles, together with accelerator support for the small firms that build AI systems and deliver AI training. Make it simple, generous enough to change decisions, and visible to the businesses it is meant to reach.

  5. Treat AI capability in education policy and professional frameworks as an ongoing competency rather than a one-off credential, prioritising pairing domain expertise with AI fluency.

    In education policy and professional frameworks, treat AI capability as an ongoing competency that must be maintained rather than a one-off credential, and prioritise pairing domain expertise with AI fluency.

  6. Fund practical training in how to scope, constrain and supervise AI systems, aimed at the people implementing them in businesses and government.

    Fund practical training in how to scope, constrain and supervise AI systems, aimed at the people implementing them in businesses and government.

  7. Treat model building, sovereign inference and training infrastructure as separate policy problems, prioritising specialist models for Australian domains, sovereign inference for government, banking, health and critical infrastructure, and shared national compute extending existing allocation models with commercial access for Australian companies.

    Treat model building, sovereign inference and training infrastructure as separate policy problems. Prioritise specialist models for Australian domains, whether post-trained or distilled from open-weight models, or trained from the ground up on Australian data, sovereign inference for government, banking, health and critical infrastructure, and shared national compute infrastructure that extends existing national allocation models, with commercial access for Australian companies.

  8. Assess the feasibility of opening suitable Commonwealth-owned land for large-scale compute infrastructure anchored by grid-scale renewable generation and storage.

    Assess the feasibility of opening suitable Commonwealth-owned land for large-scale compute infrastructure anchored by grid-scale renewable generation and storage, where Australia holds a rare combination of natural and institutional advantages.

  9. Test whether data centre and AI infrastructure expectations deliver local research access, skills and supply chain participation, and assess whether arrangements return an appropriate share of value to the public without compromising Australia's attractiveness.

    Build on the Government's published expectations for data centre and AI infrastructure developers by testing whether they are delivering measurable local research access, skills and supply chain participation in practice, and assess whether current arrangements return an appropriate share of value to the Australian public for the land, grid capacity and energy supplied, without compromising Australia's attractiveness as a location.

  10. Make capability transfer an explicit objective in AI infrastructure arrangements, including research partnerships, skills development, access for Australian institutions and firms, and supply chain participation.

    On capability, transfer should be an explicit objective: research partnerships, skills development, access for Australian institutions and firms, and supply chain participation are all negotiable when a country has something scarce to offer.

  11. Test directly how many ongoing jobs large data centre facilities create once built, compared with construction employment.

    One question the Committee should test directly: how many ongoing jobs do facilities of this kind actually create once built? Construction employment is substantial and genuine. Operational employment at large automated data centres is, on the international evidence I have seen, considerably smaller relative to the capital involved.

  12. Incorporate AI provider dependency into critical infrastructure continuity obligations, make model portability the default for government and critical infrastructure deployments, and pursue allied access arrangements.

    Incorporate AI provider dependency into continuity obligations for critical infrastructure and essential services, make model portability the default for government and critical infrastructure deployments, and pursue allied access arrangements so Australian entities are not caught by blanket foreign controls.

  13. Establish focused national programmes for AI-enabled biological and cyber risk, each with clear agency ownership, funding and preparedness exercises, covering deliberate misuse and accidental incidents.

    Treat AI- enabled biological and cyber risk as critical acute risks requiring focused programmes, each with clear agency ownership, funding and preparedness exercises, planning for both deliberate misuse and accidental incidents arising from proliferation.

  14. Review the AI Safety Institute's funding against its mandate and the pace of capability change, and formalise and exercise its links with health security, biosecurity, cyber security and intelligence agencies.

    Review the Institute's funding against the breadth of its mandate and the pace of capability change, and formalise and exercise its links with health security, biosecurity, cyber security and intelligence agencies.

  15. Work with the EU, UK, Canada, Japan, South Korea, New Zealand and other trusted partners on an agenda beyond standards: compute access, model portability, allied access, pooled safety evaluation and joint investment.

    Work with the European Union, the United Kingdom, Canada, Japan, South Korea, New Zealand and other trusted partners on an agenda that goes beyond standards: compute access, model portability, allied access, pooled safety evaluation and joint investment in the specialist middle.

Read the original

Open in the interactive guide