From time to time, I’m going through observations of a moth (Hylesia nigricans) that people often misidentify. The basic problem is that the common name in Latin America (azotator, gusano quemador, gato peluda) of the caterpillar can be used for several Lepidoptera by the public, one common name includes many species. Very interesting is, that if you google Hylesia nigricans in Mexico for example Googles AI bounces almost daily between: Yes, Hylesia nigricans is present in Mexico and No, presenting some facts (but without a source eg. to a Lepidoptera authority or an expert). Whenever the Signs are on “Yes” there appears an increase of misidentification out of range on iNat. This moth is not a very important species, but I have the feeling that because there is not much information about this species available, the struggle/juggle with (non)-facts of an AI can be probably seen well. What are your experiences ?
I stopped fully trusting AI overviews from search engines a long time ago, but before I did I searched a question (not verbatim probably, because I forget): “Are male seahorse and pipefish really considered pregnant when the female gives them the eggs?” The AI overview basically said yes and that they usually give live birth. That’s basically it. The AI overview should have added a lot more nuance than that:
- in the egg-bearing animal world, the term is gravid, not pregnant (I did not know this at the time I asked)
- certain species fertilize the eggs in the female’s body and the female transfers the embryos to the male; others allow the female to transfer the eggs to the male and they get fertilized within its body
- no ray-finned fish have uteri, so they cannot “give birth” in any way; they simply incubate the embryos in a bodily cavity
- zoologists tend to classify the animal with the larger reproductive cells as the females and the smaller ones as the males, but currently evolutionary biology is unclear if this is an exception and the male in fact has the larger gametes
I also asked a similar question about a species’ range in a search, and the AI overview and it got the answer completely wrong (Eurasia instead of North America), but I can’t remember what species it was, although it was likely a vertebrate or a flowering plant.
I specifically disable all AI summaries(many instructions available online) provided by the search engine and use a AI websites & generated websites blacklist so they don’t clutter my search results https://ublacklist.github.io/docs (unless I specifically want them then I just go to the ai’s website) to avoid this exact issue.
This way I can search effectively for the information I need just as I have for many years prior, yet the companies have a need to fix what wasn’t broken and break it in the process.
Hope this helps somebody
I think, unfortunately there is little we can do about it.
We can and should of course educate people, but there’s no real chance we have against automated processes like AI summaries, especially since a lot of people do not question them and even become upset at you if they are made aware of how unreliable AI is.
Easily understandable answers that make sense if you know nothing about a topic versus nuance that you have to spend effort to fully comprehend. It’s obvious what people tend to choose. It’s a phenomenon that was widely observable even before all the LLMs existed. The problem is that now those simple answers are automated, readily available, and free.
All we can do is wait until either the LLMs will only provide factual answers (which I doubt will ever happen), or until the clutter and slop generated by them will poison the internet to the point of becoming unusable (towards which we are well on the way).
This. 100%.
This is nothing new. Before LLM many people use to read just the headline or first paragraph of a random blog and consider it factual. The more entertaining the story, the better. We can’t blame this on LLM and I envy those who were not exposed to that until now. I work in a data driven industry and yet I had to learn that the short inaccurate version is the one people listen to, not the nuanced and precise one. We keep the “technical” details and modest conclusions to “ourselves”, among peers.
People who wanted a short version right now will take one, whether it’s a LLM or a Tik Tok video. But those who want a deeper information will also have more tools to access it.
i have never and will never trust anything an AI chatbot spits out at me. i have stopped using several services when they added AI features (i’m still pissed off at iNat for the google kerfuffle and don’t plan to donate/buy merch again for now…). frequently on google/youtube services i ‘submit feedback’ saying to let users disable such features.
i’ve gotten to the point where if somebody says ‘i asked chatgpt for…,’ i stop reading/listening.
destroying the planet faster for a bunch of bullshit nobody with a brain wants.
From a (probable) technical point of view: Here I did not mention “Hylesia nigricans” and “Mexico” in the caption. In one of the bouncing AI summaries, an older iNat Forum post by me was used as one source of the AI summary, therefore I wonder wether that AI would use this post again as source, probably summing up it’s own role within this kind of probable misunderstanding (in the case this post would contain “Hylesia nigricans and “Mexico” in the title) in the future. It’s just strange to think about it.