Chef and forager Alan Bergo says he first heard about the books from a worried Facebook post by John Kallas, a fellow forager and author, who runs Wild Food Adventures in Portland, Oregon. Kallas wrote, “We are in a new era of scams,” calling the books a “disturbing trend” and the product of people “that don’t give a hoot about wild foods, don’t know wild foods, and want to drain you of your money.”

But Bergo didn’t think much of this complaint until he was reviewing a manuscript for a new foraging book and saw that the author had referenced a book that he knew from Kallas was “plagiarized blatantly” from the work of fellow forager Sam Thayer.

“I was just incensed, and it was on from there,” Bergo says. The reference was a wake-up call, he says, that “this is way more of a pervasive problem than I had imagined, and we need to squash this right now.”

The alarm in the wild food community goes well beyond plagiarism, however. A trained forager—the sort that would usually publish a book on the topic—would not only guide readers towards foods that are safe to harvest and detail how to prepare them for consumption but would also talk about how to avoid overharvesting for the health of the wider ecosystem. By scraping and collating information without an expert’s guiding hand, these AI books could lead novices to make damaging and potentially deadly mistakes.

Generative AI, the technology behind the popular ChatGPT tool, can be used to “write” books in seconds by drawing from text available online and in published manuscripts. With the use of self-publishing tools like Amazon’s Kindle Direct Publishing, those books can be made available for purchase within a few hours. At present, there are over 3,000 books on Amazon that list ChatGPT as an author or co-author, less than a year after the tool’s debut.

The book world is already grappling with controversy around AI as a result, including a scandal around allegedly AI-written books published under a real author’s name without her consent, and a New York Times investigation into a flood of AI-written travel guides. Comedian Sarah Silverman is one of three authors suing both OpenAI and Meta for copyright infringement, alleging that the companies trained their systems on datasets illegally containing their books.

In July, nearly 8,000 authors signed a letter asking AI companies to stop using their writing to generate text without the authors’ consent and without compensation; this summer, these complaints received some solid backing when an journalist from The Atlantic obtained the Books3 data set, which listed 183,000 copyrighted ebooks that had been pirated and used to train generative AI systems at Bloomberg and Meta.               .

And while numbers are hard to come by, the popularity that foraging saw during the worst of the pandemic seems to have continued: foraging accounts are booming on social media, and those who run wild food workshops continue to report record attendance.

After Bergo posted a video about AI-generated foraging books on Instagram, Alexis Nikole Nelson, a forager with a cumulative 5.9 million followers on her popular @blackforager Instagram and TikTok accounts, posted her own warning about the books. Since then, several have been taken down by Amazon.

“Amazon is constantly evaluating emerging technologies and is committed to providing the best possible shopping, reading, and publishing experience for authors and customers,” Amazon spokesperson Ashley Vanicek told Civil Eats in an email. “All publishers in the store must adhere to our content guidelines, and while we allow AI-generated content, we reserve the right to reject or remove AI-generated content that does not meet those guidelines. We’re committed to providing a safe shopping and reading experience for our customers, and we take matters like this seriously.”

Amazon has recently taken some steps towards controlling the proliferation of these books, by limiting the number of books an author can self-publish through the site to three per day and requiring that self-publishers disclose whether their work was AI-generated. However, the company did not say if they have any method of verifying these disclosures, and copyright compliance is entirely the responsibility of the seller.

In other words, new books appear to replace the ones removed quite easily. In her video, Nelson compared them to the mythical hydra, which grows two heads every time you cut one off.

Computers Don’t Make Good Foragers

“Saying [language models] have human-like intelligence, human-like, fact-based reasoning—that put forward to the general public that language models are factual,” Mitchell adds. “I think that the larger companies that put forward this technology for search did the public a great disservice in giving this impression.”

One of the most obvious concerns about AI-generated foraging books is around misidentification: the potential that they provide incorrect or incomplete information, which could cause newbies to misidentify dangerous plants. Bergo gave the examples of wild carrot, commonly known as Queen Anne’s Lace, which has edible roots and flowers; and wild parsnip, which has edible roots. Both can easily be confused with poison hemlock and water hemlock, all parts of which are toxic enough that consuming even small amounts could be fatal.

“I think there’s a big possibility of missing small details that can make people really sick, or injured in some way,” Bergo added in a follow-up email. He pointed to an instance from a few years ago, when a publisher recalled a foraging blogger’s book because it contained recipes with raw morel mushrooms, which can cause illness.

Mushrooms are at the center of the conversation about AI-written books: In August, the New York Mycological Society warned followers about the new books in a tweet, noting that a mistake could “literally mean life or death.”

