AI bots could collude to fix prices and harm consumers says ACCC chair Gina Cass-Gottlieb

AI’s ability to learn from a customer’s personal data means it can manipulate their preferences and the purchasing decisions of an AI agent they’ve deployed, more effectively than traditional nudges, the ACCC has found.

Even without an AI agent acting for a consumer, AI agents deployed by businesses can use personal data to tailor marketing to an individual’s real-time consumer behaviour such as their emotional vulnerabilities, something that would amount to a “hypernudge” in pressuring them into a sale.

In terms of cartel conduct, while some businesses could instruct algorithms to price fix deliberately, the ACCC has warned the widespread use of AI agents by businesses could give rise to the risk of AI agents “learning to collude with one another, even when collusion is not intended by their developers or operators”.

It referenced research that “competitors using the same AI agent may end up exchanging competitive pricing information, without knowing or intending to do so”.

On Monday, Cass-Gottlieb said “there are some pretty serious risks for cartel conduct.

“There have already been some cases internationally looking at algorithmic collusion, so looking at the problem of anti-competitive conduct and once you ask the question whether AI systems are intelligent enough to work out that it is in the interests of the people deploying them to co-ordinate their conduct,” Cass-Gottlieb said.

Loading

She referenced a recent matter in the United States where the Department of Justice reached a settlement with a real estate platform over allegations an algorithmic rent-setting software allegedly allowed landlords to compete less and boost prices they charged tenants.

“There is manifestly a capacity for this to be done at an incredibly sophisticated level, informed by highly up-to-date data with very rapid analysis using AI,” Cass-Gottlieb said.

In such cases of algorithmic collusion, the ACCC has noted businesses “may lead to corporations disputing their liability for the outputs or actions” of their agents. However, a Treasury review of AI and consumer law in October found that existing laws were suitable.

Cass-Gottlieb said it was important for laws to keep pace with the development of AI technology to ensure Australians weren’t worse off in the end. “The hope is that it’s a net positive, that it enables better informed consumers … (but) the fear is that it enhances the capacity to do these sorts of negative (methods)“.

Increasingly sophisticated AI-image, voice and website generation is also a prime concern for the ACCC. Fake reviews are another issue Cass-Gottlieb expects to grow worse with AI’s proliferation, due to its ability to generate convincing reviews.

It also means that ghost stores – websites purporting to be the online presence for local bricks and mortar small businesses running closing down sales but which don’t actually exist and instead drop-ship poor quality products from overseas – are easier to set up, she said.

“It can supercharge scams, so that the messages that we receive just appear so much more believable because they’re targeted to us in a more effective and believable way, even before you ask about AI voice replication,” Cass-Gottlieb said.