AI tools can lead to severe mushroom poisoning

A surge in foraging for wild mushrooms has raised concerns among experts, who warn of potentially disastrous consequences, particularly when combined with our overreliance on AI.

“Individuals relying solely on AI technology for mushroom identification have been severely sickened and hospitalized after consuming wild mushrooms that AI systems misidentified as edible,” Public Citizen, a national non-profit consumer rights organization, said in its recent report on the mushrooming risk.

In North America, there are approximately 5,000 “fleshy” mushrooms, and while 15 species are deadly and another 60 toxic, the majority of them have never been tested for toxicity.

Recently, foraging for mushrooms has increased, and so has mushroom poisoning. For years, various internet resources, including dedicated social media groups, have been extremely helpful in educating foragers about mushrooms and staying out of harm’s way.

However, it seems that with the boom of AI, the situation is getting out of control. People now consult AI on mushrooms just as they would consult it on any other aspect of their life. Unfortunately, AI tools are far from reliable at this point.

As per the Public Citizen report, there are about a dozen mushroom identification apps, and the most powerful of them rely on AI to identify a certain species.

“Because of how these apps are marketed, users may understandably believe that identifying a mushroom is as simple as snapping a photo of the mushroom and allowing the AI to deliver a reliable identification,” it reads.

Apparently, most AI-based apps can’t take into account the factors crucial to identifying a mushroom with confidence, like the mushroom’s substrate, bruising features, and smell. What’s more, even if the mushroom is edible, it doesn’t mean that it won’t make a forager sick.

“It may be found in a condition in which it should not be consumed. A piece of fruit that became rotten while on a kitchen counter is, as far as a species-identifying AI tool is concerned, edible. Human judgment is required to recognize that the rotten fruit could cause sickness,” the report reads.

It quotes research by Australian scientists who, prompted by the increase in wild mushroom poisonings, tested three AI apps by “feeding” them pictures of 78 mushrooms. The apps were tested on digital photos of 78 mushrooms in 2020 and 2021.

The best-performing app at the time (Picture Mushroom) accurately identified mushrooms only in half of the cases and identified toxic mushrooms only 44% of the time.

“To be clear, the AI apps’ ability to identify nearly half of all wild mushrooms tested is impressive by any human standard. But, counterintuitively, this impressiveness is part of the reason why using these apps as one’s sole or primary mushroom identification tool carries such a high risk,” Public Citizen said.

Human overtrust in AI-based applications is, indeed, a cause for worry.

“I think if I hadn’t had my phone, I wouldn’t have picked them,” an Ohio man said in 2022 after he required emergency treatment after consuming deadly mushrooms based on an app’s assumption that they were edible.

“Mushrooms offer just one example of how overreliance on AI technology for truthful information can cause harm,” the report concludes.