I do not know the details, but I am ABSOLUTELY against ANYTHING to do with big-tech AI. This is a line I am not willing to cross, and many other users feel the same.
All I can say is that iNaturalist is the epicenter of my life and I have no idea what I would do without it, but if this organization becomes involved in generative AI bullshit it will no longer be something I am comfortable being a part of.
One of the ideas from one of their annual retreat/prioritizing/brainstorming sessions (2019) was: âInvestigate ways to capture comments and ID remarks that are useful for making identifications and including them on the taxon page and in identification tools.â
it doesnât have to be binary in the way youâre suggesting. AI could point users to existing observations where identifiers are offering up identification tips. it would just be a faster, more convenient search for things you didnât even know to search for.
I do not see a meaningful distinction here. If the distinction is one of resource utilization, I am fairly certain that AI more efficiently utilizes our limited physical resources than an army of humans trying to achieve the same task. While AI has been prone to hallucination, there have been large steps in minimizing that risk, and AI seems to be in a decent place to solve specific problems like the one iNaturalist seems to be tackling (aggregating textual data in comments and descriptions). EDIT: If youâre making this distinction because of an opinion (e.g., you strongly dislike generative AI and would never use it for ethical reasons), then I suggest being realistic in recognizing that many people will have a different ethical opinion.
I canât formulate my thoughts before knowing more about the details, but itâs worth noting that thereâs already a general distrust regarding iNat IDs among workers of small invertebrates. A lot of identifications can only be done in a lab and the only path forward with gen AI comments I can think of is more confusion.
Iâm aware iNat is inherently a tech project, but I love iNat for its community. Finding real taxonomists who provide insightful comments on my photos has been a delight. Iâm afraid that making the automated tools feel more authoritative than actual experts is going to devalue real taxonomists and alienate them further.
It might be useful, but I hope the end product comes with some disclaimers telling people it is not necessarily 100% reliable. The issue is that iNat comments and ID remarks stem from discussions about a specific individual observation, not the species as a whole. For example, some iNat users might be having a discussion about whether a particular dragonfly is Clubtail species A or Clubtail species B. Well, the appearance of some species can vary widely by geographic region, age of the individual, male or female, etc. And the iNat users were only discussing that one particular observation and photograph. My fear is that AI will scrape their remarks to say Clubtail species A always looks like this, when that might not be the case. AI is very good about making confident-sounding statements that are not always correct. You need to look at context of the individual observation where the discussion was held.
What? No there havenât - the AI hallucination issue has been getting far worse. OpenAI models are nearing full model collapse at this point, and googleâs arenât far behind.
But all that is a very minor concern compared to the ethical and enviromental issues around commercial generative AI use, which is where my primary objection lies. Even if it were a perfect and amazing tool that made everything a hundred times quicker and simpler, I would still be utterly opposed to it.
Yes, there have. Retrieval Augmented Generation, chain of thought models, models trained on specific domains, and dozens of other techniquesâŚ. I use generative AI on a daily basis (double checking it, of course) and it almost never hallucinates. I have not seen any evidence that these models are ânearing full model collapse.â
As I said, the environmental concerns are a moot pointâit would likely take more environmental resources to have humans perform the task in question. I can understand having an ethical objection, but at this point, it seems that you are utterly impossible to convince (not meant as an insult) and I will be discontinuing this conversation.
I would also prefer a more in-depth response about what the conversion of identification remarks means from iNat staff before we even go into a rabbit hole about all this ânewâ meaning of AI.
I do think this may be a good idea that comes from a well-intended thought, but I think some caveats could result in a worse end product than desired if itâs done through âAI scrapingâ of what identifiers do, as exemplified by @smwhite.
Things that contribute to constructive discussions
Try to do these as much as possible!
Share your thoughts, and really listen to those of others.
Criticize ideas and opinions, not people.
Unless thereâs evidence to the contrary, assume someoneâs making a good faith effort to express their thoughts.
Keep in mind that everyone here is a real, complex person behind their username and avatar - just like you.
Give everyone a chance to speak.
Stay on topic.
Be respectful, even if you strongly disagree.
Make sure your posts help move the discussion forward.
Accept that the discussion may need to end.
Flag inappropriate content, donât respond to it.
Things that damage constructive discussions
Moderators and staff may hide or remove posts that include these actions, and silence or suspend the accounts that posted them. Keep in mind this is not an exhaustive list.
Attacking others - this includes name-calling, labeling, insults, threats, doxing, harassment, and any other negative text aimed at others on the forum.
Bad faith arguments - this includes dredging up old arguments and not adding anything new, sealioning, dishonesty, and general trolling.
Complaining - each post in a constructive discussion should further the goal of the discussion and help us better understand each other. Donât make a complaint without offering some ideas on how to move forward on the topic.
Going off topic - short tangents emerging from the discussion are fine, but introducing or feeding into a discussion that takes the thread off track is not. If you want to talk about something different, start a new topic or find a suitable existing one.
