Submission 81
Uniting Church in Australia and U Ethical
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.
Summary
Read the original submission for the full text.
Recommends an AI Safety Act, greater AI Safety Institute resources and international oversight of frontier AI. Describes harms from AI-generated child sexual abuse material, deepfakes and autonomous weapons, and risks from biological weapons and loss of control. The supplementary submission adds evidence on workers, children and health, and asks for financial support for displaced workers undertaking retraining.
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.
Argues for strict oversight of frontier AI even if extreme risks are low probability, and ideally prohibiting superintelligent AI development.
Given these extreme risks, even if they are low probability, governments should strictly oversee the development of frontier AI models. Ideally, attempts to develop a superintelligent AI model should be prohibited.
Argues that safeguards could restrict some perpetrators from generating child sexual abuse material, even if technically sophisticated perpetrators defeat them.
While it is likely that technically sophisticated perpetrators will be able to defeat safeguards and misuse some generative AI software with built-in safeguards, it is likely to restrict a proportion of perpetrators from generating child sexual abuse material.
Recommends requiring generative AI developers to take all reasonable steps against misuse and easy generation of child sexual abuse material and non-consensual intimate images.
Amend the Criminal Code to require generative AI developers to take all reasonable steps to ensure the products they are developing have safeguards against misuse and cannot be easily used to generate child sexual abuse material and non-consensual intimate images.
Recommends assurance that the ADF will not use weapons to kill without a human making informed decisions and being accountable.
Provide assurance that the Australian Defence Force will not use weapons systems to kill people without a human making informed decisions and being accountable for those decisions.
It says the effect of AI on the workforce is likely to be diverse, including job loss and worse pay and conditions for some.
The impact of AI on the Australian workforce is likely to be diverse, enhancing workplace satisfaction for some, ending employment for others and worsening pay and conditions for yet others.
Recommends empowering and resourcing existing regulators, including ACCC, ASIC, OAIC and eSafety, to address AI-related harms within their domains.
Empower existing regulators to deal with curbing AI-related harms that fall within their jurisdiction. Bodies like the Australian Competition and Consumer Commission, Australian Securities and Investments Commission, Office of the Australian Information Commissioner, and the eSafety Commissioner must be empowered and resourced to address AI-related harms within their domains.
Recommendations
Recommendations are shown beside the submission’s own words. Listed in the same order as on its page in the interactive guide.
Significantly increase resourcing to the Australian AI Safety Institute.
Significantly increase resourcing to the Australian AI Safety Institute.
Pursue international agreements for oversight of frontier AI model development to reduce harm.
Seek to work with other governments towards international agreements for the international community to have oversight of the development of frontier AI models to reduce the risks of harm.
Develop and pass an Artificial Intelligence Safety Act with obligations on AI developers to minimise risks from AI products and systems.
Develop and pass an Artificial Intelligence Safety Act with obligations on AI developers to make sure that risks from AI products and systems are minimised.
Require mandatory reporting of significant AI-related harms, with voluntary reporting for users negatively affected by AI.
Require mandatory reporting of significant AI-related harms and voluntary reporting for users negatively affected by AI.
Empower and resource existing regulators (ACCC, ASIC, OAIC, eSafety Commissioner) to address AI-related harms within their domains.
Empower existing regulators to deal with curbing AI-related harms that fall within their jurisdiction. Bodies like the Australian Competition and Consumer Commission, Australian Securities and Investments Commission, Office of the Australian Information Commissioner, and the eSafety Commissioner must be empowered and resourced to address AI-related harms within their domains.
Continue efforts to strengthen the Biological Weapons Convention with a board of experts and enhanced verification and compliance measures.
continue efforts to strengthen the Biological Weapons Convention with a board of experts and enhanced verification and compliance measures.
Increase penalties for running a business that generates AI child sexual abuse material for others, above those for generating it for oneself.
Increase the penalties for people who run businesses to generate AI child sexual abuse material for others, which should be higher than for a person who generates such material for themselves.
Require generative AI developers to take all reasonable steps against misuse and easy generation of child sexual abuse material and non-consensual intimate images.
Amend the Criminal Code to require generative AI developers to take all reasonable steps to ensure the products they are developing have safeguards against misuse and cannot be easily used to generate child sexual abuse material and non-consensual intimate images.
Include safeguards against training on databases containing child sexual abuse material and against user requests that would generate it.
Such safeguards should include not training generative AI on databases that include child sexual abuse material and not allowing users to enter requests that would result in the generation of child sexual abuse material.
Create a power to take down a generative AI model until it can be made safe from generating child sexual abuse material and non-consensual intimate images.
