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 532

Australian Renaissance Party

Submitter group: Civil society.

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 AI substitution threatens work, sovereignty and living standards, while describing benefits and recommending adoption for shared prosperity. Proposes employment incentives and floors, sovereign compute and models, a public AI utility, human review and output-stage copyright controls. Prioritises the least advantaged and participation over income transfers; describes employment projections as a stress test, not a forecast.

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 once human labour is no longer needed for production, wages stop being the channel by which wealth reaches people, and the demand that sustained producers falls away.

    Once human labour is no longer needed to produce, the wage stops being the channel by which wealth reaches people, and the demand that sustained the producers falls away with it.

  2. Describes the submission's employment projections as a stress test rather than a forecast.

    We offer this as a stress test, not a forecast, and the quarterly labour-force figures will show which path we are on.

  3. Argues that sovereign AI could retain information and money in Australia and avoid dependence on rented AI.

    This then is the core opportunity of sovereign AI: it would reduce this haemorrhaging of information and money and keep it in circulation within the country. If we do not own the AI we use, we rent it, and the rent comes out of everything we produce with it.

  4. Proposes many smaller facilities spread across the country in preference to a few larger data centres that are more efficient but more exposed.

    We therefore advocate distribution over concentration: many smaller facilities spread across the country, in preference to a few larger ones that are more efficient but more exposed.

  5. Argues that people dealing with automated systems in government or commerce should be able to elect transfer to a human at any point.

    Any person dealing with an automated system, in government or commerce, should have the right to elect transfer to a human being at any point: within two minutes by telephone and four by text.

  6. Argues that learning from lawfully acquired material should be treated as free, with infringement assessed on reproduction of expression.

    We would treat the machine as the law treats a reader. Learning from lawfully acquired material is free; reproducing the expression infringes, whether by a person or a machine.

Recommendations

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

  1. Treat productivity growth as a means and living standards as the end, judging each AI measure by what it does for the bottom fifth of households rather than the average.

    Treat productivity growth as a means and living standards as the end, judge each AI measure by what it does for the bottom fifth of households, not the average

  2. Require firms to disclose task-level automation and report AI-driven changes to the composition of their workforce.

    require firms to disclose task-level automation and report AI-driven changes to the composition of their workforce.

  3. Encourage AI opportunities in medicine and healthspan, provided the industry does not simply maximise profit at the end of life.

    It will create remarkable opportunities in medicine and healthspan, which should be encouraged, so long as the industry does not simply maximise profit and become an end-of-life money grab, as it has in parts of the United States.

  4. Levy compute excise and GST-style per-token foreign-AI charges where they displace Australian work, crediting compute and services earning overseas.

    The tax falls on compute wherever it displaces Australian work: AI compute capacity pays an excise, whether its chips are imported or assembled here, and imported AI services a small per-token levy collected like GST, so sovereign and foreign capacity are treated alike. Compute that earns abroad, whether as exported AI services or as chips shipped out inside Australian-built robots, earns credits instead. The revenue is not for welfare. It reduces payroll tax, and where warranted reverses it into a payment per full-time-equivalent employee.

  5. Pair the compute tax with a negative payroll tax, so that employing a person costs less as automation grows.

    Pair it with a negative payroll tax, so that employing a person costs less as automation grows.

  6. Use compute-tax revenue to reduce payroll tax and, where warranted, pay employers per full-time-equivalent employee, rather than fund welfare.

    The tax falls on compute wherever it displaces Australian work: AI compute capacity pays an excise, whether its chips are imported or assembled here, and imported AI services a small per-token levy collected like GST, so sovereign and foreign capacity are treated alike. Compute that earns abroad, whether as exported AI services or as chips shipped out inside Australian-built robots, earns credits instead. The revenue is not for welfare. It reduces payroll tax, and where warranted reverses it into a payment per full-time-equivalent employee.

  7. Set each larger firm an employment floor following its output, at its own 2025 jobs-to-output ratio or its industry's average if lower, with tradable employment credits and an exemption for jobs shed for safety.

    Set each larger firm an employment floor that follows its output, at its own 2025 ratio of jobs to output or its industry's average if that is lower; a firm below its floor buys employment credits from firms above theirs, as cap and trade does with emissions, and jobs shed for safety are exempt.

  8. Allow above-industry-average firms to shed to the average without charge, earning credits until then; retain each at-or-below-average firm's own 2025 ratio, with industry averages defined by ANZSIC group.

    The second is Floor and Trade. Each firm starts from its own ratio of jobs to output in 2025. A firm that employed more than its industry's average may shed down to that average without charge, but no lower, and until it does, the jobs it holds above the average earn it credits; a firm already at or below the average keeps its own 2025 ratio. The industry average is the ratio across all firms in the same ANZSIC group, and the floor then follows the firm's output.

