Study Finds AI Identification of Mushrooms Lacks Accuracy

If picking wild mushrooms and frolicking in the forest is on your bucket list, please go ahead and do it. However, when it comes to determining whether the mushrooms you pick are edible, do not rely on artificial intelligence.

A new in-depth report from the non-profit consumer advocacy organization "Public Citizen" warns against using AI identification tools for mushroom picking, as it is a "high-risk activity that requires real-world skills." The report cites data from Google Trends, which shows an increase in interest and online searches for mushroom identification software and applications in recent years, prompting the warning from "Public Citizen."

While AI-driven applications can be powerful tools for learning more about plants and animals, according to Rick Krippel, the research director at "Public Citizen," simply taking a photo of a mushroom and running it through an app is not enough to accurately identify the species.

Krippel writes in the report, "Experienced local foragers know when and where wild mushrooms appear, and they can identify mushroom species through visual, olfactory, tactile, and sometimes even taste cues. This knowledge is irreplaceable. Local knowledge is crucial."

Krippel points out that a single photo of a mushroom is not sufficient to determine its species. Identifying additional features of the mushroom is crucial. Furthermore, mushroom gatherers must pay attention to where the mushrooms grow, such as on the ground or on wood. If the mushrooms grow on wood, gatherers must also determine the type of wood.

Reportedly, there are over 5,000 meaty mushroom species in North America, most of which are still unknown. Among these mushrooms, 75 are known to be deadly or toxic, and 36 are suspected to be toxic. Additionally, 40 species of mushrooms are known to be toxic when consumed raw but safe when cooked.

Krippel continues to point out that there have been studies on the accuracy of mushroom identification applications, but the results are not reassuring. Australian toxicologists conducted a study in 2022 analyzing three AI applications, and on average, they were only able to correctly identify 50% of wild mushrooms. In some cases, these applications misidentified toxic mushrooms as edible ones.

However, the lack of expertise in mushroom identification using AI is not limited to applications alone, according to Krippel. In 2023, Amazon was flooded with AI-generated mushroom foraging books, many of which contained incorrect and potentially deadly information. According to a report by 404 Media at the time, some mushroom foraging books on Amazon allegedly did not indicate that they were written by AI.

In August 2023, the New York Mycological Society posted on X, stating, "Amazon and other retailers are inundated with AI-generated foraging and identification books. Please only purchase books from reputable authors and foragers, as this is a matter of life and death."

The "Public Citizen" report also highlights the integration of AI into search engines as a concerning issue. For example, last year, Google provided users with step-by-step instructions on how to cook the deadly toxic mushroom Amanita phalloides. According to Krippel, generative AI could also create incorrect images of certain mushroom species, which could confuse users.

Overall, Krippel states that mushroom identification is just one of many examples where overreliance on AI technology leads to harm.