I’m not sure that I have a particular goal in bringing this up, beyond sharing yet another astonishing example of the morass of supposed “information” being spewed up by allegedly innovative people armed with an LLM and some basic web/app development skills.
In this case, it appears there’s an app called “Greg” that purports to be “the easiest way to keep plants alive” allowing you to “identify your plants in seconds and get smart, personalized care info.” From what I can tell, the developer wrote an iterative script that ingested something like the POWO taxonomy and then asked ChatGPT a half-dozen questions about plant care for each and every species.
Thanks to this fantastic initiative, I can learn all the plant care requirements for my Sisyrinchium chaguaranicum houseplant. It needs “0.8 cups [of water] every 9 days”, “love[s] being close to bright, sunny windows” but “does not tolerate low-light” and should be repotted “after it doubles in size or once a year—whichever comes first”. All this info is for free, but I can sign up to pay for “Super Greg, [which] unlocks smart, personalized water, light, and fertilizer reminders for you and your plants.”
So why am I griping here? Well Sisyrinchium chaguaranicum is not a house plant. And it’s not a garden plant either. In fact, it is essentially unknown. It was described in 2006 from “sandy-clayish soil on slopes” at 2600–2800 m near Chaguarani, in Cochabamba department, Bolivia. The description was contained in a self published journal distributed to a half-dozen academic libraries at best (I traveled to Kew to find a copy). The type specimen was apparently deposited at the National Herbarium of Bolivia, with isotypes at the New York Botanic Garden and in the collector’s own herbarium, which was later destroyed. Exhaustive research published last year suggests that all of these specimens no longer exists. There are also 0 occurrences of this plant recorded by GBIF.
All of that is unfortunate, but not especially unusual given the history of taxonomy over more than 200 years. What is new is the wholesale invention of fake information that supposedly relates to every species under the sun. Obviously, a researcher coming across the Greg page for Sisyrinchium chaguaranicum isn’t going to give it much credence. But I fear that more plausible fakes will follow, trained on a mix of genuine and fake data, and that the task of research will become a whole lot harder.
I think even before AI, the internet has long been full of nonsense on many topics. Some people will always take the time to search out legitimate information for all topcis; some do for certain topics but not others; some never do.
Plants are funny things though, and some plants will wither with incorrect info and some will flourish. (My mother kept a discarded plant she found on someone’s trash alive for years on sheer willpower alone without ever knowing what it was or what the care might be.)
AI content in its current iteration generally sounds silly to me, trying so hard to be serious and important and relevant. In the example you offered, for example, I am imagining plants being repotted once a year every year, really for no reason at all, but by God it will be done because the calendar page has turned.
I cannot comment on if research will become harder because for that particular plant, it sounds like it is nearly impossible anyway, for which I am so sorry.
I cannot say if your concerns are overblown but I do understand them.
True. “Content farms” have generated pretty low-quality screeds for many years.* And yet, this stuff has mostly been limited to topics that would attract blog viewers to generate ad revenue. Even at less than a dollar per article, the cost of human time to generate content meant that unscrupulous publishers had to pick and choose topics and pay at least some attention to article quality.
But LLM-generated content costs almost nothing and is based only on generating plausible sounding text, not real information by any measure. I feel that’s causing a step change in the amount of false “facts” being fabricated about things that very few people care about (e.g. obscure plant species).
I guess we learned to live with all the previous challenges to the quality of published information. Maybe this is just another iteration of the panics that erupted around blogging, Wikipedia, etc. And any suitably educated person should be able to judge the quality of the information sources they’re working from. But non-human “intelligences” may have trouble detecting those signs. Perhaps our fate will be to watch LLMs tell each other increasingly untethered fictions until the supposed promise of generative AI becomes unattainable. It seems like the information parallel of the Kessler syndrome.
* Note: When writing this, I wanted to check whether I was using the term “content farm” accurately, so I looked it up on Google Search, which helpfully gave me an AI Overview, thereby depriving a content farm worker of the few cents they might have been paid to write an article about content farms.
expected to deorbit within five years even without propulsion
Does that mean the bits will crash land on Earth, to free up space for more Starlink satellites?
But eventually it comes back down to - does someone care whether that is crud or true? Do they want to find out? Had someone on FB yesterday vehemently insisting that the fallen crepe myrtle flower IS PART of the groundcover with little leaves … sigh.
On a flippant level I could blame the Geomodel Anomalies on iNat’s CV. But you told me - that ‘rare Himalayan wotsit’ is this one here. We can find and resolve those ‘Anomalies’.
The other day I was curious to see what the growing requirements were for a couple Parnassia species and Trichophorum alpinum since they’re among my favourite plants and it would be neat to have them at home. I’ve seen photos of them growing both in gardens and indoors so I think it is possible, but there’s very little legitimate information out there about their growing requirements in cultivation because they’re pretty obscure.
But as you said, there are plenty of search results regardless. The websites “greg.app” as well as “picturethisai” and “plantingo” all have auto-generated fake growing requirements for any plant you search for. “GardenersHQ” seems older and probably actually typed out by a human, but the fact that it has standardized pages for so many species and only photos taken from elsewhere makes me suspicious that they’ve never actually tried growing most of the plants they cover.
Similar to the spider website, I assume these pages are all designed to get clicks either for ads or to funnel people towards some scam where they’ll pay for “expert advice” that doesn’t exist.
By the way ChatGPT was surprisingly accurate at identifying a plant photo I tried. Claude.ai was not, it thought the Parnassia were various carnivorous plants.
Misinformation has always been around, but the nice thing about the current setting is that most people are now much better equipped to get misinformation corrected.