While there are fewer toxic mushrooms than popular culture suggests, eating inedible fungi can lead to intense gastrointestinal distress at best and organ damage or death at worst. Some online sleuths have already found AI-generated books misidentifying mushrooms in their pages, including labeling non-edible fungi as edible.

The biggest species of concern is the death cap mushroom, which can easily be confused with edible lookalikes, and which is becoming more common in North America. The Canadian province of Quebec recently reported that there has been an increase in hospital visits for mushroom poisonings as foraging becomes more popular there.

Nathan Wilson, an experienced forager who runs the website Mushroom Observer, isn’t specifically concerned that AI-generated books could lead to mushroom poisoning. He points out that plants actually tend to be much more dangerous than mushrooms and thinks that the concerns around mushrooms can mostly be chalked up to “fungi-phobia.” To him, AI-written books likely aren’t a huge departure from other books and field guides, which have “common and well documented errors” in mushroom identification.

“There have always been charlatans, and one of the challenges of AI is it potentially masks those charlatans in a way that’s harder to dig into, at which point you just have to be skeptical,” he says.

Beyond concerns about identification, experienced foragers and their books often teach readers about the ethics of harvesting, such as leaving enough of certain species for other foragers and for the wider ecosystem. Interest in wild food has grown enough that it has put certain species at risk. For example, ramps and wild onions have become darlings of the culinary scene even as botanists have warned they grow too slowly to sustain heavy foraging.

They’re considered plants of “special concern” in three states, and there’s a law regulating ramp collection in Quebec that comes with a steep fee for violators. According to a 2021 study, “crop wild relatives” hold pest and disease-resistant traits that could be essential to protecting the wider food supply, yet many of these plants are endangered.

Additionally, many in the foraging field are taking a decolonizing approach. Modern foraging often comes with an acknowledgement of the colonial and white supremacist laws that intentionally prevented marginalized groups from gathering wild food. Experts in the field are highlighting the Indigenous knowledge that underpins foraging, in which respect for plants’ role in the larger community builds in a naturally cautious approach to harvesting.

In a 2020 panel by the Indigenous food sovereignty group I-Collective, ethnobotanist Linda Black Elk of the Catawba Nation highlighted the need to return this perspective to modern-day foraging: “If non-Indigenous people can go around knowing plants are relatives, people are way less likely to exploit,” she said.

These are perspectives that artificial intelligence can’t bring to the table. And Bergo is concerned that books written by computers could also be extremely damaging to real authors’ livelihood if they flood the market.

“These are the people that would be hurt the most by something like this: small independent authors,” he said. He suspects that whoever is responsible for the new books moved into foraging quite intentionally. “I think they . . . identify literary niches that have high sales, high interaction and engagement, and low competition. So, they spit out a bunch of books to create competition in that market to skim some cream off the top,” said Bergo.

Reining In the Risks

“The rates at which [AI interfaces] are growing is terrifying, and the way that these technologies are learning is beyond human comprehension,” he says. “I am not at all scared of a book written on foraging, but if someone actually created an interface whose purpose was to persuade people to do something deadly? That’s the kind of thing I am terrified about AI doing.”

The lack of regulation around language models is particularly frustrating to Margaret Mitchell, as she’s worked for much of the last decade on guardrails that could be used to prevent AI from putting human life at risk. One form of this is a type of documentation called model cards, which tell a developer the potential uses, limitations, and ethical considerations of a program.

Developers are then directed to display a disclaimer for the public—Mitchell compares them to the choking hazard warnings you might find on toys with small parts—that explain what the models shouldn’t be used for. This would include any text generated about activities that could lead to someone poisoning themselves, including cooking, mixing household chemicals, and yes, foraging.

“The fact [AI is currently] being used in a domain where poison is involved says that they’re being used out of scope,” Mitchell says. “Any situation where you’re going to be giving advice on eating things, you need to be grounded in a knowledge base.”

Currently there are no repercussions for companies who allow their software to be used in this way, as model cards are only adopted voluntarily. However, there’s a movement aimed at  changing that: the European Union is currently in the process of implementing the EU AI Act, which would require model cards along with other regulations that make AI “safe, transparent, traceable, non-discriminatory, and environmentally friendly.” According to Mitchell, the U.S. and the U.K. are considering similar policies.

Until then, Wilson’s number-one recommendation for foragers continues to be skepticism; he notes that even experienced wild food gatherers often cross-reference multiple sources before they decide to eat something unfamiliar.

For newcomers who might be most at risk from falling for computer-generated information, Wilson has a standard suggestion: “When I was first starting, my iron-clad rule was that I would not eat something unless I have been taught what it is by a person—and if someone is a beginner, I strongly recommend that.”