All election material generated by artificial intelligence technology should come with a clear warning, Australia's Electoral Commissioner has told an inquiry, amid concerns the next federal poll will be swamped with high-tech disinformation.
But Commissioner Tom Rogers stopped short of calling for a blanket ban on the use of generative AI tools during elections, even though he admitted the commission did not have the tools or laws to tackle artificial disinformation online.
The statements were made at the Senate Select Committee inquiry into Adopting Artificial Intelligence on Monday, which also heard concerns the social media firms were doing little to remove AI-generated disinformation and that Australian organisations were slow to adopt and build their own AI technology.
Mr Rogers told the inquiry deceptive AI material had been detected in recent election campaigns in the US, Indonesia, Pakistan and India and Australian voters should "expect things like that to occur at the next election".
Examples of AI-generated electoral misinformation include deepfake videos pretending to deliver messages from candidates and robocalls misleading voters about how to participate in elections.
But Mr Rogers said the Australian Electoral Commission was not equipped to highlight AI-made content, even though it was now being used to spread digital disinformation further and faster than before.
"The AEC does not possess the legislative tools or internal technical capability to deter, detect or then adequately deal with false AI-generated content concerning the election process, such as content that covers where to vote, how to cast a formal vote, and why the voting process may not be secure and trustworthy," he said.
Mr Rogers told the inquiry making digital watermarks mandatory for all AI-generated election material could help voters decide whether to trust claims made in it, but he stopped short of recommending an AI election ban.
He said some uses of AI tools during election campaigns, such as translating messages into other languages, could be useful and regulators had other options.
"A blanket ban on all AI-related activity would be very, very hard to enforce, very impractical, but we would like to see some changes, particularly potentially legislative changes," he said.
"Some overseas election management authorises have asked candidates and parties to sign a voluntary code of conduct ... that's something we could introduce and we don't need legislation for that."
In additional to AI labels, Mr Rogers said a national digital literacy campaign could help to prepare voters, though he warned social media companies such as X were proving less willing to remove disinformation than they were during the 2022 federal election.
CSIRO National Artificial Intelligence Centre director Stela Solar told the committee Australian companies were proving slower to adopt AI technology than those in other nations, although 544 local firms were using it, and another 254 start-ups had applied for CSIRO funding.
Ms Solar said questions remained about the data used to train AI models, but her team was developing a voluntary AI standard that would include guidelines for watermarks and labelling AI content.