When I was a kid, I spent a lot of time reading books at the library. But when I look into some of those same things currently, I have the realization that almost everything I knew is actually wrong. But I still hear people saying things that were “common knowledge” 20 years ago, that have long been de-bunked.
Even in academic publications, I read one suggesting to plant honeysuckle in the the US, because it was “easy to grow and noninvasive”.
Well, mainstream press often invents news and someone claimed that a certain percentage of research is non-replicable (in certain cases invented) so, why shouldn’t AI have the right to invent things?
There will always be someone, even considered authoritative by most, who claim something that is clearly fake. It would be important that readers would have the necessary criti cal spirit to distinguish what is fake from what is genuine or, at least, to have doubts.
See the mess from the paper that made Thayer’s Gull a species. Later it was said of that research: “Smith (1966) published his now famous (infamous) study showing that kumlieni and thayeri were reproductively isolated. Pittaway (1999) pointed out that no subsequent researchers have reached this same conclusion.”
I think I have to disagree with the second half of that statement. Yes, misinformation has always been around, and yes, we have tools to combat misinformation that we did not have access to in the past. Unfortunately, though, there are (at least) two major problems with that. Firstly, the volume of misinformation that is out there has grown exponentially over the years. Debunking one false statement doesn’t take that much time and effort. Debunking millions, billions, trillions of false statements even with the tools we have is a Sisyphean task, and made much more difficult by the AI programs that Rupert brought up in the original post. We just don’t have the resources to fight misinformation at that scale, especially as it continues to be generated at ever increasing rates. The second issue is the willingness of people to correct misinformation. We’re at a point in history where a worrying fraction of the population seems to be uninterested in sorting out the truth from the garbage. Many of them even are utilizing this ‘misinformation age’ for their personal benefit. Even if we had better tools for combating misinformation available, they do nothing to help if people do not choose to utilize them. Rupert wondered if his concerns were overblown. Honestly, this is an issue that will affect far more important things than plant identifications and care tips or niche academic studies. Just think of the misinformation that is already metastasizing regarding response efforts to natural disasters like the California wildfires or the North Caronlina hurricane. If it’s not already to that point, it will become a life-threatening national security threat in the United States and a global security threat beyond that if left on its current course.
Factor in SEO, too. Everyone putting up a job posting for a content writer expects SEO expertise. But I see too many garbage articles at the top of my search results because SEO, not content quality, was the only concern. This can only lead to the eventual dominance of crud content because writers who prioritise SEO over other aspects of content will be the only ones getting the jobs.
Difficult today, almost impossible in the not so recent past. Consider that if your only source is the library, and you read a book on the topic, what you learn is at the whim of the author (especially the author’s bias). That is still the case, with the exception that it is much easier to access a wider array of resources, and making such easy access means that people don’t have to publish to make information available now.
In the past if I learned something wrong, I’d have to pretty much hope that I ran into someone who knew what was correct who could show me. Which probably means, they go to a different library.
Which isn’t really that different from what we had in the past, except then: it was only those who could publish a book that could counter a narrative.
But again we’ve always had that. Except before the misinformation was very largely coming from the top down with little that could be done to counter it (if you’ll allow me a facetious example: “of course we’re fighting WWI to save poor Belgium from the evil hun, no it has nothing to do with the fact we’ll lose our African holdings if we don’t”).
Misinformation has certainly evolved as well, but I’d much rather have the ability to easily check, and now we do.
YMMV. Having “easily” access to much “more” info is a nice thing… but is of little help if you lack the expertise to doubt dubious sources, and to reject what must be rejected. How to know? ask some chatbot?
With many scientific papers and wikipedia pages and serious-looking websites generated in whole or in part with the help of Large Language Models, illustrated with perfect-looking photos of completely imaginary flowers or anatomical drawings of gigantic rodent testicles… how to confidently assess info about unfamiliar topics?
I’ll keep my expert (human) friends and university (paper/electronic) library, thanks. (And, yes, I do know that even reputable textbooks can have typos, and that Piltdown was a fraud)
Lately, on Facebook, I’ve been seeing posts supposedly about archeology and human origins. (I don’t subscribe to these; FB just sends them to me.) I’ll see something interesting and then think, if that were real, I’d probably have see it / read about it before. I’m not sure it’s AI (or human lying) but I think it probably is. Perhaps I could track it down and be sure, but the trouble is that’s not simple, at least for me. So I’m left distrusting all I see that doesn’t fit what I already know (or think I know). That makes it impossible to learn new true things.
Oh, I do know and use some trusted sources! But that base seems narrower and narrower against all this (probable) bullshit. So yes, AI worries me. We can’t make good decisions without fairly accurate information and that’s becoming harder and harder to get.
And consider that the same tools that have been used for this explosion of misinformation, have also equipped you to meet more experts.
Prior to using iNat for example, I had significantly lower access to experts than I do today. Which now makes it so much easier to spot the less obvious errors. Many of the examples being brought here (giant rodent testicles), are only fooling the same people that were buying National Enquirer before. But before, an error wrapped in the guise of a scientific paper was much harder for regular people to catch.
I have yet to meet a single expert arising thanks to “easy online means”. All the “experts” I know (and who happen to also be on iNat) have been crafted and approved IRL at brick-and-mortar institutions (the ones that the average Joe watches disappear without lifting a finger); not by internet votes.
We do have taxon specialists who are not university educated (yet?)
Think of the ‘Danko’ brothers for flies. They were still at school when they first came to iNat. On iNat you prove your skills by your IDs, especially your comments (and your obs if you do).
Explore profiles. Some of your ‘experts’ may also be still at school. Or retired from a quite different occupation and now enjoying biodiversity instead.