There needs to be a power to take down a generative AI model until it can be made safe from generating child sexual abuse material and non-consensual intimate images.
Where possible, require generative AI developers to nominate individuals or teams accountable for ensuring the model's safety.
Where possible, generative AI developers should be required to nominate individuals or teams accountable for ensuring the model's safety.
Create an offence to use personal data or digital information to create digital models or files that facilitate creation of AI or computer-generated child sexual abuse material.
Create an offence to use personal data or digital information to create digital models or files that facilitate the creation of AI or computer-generated child sexual abuse material.
Criminalise developers' failure to stress test safeguards against generating child sexual abuse material, while considering a testing mechanism that avoids unlawful generation during testing.
Create an offence for generative AI developers not to stress test their software to ensure that it has effective safeguards against generating child sexual abuse material. The eSafety Commissioner has recommended such stress testing and the need to assume there will be malicious users. The mechanism for such testing will need consideration to ensure that developers do not break the law by generating child sexual abuse material during testing.
Create an offence to provide advice, tools or manuals that explicitly advise on how to use AI tools to produce child sexual abuse material.
Create an offence to provide advice, tools or manuals that explicitly advise on how to use AI tools to produce child sexual abuse material.
Create an offence for a search engine to return results for generative AI services to produce child sexual abuse material, with a defence for meaningful steps taken to avoid this.
Create an offence for a search engine to return results to users for generative AI services to produce child sexual abuse material. Search engines should have a defence that they took meaningful steps to ensure they are not returning such results.
Create an offence to create a generative AI chatbot that allows sexual communication with a simulated child.
Create an offence to create a generative AI chatbot that allows for sexual communication with a simulated child.
Expand offences on sexually exploitative deepfakes to include audio-only depictions, and make creating such material an offence.
Expand the offences on generating sexually exploitative deepfakes, including audio-only depictions and ensure it is an offence to create such material.
Ensure standardised national school response protocols to students' use of exploitative or abusive deepfakes, building on the eSafety Commissioner's Toolkit.
Ensure there are standardised national school response protocols to incidents of students using exploitative or abusive deepfakes, building on the eSafety Commissioner's Toolkit.
Create a Commonwealth survivors of crime compensation scheme for survivors of online child sexual abuse, including sexually exploitative deepfakes.
Create a Commonwealth survivors of crime compensation scheme for survivors of online child sexual abuse, including survivors of sexually exploitative deepfakes.
Create offences for non-sexual deepfakes used for criminal or harmful purposes, with aggravated penalties where significant harm is proven.
Create offences to address non-sexual deepfakes used for criminal or harmful purposes, such as impersonation for fraud, election interference and fabrication of defamatory “evidence”. There should be aggravated penalties where significant harm is proven.
Provide assurance that the ADF will not use weapons systems to kill without a human making informed, accountable decisions.
Provide assurance that the Australian Defence Force will not use weapons systems to kill people without a human making informed decisions and being accountable for those decisions.
Not allow the sale of weapons systems that can kill people with no person involved in the decision, placing them on a banned list.
Not allow the sale of weapons systems that can kill people, with no person being involved in the decision to do so, placing such weapons on a banned list.
Support negotiations for international law on autonomy in weapons systems, ensuring human control of force and prohibiting machines that target people.
Support negotiations for new international law on autonomy in weapons systems - to ensure human control in the use of force and to prohibit machines that target people, reducing us to objects, stereotypes, and data points.
Set penalties for businesses producing child abuse material above those for people purchasing their services.
There are businesses that provide services to produce child abuse material. People engaged in this activity earn significant incomes. Penalties for such businesses should be higher than those for people who purchase their services.
Even if extreme risks are low probability, strictly oversee frontier AI development and, ideally, prohibit attempts to develop a superintelligent AI model.
Given these extreme risks, even if they are low probability, governments should strictly oversee the development of frontier AI models. Ideally, attempts to develop a superintelligent AI model should be prohibited.
Make it an offence to produce a chatbot that provides advice on how to use AI tools to make child sexual abuse material or on how to abuse children sexually.
it should also be an offence to produce a chatbot that provides advice on how to use AI tools to make child sexual abuse material or advice on how to abuse children sexually.
After this inquiry, hold further inquiries into specific AI aspects needing deeper consideration as identified through this inquiry's high-level evidence.
After the Committee completes this inquiry, we would urge the Committee to hold subsequent inquiries into specific aspects of AI that need further deeper consideration as identified through the high-level evidence provided to this inquiry.
Provide job security measures that financially support workers retraining after displacement by AI, as exists in Sweden.
the Committee should recommend the provision of greater job security measures that will financially support workers undertaking retraining due to being displaced by AI. Such a safeguard exists in Sweden.