This is the ADSEI submission to the Australian Parliamentary Inquiry into Artificial Intelligence and Data Centres.
The Australian Data Science Education Institute has significant concerns about the unconstrained proliferation of data centres in Australia, as well as the unregulated and unethical implementation of Large Language Model Chatbots (LLMs) by commercial organisations.
Our primary concerns centre around these issues:
- Environmental impact, particularly (but not only) CO2 emissions and fresh water consumption
- Data centre CO2 emissions are likely to be catastrophic for our attempts to limit emissions and mitigate climate change. See the report co-authored by Global Energy expert Ketan Joshi with Greenpeace for details. We need laws mandating that, if they are built, they must build new renewable energy capacity to meet their needs at the same time. They must not be allowed to operate off fossil fuels, or to consume existing renewable energy capacity destined for the Australian power grid.
- Similarly, we urgently require laws stating that data centre cooling systems must be closed loop, and that they are not permitted ongoing access to community fresh water supplies, or environmental water flows. If this makes them too expensive to run, then they are too expensive for our environment to sustain.
- The impact of large scale data centres on our power grid must be carefully modelled and planned for. They are unprecedented loads that we do not currently have capacity for. See expert advice here: https://www.ausnetservices.com.au/-/media/project/ausnet/corporate-website/files/our-insights/media-folder/2026-040—integrating-data-centres-whitepaper.pdf?rev=b83267cbc8dc495f861413de7d5be241&hash=20BA4B318B485142CFA208FB952E6A2D
- Unregulated marketing and use of LLMs for inappropriate purposes including (but not only) medical, mental health, education, critical analysis, and decision making
- Large Language Models are not analytical systems or search engines. They are statistical pattern matching machines. The errors they make (so called hallucinations) are an inevitable feature of the technology. They are extremely inappropriate for, among other things, safety critical uses such as healthcare. ADSEI director Dr Linda McIver recently had a medical appointment that was “summarised” by an LLM based system. The summary contained 9 significant errors, with the potential for considerable harm, that were not detected by the specialist, despite his claim to have checked the text carefully. We urgently need legislation regulating their use, and limiting their marketing to applications they are actually suitable for.
- Lack of accountability for harms caused by these systems
- LLMs cannot reason, judge, or assess. They should never be allowed to be marketed for purposes that require those skills. They must never be used to make decisions, for those reasons, but also because they can be used to sidestep accountability. Humans must always be responsible for decisions impacting humans.
- The rise in AI psychosis and suicides clearly attributable to interactions with LLMs is deeply concerning, and a direct result of the sycophancy and human-like interaction, which is a design choice intended to maximuse profit. Companies must be held accountable for these outcomes.
- Unethical practices in building and training LLMs, including (but not only) copyright violations and bias
- There is no reason why commercial companies should be entitled to violate copyright and profit from the intellectual property of others. This must not be allowed.
- Materials scraped from the internet inevitably contain harmful biases and stereotypes. AI companies must be held accountable for addressing these issues.
- Artificial urgency creating a push to circumvent planning controls and consultation processes
- There is no urgency to build data centres for AI
- There is no evidence of economic benefit of building data centres – after building they have minimal staff and therefore minimal jobs, and as foreign owned entities, any profits from their operation go overseas
- There is no evidence that the LLM companies that are driving the data centre push can be profitable.
- It is extremely likely that the AI bubble will collapse, the LLM companies are manufacturing debt faster than profit. Dead companies pay no tax and run no data centres.
- A pause on building data centres until the technological and economic landscape is better understood makes sound economic sense
- Lack of Transparency and Community Input
- Data Centres must not be allowed to circumvent planning processes or regulations
- Communities – both nationally and locally – must have a voice in what gets built around them, and who benefits from it
- For communities to have an authentic voice, there must be transparency, particularly with respect to noise, energy & water use, jobs to be created, and ownership.
- There must be strict and meaningful consequences for not keeping promises made during the application/proposal process
- Lack of regulations around Data Ownership and Stewardship
- We must legislate for
- Meaningful and accountable transparency on data ownership
- Maintaining copyright with no exceptions for training data
- Transparency around ownership of data fed into LLMs, both in training and by users of the models
- Transparency and accountability around the stewardship of data fed into LLMs – who is responsible for data exposed by LLMs?
- Laws around management of Protected Personal Information, what can be fed into LLMs, and who is responsible for protecting that data?
- We must legislate for