  9. Permit government to reset the employment-floor base year by later decision.

    The government may reset the base year by later decision.

  10. Exempt jobs shed to remove danger and count contractors towards headcount; begin with larger employers.

    Jobs shed to remove workers from danger are exempt, contractors count towards headcount, and the scheme begins with larger employers.

  11. Start laying the groundwork for a shorter working week, to spread the available jobs.

    Start laying the groundwork for a shorter working week, to spread the available jobs.

  12. If AI displaces work at scale, consider a three-day standard week with two shifts across six days to spread participation.

    If AI does displace work on a large scale, a further measure becomes available: sharing the work. A standard week of three days per worker, with workplaces running two shifts across six days, keeps each workplace operating as it does now while doubling the number of people it engages.

  13. Build Australian-owned compute operators anchored by government workloads, with a national GPU reserve for crises, favouring many smaller facilities and a scheme paying owners to make idle compute available to sovereign workloads.

    Build Australian-owned compute operators, anchored by government workloads, with a national reserve of GPUs that can be called on in a crisis. Favour many smaller facilities across the country over a few large ones, and establish a national scheme that pays owners to make their idle compute available to sovereign workloads.

  14. Run home/office spare-compute workloads only with non-sensitive data; keep sensitive work in accredited Australian facilities.

    Home and office machines would take only work that carries no sensitive data; sensitive work stays in accredited Australian facilities.

  15. Build or distil an openly held, version-fixed Australian foundation model, with mining, agriculture and aquaculture models trained on Australian data; publish government AI procurement sovereignty criteria.

    Build an Australian foundation model, or distil one from open-weight models, openly held and fixed by version, with domain-specific models for mining, agriculture and aquaculture trained on Australian data, and publish sovereignty criteria for government procurement of AI.

  16. Assure sufficient chip stocks held by Australian-owned operators; keep data-centre racks, networking, storage, cooling, power supplies and facilities Australian-owned and operated.

    What we need is an assured supply and a sufficient stock of chips, held by Australian-owned operators. Compute is more than chips. It includes the racks, networking, storage, cooling and power supply that make up the data centre, and the facility itself, much of which can be built and maintained here, and all of it must be Australian-owned and operated.

  17. Renew GPU stocks on a normal commercial cycle as designs are overtaken.

    GPUs are overtaken by faster designs well before they wear out, so the stock must be renewed on a normal commercial cycle.

  18. Secure sovereign baseload power as the foundation of sovereign AI.

    Sovereign baseload power is the foundation of everything above it.

  19. Catalogue and protect national data assets such as Geoscience Australia's mapping and the Reef monitoring corpus and use them to train Australian models, with commercial AI training by others licensed, not free.

    National data assets such as Geoscience Australia's mapping of the continent and the Reef monitoring corpus should be catalogued, protected and used to train our own models, with any commercial AI training by others licensed, not free.

  20. Align everyday Australian-used models with Australian values.

    A model's alignment is its flag, and the models Australians rely on every day should be aligned to Australian values.

  21. For defence, security and critical infrastructure, use small models trained from scratch on controlled data, rather than rely on retraining foreign models.

    Constraint falls short of trust, however. Behaviour hidden in a model's weights can survive retraining, so for defence, security and critical infrastructure we should use small models trained from scratch on data we control, which, being small, we can afford to build.

  22. Use a sovereign model for policy advice inside the ABS under the DATA Act 2022, accrediting CSIRO alongside it; exclude enforcement and national-security uses, without new legislation.

    Let a sovereign model advise policy from inside the ABS under the DATA Act 2022, with the CSIRO accredited alongside it. The Act already rules out enforcement and national-security use, so no new legislation is needed.

  23. Legislate a right of human election for government and essential services: a person on the line within two minutes by phone or four by text, and a human arbitrator able to override any automated decision.

    Legislate a right of human election for government and essential services: a person on the line within two minutes by phone or four by text, and a human arbitrator able to override any automated decision.

  24. Give anyone dealing with automated systems in government or commerce transfer to a human at any point, within two minutes by phone or four by text.

    Any person dealing with an automated system, in government or commerce, should have the right to elect transfer to a human being at any point: within two minutes by telephone and four by text.

  25. Provide independent human arbitrators able to override automated claim denials, application rejections or service withdrawals.

    Where an automated decision denies a claim, rejects an application or withdraws a service, the person affected should have recourse to an independent human arbitrator with the power to override it.

  26. Require the human fall-back for automated systems to be based in Australia and staffed by Australians.

    The human fall-back must be based in Australia and staffed by Australians.

