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 486

Mr Peter Horniak

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.

A PauseAI Australia director, submitting personally, supports narrow AI under adapted existing frameworks but argues frontier development poses extinction risks. States uncertainty about near-term superintelligence and extinction. Recommends separate treatment of frontier systems, catastrophic-risk evidence, standards covering frontier training, stronger AISI funding and reporting by large-scale Australian compute operators; says these measures would not slow adoption.

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. Says enterprise AI is governable through adapted existing frameworks, whereas frontier capabilities are discovered during and after training.

    Enterprise AI systems are useful, largely boring, and governable by adapting existing frameworks. That is proportionate to the damage they can cause. A frontier model’s capabilities are not specified in detail in advance, but discovered during and after its training.

  2. Cites a survey of AI researchers giving a median 5% chance, and a mean of 14.4%, of AI causing human extinction or similarly permanent and severe disempowerment within the next 100 years.

    The largest survey of AI researchers gives a median chance of AI “causing human extinction or similarly permanent and severe disempowerment of the human species within the next 100 years” at 5%, with mean 14.4%.

  3. Argues companies failed to slow frontier development voluntarily and that governments have to build a brake.

    Companies had years to voluntarily slow down and they failed to do so. We need a brake, and governments have to build it.

  4. Argues Australia, as a US ally and major trading partner of China, can convene discussions that neither of them can convene themselves.

    Australia is a US ally and major trading partner of China. We can convene the discussions that neither of them can convene themselves.

  5. Says the recommendations would not slow AI adoption and apply only to frontier training runs and large-scale compute.

    My recommendations for the committee would not slow AI adoption. They apply to frontier training runs, which almost no Australian organisation performs, and to visibility over compute at a scale only a few operators possess.

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 frontier systems separately from enterprise AI.

    The systems being built at the frontier are an entirely separate category, and I believe the committee should treat them separately.

  2. Have governments build a brake where companies have failed to slow down voluntarily.

    Companies had years to voluntarily slow down and they failed to do so. We need a brake, and governments have to build it.

  3. Include catastrophic and extinction risk within the committee's scope and take evidence from existential risk researchers.

    Explicitly include catastrophic and extinction risk within the committee’s scope, and take evidence from researchers working on existential risk from AI.

  4. Apply the Australian AI Standards to frontier AI system training within data centres.

    Recommend that the Australian AI Standards apply to frontier AI system training within data centres, not only to deployment and use of them.

  5. Fund the Australian AISI at least equivalently to the UK AISI, preferably consistently with how other portfolios price initiatives that save Australian lives.

    Recommend that the Australian AISI is funded at minimum to the equivalent level of the UK AISI, and preferably to a level consistent with how other portfolios price initiatives that save Australian lives.

  6. Require operators of large-scale AI compute in Australia to report compute capacity, frontier training and safety incidents to a Commonwealth regulator.

    Recommend that operators of large-scale AI compute in Australia are required to report their compute capacity, frontier training, and safety incidents to a Commonwealth regulator.

Read the original

Open in the interactive guide