AI-powered mushroom ID apps are frequently wrong

Mike Powers
AI-powered mushroom ID apps are frequently wrong

In mushroom foraging, there’s little room for error. Researcher Rick Claypool learned this the hard way.

A few months into his foraging hobby, Claypool picked a basket of what he thought were honey mushrooms, fried them in a pan and ate them with ramen noodles. Then his stomach felt weird.

Fast-forward through some frantic Googling and a trip to the emergency room, Claypool learned he’d been right in the first place — the mushrooms weren’t poisonous. Doctors labeled his symptoms as a panic attack and sent him home.

Others haven’t been so lucky. An Oregon family was hospitalized in 2015 after eating mushrooms an identification app indicated were safe, according to news reports. An Ohio man became seriously ill in 2022 after eating poisonous mushrooms, also misidentified by an app. Confidently identifying wild mushrooms requires expertise, Claypool said, and tech tools haven’t measured up.

Now, a new assortment of AI-powered mushroom identifiers are popping up in the Apple, Google and OpenAI app stores. These tools use artificial intelligence to analyze photos or descriptions of mushrooms and compare them to known varieties. Like past mushroom identification apps, the accuracy is poor, Claypool found in a new report for Public Citizen, a nonprofit consumer advocacy organization. But AI companies and app stores are offering these apps anyway, often without clear disclosures about how often the tools are wrong.

Apple, Google, OpenAI and Microsoft didn’t respond to requests for comment.

The mini-explosion of AI mushrooms apps is emblematic of a larger trend toward adding AI into products that might not benefit from it — from tax software to therapy appointments. Powerful new technology such as large-language models and image generators are good for some things, but consistently spitting out accurate information is not one of them. With its high stakes and frequent mess-ups, mushroom identification is a bad candidate for automation, but companies are doing it anyway, Claypool concluded.

“They’re marketing it like, ‘This is a source of knowledge,’ like it’s the Star Trek computer,” he said. “But the reality is: These things make mistakes all the time.”

Despite the risks, budding foragers seem to increasingly turn to apps for help identifying mushroom species. According to Google Trends, three of the five top searches related to “mushroom identification” mention apps or software. A search for “mushroom” on OpenAI’s GPT Store — where users find specialized chatbots — immediately surfaces suggestions such as Mushroom Guide, which claims to identify mushrooms from pictures and tell whether they’re edible. On the Apple or Google apps stores you’ll find dozens of apps claiming to identify mushrooms, some with “AI” in the names or descriptions.

When Australian scientists tested the accuracy of popular mushroom ID apps last year after a spike in poisonings, they found the most precise one correctly identified dangerous mushrooms 44 percent of the time.

Even low-accuracy AI products can quickly gain consumer trust, however, due to a cognitive distortion called automation bias. As early as 1999, scientists found that people tend to trust a computer’s decisions, even if its recommendations contradict their common sense or training.

In some contexts, AI improves accuracy and outcomes. For example, a February study in JAMA Ophthalmology found that a large-language model chatbot was just as good as eye specialists at recommending diagnoses and treatment for glaucoma and retina diseases.

“Our findings, while promising, should not be interpreted as endorsing direct clinical application due to chatbots’ unclear limitations in complex decision-making, alongside necessary ethical, regulatory, and validation considerations,” the authors note.

When Claypool checked out other avenues of AI mushrooms identification, he found the Amazon bookstore offering what appeared to be AI-authored mushroom field guides. He also tested Microsoft’s Bing Image Creator, asking it to generate and label images of different mushrooms. It made up new mushrooms parts, such as the “ging” and “nulpe,” which do not exist. Inaccurate, AI-generated images of mushrooms could poison future AI training data and search engine results, making these AI systems even less accurate, Claypool wrote.

“We are committed to providing a safe shopping and reading experience for our customers, and we take matters like this seriously,” Amazon spokeswoman Ashley Vanicek said. The company’s guidelines require authors to disclose the use of AI-generated content, and Vanicek noted that Amazon both prevents books from being listed and removes books that do not adhere to those rules.

(Amazon founder Jeff Bezos owns The Washington Post.)

The takeaway: Don’t eat a wild mushroom unless you’ve consulted an expert. (A real, human one.)

Source link

Share This Article
Leave a comment

Leave a Reply

Your email address will not be published. Required fields are marked *