  27. Establish an audit authority with a statutory right to inspect any AI deployed at scale, and give regulators and the AI Safety Institute authority and measurable mandates.

    Establish an audit authority with a statutory right to inspect any AI deployed at scale, give regulators and the Australian AI Safety Institute authority and measurable mandates

  28. Appoint a standing parliamentary committee on sovereign AI.

    and appoint a standing parliamentary committee on sovereign AI.

  29. Treat AI learning from lawfully acquired material as free, with infringement assessed on reproduction of expression.

    We would treat the machine as the law treats a reader. Learning from lawfully acquired material is free; reproducing the expression infringes, whether by a person or a machine.

  30. Enforce copyright for training acquisition and outputs; request EU-aligned training-content disclosure and screening of Australian-supplied AI services' outputs against a national registry of enrolled works.

    Enforce existing copyright law on how training material is acquired and on what models output, request disclosure of training content aligned with the European Union's, and require AI services supplied in Australia to screen their output against a national registry of enrolled works.

  31. Use an independent checking model and a publicly held registry for enrolled authors/publishers; block, attribute or license matches with royalties, blocking only clear reproduction to preserve fair dealing.

    AI services supplied in Australia should screen what they produce, using an independent checking model, against a national registry in which authors and publishers enrol their works. The registry would be held by a public body. Any match would be blocked, attributed, or licensed with a royalty to the author, much as YouTube's Content ID has done for video since 2007. The checker should flag possible matches but block only clear reproduction, so that fair dealing for criticism, review and news reporting is not chilled.

  32. Have Parliament confirm that Crown use of copyright material under the Copyright Act, with payment through a declared collecting society, extends to training models for public purposes.

    The Crown may already use copyright material for its services under section 183 of the Copyright Act, paying the owner through a declared collecting society such as the Copyright Agency. A registry that marks and meters each use would let a sovereign model draw on Australian work lawfully while every author is paid, and for the security services and law enforcement, whose use the Act already allows the government to keep from the owner where disclosure would be contrary to the public interest, the record could be held in confidence and settled in aggregate. The Parliament should confirm that this extends to training models for public purposes.

  33. For Crown uses, mark and meter each use for author payment; where disclosure would contravene the public interest for security/law-enforcement uses, keep records confidential and settle in aggregate.

    A registry that marks and meters each use would let a sovereign model draw on Australian work lawfully while every author is paid, and for the security services and law enforcement, whose use the Act already allows the government to keep from the owner where disclosure would be contrary to the public interest, the record could be held in confidence and settled in aggregate.

  34. Provide AI free at the point of use, like Medicare or the roads, so no Australian business or citizen has to enter a foreign relationship to use it.

    Provide AI free at the point of use, like Medicare or the roads, so that no Australian business or citizen has to enter a foreign relationship to use it.

  35. Preserve conditions for productive participation, rather than provide income alone.

    Participation in the productive life of a society is, for most people, a psychological necessity, and policy must preserve the conditions for it, not merely the income.

  36. Recognise child-rearing as a productive occupation and compensate it as one.

    Recognise child-rearing as a productive occupation and compensate it as one.

  37. Exempt safety-improving automation from compute tax and Floor and Trade on application.

    Automation that makes work safer, in mining, hazard response and dangerous industrial processes, would be exempt from both the compute tax and Floor and Trade on application.

  38. Prefer nearby people for mining-related work where they can fill or train for roles, while permitting specialists from elsewhere; site remote operations centres in regional cities.

    That work should be open to the people who live near it. Where the nearby pool can fill a role, or be trained to fill it, it should be preferred; specialists will reasonably come from where they are concentrated. Remote operations centres should be sited in regional cities.

  39. Back automation of mining, agriculture and aquaculture, opening resulting work to nearby residents.

    Back the automation of mining, agriculture and aquaculture, exempt automation that improves safety from the compute tax and Floor and Trade, and open the resulting work to the people who live near it.

  40. Require banks, telecommunications carriers, platforms and government to deploy AI-based detection of scams and fraud.

    Many of the listed harms can be met by AI itself, in models that detect scams and fraud as they happen. That defence is not automatic: attackers mechanised first, and banks, telecommunications carriers, platforms and government must be required to deploy it.

  41. Consider using independent checking to screen AI harms, employ capture-time content signatures against deepfakes, and keep AI-generated abuse material criminal.

    The independent checking layer we propose for copyright could screen for these harms as well. Where detection is weakest, as with deepfakes, provenance is the more durable defence: content signed where it is captured, so that its origin can be proved. And AI- generated abuse material must remain the crime it is.

Read the original

Open in the interactive guide