Additional comments from the Australian Greens

1.1This inquiry was an extremely timely exploration of a broad suite of issues facing society as a result of the growth and development of artificial intelligence (AI) including regulatory, human rights, and environmental challenges.

1.2The submissions received by the committee were extremely informative and from a particularly wide range of organisations. Likewise, the inquiry was undertaken with a genuine curiosity from the Senators involved and with a view to finding paths forward.

1.3It is therefore a little disappointing that the final report does not recommend an overarching strategy that would bring Australian regulation of AI into line with the UK, Europe, California or other jurisdictions. The current focus on high-risk AI means many of the more mainstream uses, many of which are positive but none of which are without risk, have not been adequately considered.

1.4To this end the Greens propose the following recommendation.

1.5That the Australian government begin consultation on a National Strategy on AI to develop sovereign capacity for human-centric AI in Australia including innovation and investment, transition to an AI enabled economy and workforce and prioritisation of environmental goals.

1.6While the report grapples with the issue of copyright and proposes government focus on consulting with industry and a plan for transparency for AI models it fails to address the key issue raised regarding unfair contracts. Many submissions and witnesses raised the issue of previously signed contracts (many of which were signed before generative AI was even a thing) being re-read to authorise the use of creative content including voice recordings and publications for purposes from developing large language models (LLMs) to directly creating voice prints or likenesses for use in generative AI models.

1.7To address copyright concerns the Greens propose the following recommendation:

1.8That the Australian Government provide regulatory guidance to ensure contracts in creative industries are not unfairly and inappropriately used to acquire a person’s likeness, inherent personal attributes or creative works for the purposes of machine learning or exploitation using AI.

1.9Automated decision making has been a subject considered at some length including in this inquiry and we support the recommendations in this report regarding implementing the recommendations from the Privacy Review and Royal Commission into Robodebt.

1.10Community mistrust in automated decision making, particularly following Robodebt, means the government should assure the Australian community that where it is used it will not only be transparent but will be in their interest. To this end we propose the following recommendation:

1.11That the Australian Government ensures Automated Decision Making is only used without human intervention in situations where it would provide that the people the subject of the decision making are better off.

1.12Finally, the environmental impacts of AI were canvassed at some length in this report. They are significant and they are increasing. Datacentres are getting bigger, and their footprints are not just in their construction but also in their use of energy, of materials upgrades and water.

1.13Predictions of exponential power use by generative AI and data centres make it clear that planning to reduce power use, operational requirements that focus on minimizing energy use and ensuring that power is sourced from zero carbon renewable power will all be needed to limit the environmental impacts going forward.

1.14As more super powered chips enter the market in Australia the likelihood is that water use will also significantly increase.

1.15The current building standards for data centres are not fit for purpose for AI and current development application rules don’t proactively look into the operating environmental impacts of these centres. That has to change. The Greens propose the following recommendation to address this:

1.16That the Australian Government develop a comprehensive roadmap to address environmental impacts of AI including updating building standards specifically relating to data centres, ensuring data centres are incentivised to access renewable energy, maximising code efficiency, water recycling and hardware reuse and recycling plans. This should include comprehensive environmental reporting of the life cycle impacts of AI data centres and benchmarks for energy and water use.

1.17AI is here to stay and Australia can grow our already strong fundamentals to become a world leader in human-centred and environmentally-responsible AI, but only if we get the right regulatory settings in place and soon.