Predicted benefits from artificial intelligence should be tempered, the leading public service union says as it advocates a considered rollout of the technology in government agencies.
A federal inquiry into AI is gearing up for its 11th hearing, bringing back a number of government departments including Home Affairs and Defence after appearances in September.
Rebecca Fawcett, the new leader of the Community and Public Service Union, will also front the hearing on Friday.
Ms Fawcett has warned of governments misinterpreting increased capacity through AI augmentation as budget savings.
Uncertain costs of AI rollout in businesses to date should inspire caution throughout the public service with the union fearing government costs could increase with no reasonable return on investment.
The submission highlights recent UK government analysis that estimated billions of dollars in savings through automating public service delivery without considering if existing AI models were capable of delivering.
"A focus on efficiency must not come at the expense of lawful, accurate decision making and reliable public services," Ms Fawcett said in a submission to the inquiry.
The union boss called for a sovereign AI capability to ensure the data of Australians was kept onshore with safeguards against overseas access.
It follows warnings from the peak cybersecurity agency on Tuesday which urged the public and private sectors to improve baseline technology to curb an impending wave of AI driven cyber attacks.
The Australian Signals Directorate anticipated a lot of work would need to be done across government agencies to safeguard systems in the coming decade and that would likely add to public service budgets.
Exactly how Australian creators should be renumerated for content being used to train AI models has also been a sticking point for the government.
The union's submission has suggested a system of "data trusts" to allow individuals to pool their data rights to collectively determine access and monetisation with guiding principles used in the superannuation system.
"Such a model could help ensure Australia retains and distributes value created from Australian data more fairly across the community, rather than being concentrated in a small number of multinational firms," Ms Fawcett said.
