A book on Manitoba mushrooms sold on Amazon likely written by artificial intelligence (AI) is worrying mycophiles. In addition to inconsistent content, it poses a danger to readers who might use it as a guide to mushroom picking.
When the dean of the Faculty of Arts and the Faculty of Sciences of the University of Saint-Boniface and mushroom lover, Alexandre Brassard, learned of the book’s existence Mushrooms of Manitoba by Jay O. Mark, he was intrigued. For years he has wanted to publish a practical guide for mushroom pickers in Manitoba.
The academic is the creator of a Facebook group which brings together more than 11,000 Manitoban mycophiles. He also gives introductory lessons in mycology and it was one of his students who recommended the book to him.
On Amazon, we do not have access to the table of contents, we do not have a biography of the author. It was $20. I said to myself, why not, let’s go for it. I ordered it. And the work is very strange
says Mr. Brassard.
A book of this type will normally review species that are common, that are well known, that are popular. [Le guide] reviews around fifteen poorly chosen species [qui ne] are not necessarily common species. There are even several species that are [évoquées] and which don’t even grow in Manitoba
he continues.
There are large sections on the medicinal virtues of certain species with broad scientific terms, without any resources, without any citations, without any bibliography.
The book omits the province’s most dangerous species, including several deadly amanitas and lepiotes. We have about a dozen deadly species, so [dans un livre pour débutants] we should mention them all
underlines Alexandre Brassard.
Some passages in the book are very strange. We talk about “melting the honey mushroom in water”. I imagine we want to talk about boiling, but a human being, an English speaker, would not write that way
notes the mycophile.
The author, moreover, cannot be found and the book was self-published. Alexandre Brassard suspects that it was written by artificial intelligence software.
Open in full screen mode
The dean of the Faculty of Arts and the Faculty of Sciences of the University of Saint-Boniface and mushroom lover, Alexandre Brassard.
Photo : Radio-Canada / Gavin Boutroy
The average text of humanity
This is a phenomenon that worries mycophile groups around the world. Even before mushroom identification, amateur mycophiles must now know how to identify a reliable guide among those written by artificial intelligence.
The Mycological Society of New York recently issued a warning on this subject. In the front page of her X account, she underlines: please only buy books from known authors or gatherers, it can literally mean life or death
.
Start of the Twitter widget. Skip the widget?End of Twitter widget. Return to start of widget?
The Toronto Mycological Society even felt the need to publish an article explaining how to choose a practical mushroom guide in the age of AI
.
Open in full screen mode
Mushroom picking is popular in Manitoba, but there is no local how-to guide in English, according to Alexandre Brassard. (Archive photo)
Photo : Radio-Canada / Gavin Boutroy
The head of the computer science department at the University of Manitoba, David Gerhard, has been interested in artificial intelligence for 25 years. As with any new tool, people will find a way to make money quickly
he emphasizes. There are a lot of websites that contain artificial intelligence content, recipes, guitar chord tour guides… Much to my frustration.
David Gerhard is also an amateur mycophile. In his opinion, the book Mushrooms of Manitoba is the product ofIA. Without extensive adjustment, the texts generated by the softwareIA as ChatGPT ont a certain feeling
notes the professor. It’s a little pretentious, a little tasteless. It tries to say a lot without saying anything. […] It’s difficult to describe.
The result is a bit like the average text of all humanity, the sort of thing the average person would say, which is, almost by definition, interesting
adds David Gerhard.
He gives the example of a passage which sent
l’IA : the beginning of the conclusion of the book. At the end of our investigation into the mystifying mushroom kingdom located in the heart of Manitoba, we are once again equipped with a comprehensive understanding of the complex environment beneath our feet
we can read.
The professor emphasizes that the texts generated by theIA do not always have this feeling and that it is impossible to determine with complete certainty that a text was produced by artificial intelligence.
Open in full screen mode
The head of the computer science department at the University of Manitoba, David Gerhard, has been interested in artificial intelligence for 25 years.
Photo : Radio-Canada / Gavin Boutroy
Fight againstIA with theIA
However, several firms are trying to do so. This is the case of the Canadian company Originality.ai. In particular, it performs text analyzes to determine the likelihood that they are written by artificial intelligence. The founder and CEO ofOriginality.aiJonathan Gillham, compares it to weather forecasts.
His company analyzed five passages from the book Mushrooms of Manitoba for Radio-Canada. Analysis of each passage concluded that it was likely a text written by AI, with 100% trusted
.
Jonathan Gillham has faith in this result, although he too recognizes that no artificial intelligence detection tool has perfect accuracy.
He explains that his company’s classification tool does not look for a list of perceptible traits to identify the writing of theIA. This is one of the disconcerting characteristics of AI […] the creators of these machines are unable to communicate the exact reasons why something happens
says Jonathan Gillham.
Open in full screen mode
An image search shows that an image of the book Mushrooms of Manitoba came from an Eastern European website, without giving credit.
Photo : Radio-Canada / Gavin Boutroy
L’IA d’Originality.ai examines a dataset of texts written by humans and byIA. What she learns about the characteristics of these two types of texts allows her to make predictions about the origins of writings. It’s a bit like the good Terminator in the movie
launches the founder ofOriginality.ai.
Radio-Canada requested an interview with Amazon about the measures the company is taking to combat the proliferation of books generated by artificial intelligence, especially when they pose a risk to the public.
Spokesman Tim Gillman says Amazon has content guidelines that govern which books can be sold
. He adds that We have proactive and reactive methods for evaluating content in our store, whether generated byIA or not
.
The book Mushroom of Manitoba was removed from the Amazon site after Radio-Canada’s interview request. The company confirms that it has reviewed the listing and that this content is no longer for sale.
Mycology, a haven of caution
However, at the time it was withdrawn from sale, the book was ranked second on the bestseller list in the category of Canadian practical guides
on the Amazon website.
Alexandre Brassard hopes that there should be no serious consequences following its publication. If it falls into the hands of a careful person, there shouldn’t be too much damage.
he said, noting that nowhere does the book say that a poisonous species is edible.
Indeed, amateur mycophiles normally follow a relatively elaborate mushroom identification process. I pick and eat mushrooms regularly, they are delicious and wonderful and I am very, very careful
souligne David Gerhard.
Open in full screen mode
Armillaria in a forest near Dauphin, Manitoba. This type of mushroom, prized by pickers in the province, does not appear in the book Mushrooms of Manitoba. (Archive photo)
Photo : Radio-Canada / Gavin Boutroy
He always relies on identification guides from three or four different authors. He also does a confirmation process where he tastes a small piece of mushroom and waits 24 hours to make sure he doesn’t have a reaction. Despite this process, it is limited to six easily recognizable species and without particularly dangerous look-alikes.
If a person was just getting started, didn’t know they had to confirm in three ways […] and she bought this book on Amazon and used it to identify a mushroom and then ate that mushroom, it could be very serious
he notes.
The computer science professor emphasizes, however, that he is generally a big advocate of AI. As the only source of information, it is unreliable. [Or,] it is a very powerful creative tool that opens up new ways of exploring the world
he specifies.
The big gap is that we, as a society, are used to computers being right. We need to build a new mental model where we have this powerful computer tool that gets it wrong from time to time.