AI used to target election fraud and criminal deepfakes

Artificial intelligence tools are being deployed to identify deepfake images of child exploitation and ward off attempts to mislead voters at upcoming Australian elections.

But the technology is also being primed for use in everyday life in Australia, helping shoppers navigate the aisles of Bunnings Warehouse and choose appropriate pet products.

The examples were revealed at the Microsoft AI Tour in Sydney on Wednesday, which also heard about AI trials and experiments from organisations including the Commonwealth Bank, MinterEllison and CSIRO.

The discussions come as the federal government considers mandatory restrictions on the use of AI tools and after a parliamentary inquiry recommended a dedicated law to govern risks and demand greater transparency from AI providers.

But generative AI technology had already begun to transform Australian businesses, government departments and practices, Microsoft ANZ managing director Steven Worrall told attendees.

Microsoft has consulted with state and federal government agencies about potential risks, benefits and guardrails for its use.

"We're working with the Australia Electoral Commission ... because we've seen in other jurisdictions how misinformation, how deepfakes can interfere with the functioning of our elections," he said.

"We know in our organisations, large or small, public sector or private sector, we need to deploy services that are trusted and that there is confidence in the community in regards to the outcomes being achieved."

Microsoft also revealed it had worked with the Australian Federal Police to develop customised AI software to identify deepfake child exploitation images for the Australian Centre to Counter Child Exploitation.

AI software was also being deployed in other areas of the police service, AFP Commander Helen Schneider said, to analyse and quarantine seized material in a secure environment, and to produce warnings or reduce the impact of images of video inspected by officers.

"AI is a powerful tool for detecting and addressing issues relating to deepfakes and other crimes," she said.

"It also offers many possibilities for making us more efficient and effective as a police force."

The software is also being deployed in everyday scenarios, with Bunnings chief transformation officer Leah Balter revealing Wesfarmers had considered more than 130 potential AI projects before focusing on 34 "promising use cases".

The hardware retail chain was testing ways to translate foreign languages for customers in real-time using AI, and one of their "most developed" AI projects was an information service for staff called Ask Lionel.

Microsoft had also collaborated with Petbarn to develop a pet care recommendation engine, Mr Worrall said, and had signed deals with Brisbane Catholic Education, Suncorp and Coles for access to its AI tool Copilot.

National science agency, the CSIRO, had also extended its Microsoft Copilot trial for a year, chief digital officer Rebecca Ostergaard said, but would be strictly auditing the results to look for unintended problems.

"We have to manage the risks with the benefits and there are risks that come with generative AI," she said.

"Perhaps in five years generative AI will just be part of all our lives and we won't know how we got by without it, but at the moment we're in a situation where people are learning to adopt it."

The federal government is considering mandatory AI restrictions for high-risk environments following a public consultation that closed in October.