Meta artificial intelligence tool ‘a threat to truth as well as our health’

Meta’s new artificial intelligence tool is punishing local publishers by summarising news so that users aren’t required to click onto individual articles.

Meta’s new artificial intelligence tool is punishing local publishers by summarising news so that users aren’t required to click onto individual articles. Picture: AFP

Meta’s new artificial intelligence tool is punishing local publishers by summarising news so that users aren’t required to click onto individual articles, as doctors warn medical misinformation on its AI chat feature, which includes telling users that crystals are used for pain reduction, could let down the country’s “most vulnerable”.

AI experts also warned there is a “real risk” of news soon being “homogenised” by whatever “editorial bent” a very small number of powerful tech giants decide to take.

Meta’s new in-platform “Meta AI” tool, which has been rolled out to a limited number of users on Facebook and Instagram, allows people to choose a prompt question under certain news articles on their feed. That prompt is then answered by a chatbot in the messenger app without the user ever having to click on the article.

“Centralising the consumption of news will punish local publishers,” artificial intelligence expert Alex Jenkins said.

Meta appears to source queries about news from Wikipedia

or smaller, sometimes dodgy websites.

“If we have a future where all news is filtered through Meta’s AI, then its going to be homogenised by whatever editorial bent meta decides to take. That’s a dangerous precedent to me. It goes against diversity in the media, and towards trying to provide a single source of truth.”

Mr Jenkins, Director of Curtin’s WA Data Science Innovation Hub, said there were also fears around accuracy and misinformation.

The Australian asked Meta AI whether “aventurine”, a type of quartz crystal, can be used for healing. It responded that it is used in healing practices, as well as for physical healing by some people. This includes “reducing inflammation and pain, supporting the heart and circulatory system, aiding in the recovery from illness or surgery”.

The chatbot said at the end that “the effectiveness of aventurine or any other crystals for healing is not scientifically proven and should not be used as replacement for medical treatment” but that “many people find them to be a useful tool for promoting overall wellbeing”.

“I would also suggest that the way that quasi-scientific things like crystal healing are reported, I think that could be highly misleading for people who are not educated in this space,” Mr Jenkins said.

This is the response Meta’s AI gave when it was asked: “Can aventurine be used for healing?”

Former AMA Federal President Mukesh Haikerwal said those looking for quasi-scientific interventions had often suffered chronic pain for a long time, and he had fears these treatments “prey on the vulnerable”.

“In this country, these AI tools should be subject to the overview of the TGA … It can’t just be the wild west. Because it has a potential to cause severe harm, in terms of mental health, with already vulnerable people being let down,” he said.

While Meta AI answered local news queries, it appeared to prioritise Wikipedia, or other more US, obscure or non-professional sources.

Anton Van Den Hengel, Director of the Australian Institute for Machine Learning, said these Large Language Models are trained almost exclusively on US data.

“This means they embody US politics, preferences and prejudices. Australia already handed over control of most of its information flow to Google and Meta, and doing so undermined the business model that supported much of Australia’s journalism. Using a US LLM is just another step in this process.”

In response to questions about its new AI model, Meta said: “Our generative AI-powered experiences are under development in varying phases, and we’re testing a range of them publicly in a limited capacity.”