Rudeness - such as shouting (all caps), continually posting the same argument, extremely vulgar language, or calling out specific users or obsevations in a negative fashion.
Arguments between a small group of people - these make others feel as though they canât participate. Discuss via private message rather than publicly.
This echoes the ethical concerns visual artists have regarding gen AI: credit, consent, compensation.
People are fed up by anything labeled gen AI due to the overall disregard of consent in data scraping schemes. The comments on twitter and bluesky regarding this situation have been scathing. Regardless of iNatâs intent and eventual use, thereâs a big trust issue growing here, which will put people off from using the platform.
Edit: Do note that us nature folks donât live in a science bubble. A lot of us know, or ourselves are writers, visual artists, musicians whose work have been scraped, stolen, and regurgitated by Open AI and Google AI without the three Câs above. When we say weâre boycotting Gen AI on principle, this is what we mean.
for me, the biggest ethical challenge for a lot of AIs is that they often do things in a black box way that prevents / disincentivizes human users from ever learning for themselves (because AI takes a lot of the grunt work that people would have learned from in the past on their way to becoming experts).
in the context of iNaturalist, i see lots of people just pointing their phone at a plant or a bug to identify it, without ever doing any additional verification of the AI identification. in that sort of case, itâs the AI thatâs doing the identification, and the human will never learn how to really identify from that. but if the AI also offers up text explanations of how to identify and points the human to example observations where folks have had discussion on how to identify, then i think that puts humans in a better position to actually learn something. in other words, i think that sort of thing could help partially resolve the black box problem.
done thoughtfully, i think it could multiply the power of the relatively few expert identifiers we have out there. instead of having the human experts make the same correction over and over, the AI could add some context upfront and hopefully prevent some of those mistakes in the first place.
it could also speed up the process for people who really want to learn. i know that a lot of identifiers often donât add comments unless asked because it just takes too much effort. and others are sometimes shy about asking for help. but an AI could always provide instant help, as needed.
this effort could even lead down the path of one day making the AR camera more useful. for example, if AI can learn from identification notes that within a particular genus, species X and Y are distinguished by features A and B, respectively, and it can learn to recognize what features A and B look like, then the AI can guide an observer while theyâre taking photos to get photos of feature A or B. being able to get to that level of certainty when observing / identifying could have all sort of practical applications for automated monitoring for invasives, disease vectors, pollinators, etc.
I am distressed by this because generative AI tends to be reinforced by reading its own output, or by uninformed users reinforcing it themselves. Right now, when I get the communityâs input on ids, I feel confident that itâs a self-correcting system. With AI, it risks getting swamped by nonsense that is then reinforced by more nonsense, faster than the human community can correct the errors.
I agree entirely. I immensely value the knowledge I have gained from iNaturalist, but that knowledge has come from the generosity, patience, and enthusiasm of people who have freely shared their knowledge and experiences with me, not from a proprietary software which profits anti-nature megacorporations. I have several objections:
The basis of iNaturalist is free contribution to science. The integration of large language-model generators calling themselves AI adds a profit motive --whether for iNat itself or Google-- which I reject as unscientific and pernicious.
The proposed iNat use seems little better (in ideal circumstances) than what dedicated users already do with native site tools and their own generosity in suggesting IDs and holding discussions in the comments. I trust Rupert Clayton explaining the differences between ookow and blue dicks far more than I would trust any LLM which is well known forâŚ
False positives AND false negatives. iNat has enough problems with bad data, as well I should know, because I am still trying to fix a bunch of idiotic confirming moss IDs I made when I was a brand-new user. (Thanks to graysquirrel, by the way, for informing me through patient and informative comments why mosses were tricky.) Why add more?
Even if it were 100% reliable, what is even the point? When I write comments, I am already using ânatural language.â I have had firsthand experience with my captions leading to helpful IDs from others, however jokey or sparse, because they were effective communication and IDers are often experts. They donât need to be cleaned up.
Furthermore, even if all these other issues were moot, the labor issues surrounding generative AI disgust me. Google has circumvented labor law in multiple ways for years to develop its AI tools. No institution should reward that behavior.
If your argument is that iNat benefits financially via Google dollars, all I can say is that iNat would benefit more if Google were not financially supporting and desperatelyingratiating itself with a climate-suicidalist wannabe dictator⌠whom they helped elect.
Hm. I would hope that iNat could be âone of the good onesâ when it comes to utilizing genAI - I donât think the technology is inherently evil, itâs all about the context in which itâs being deployed - but Iâm unfortunately not feeling hopeful about it. Does this Google.org Accelerator program offer funding? I can, cynically, understand participating to get some extra income. Plus itâs not like iNat isnât utilizing machine learning already; its computer vision model has been around for quite a while. So I understand doing what they gotta do to keep the lights onâŚ
Maybe I will be pleasantly surprised by the outcome of this. Weâll see. Not holding my breath, in any case.
Edit: if the ultimate goal is to assist with IDs via synthesizing information from ID remarks, comments, etc so that it can give a user info (with citations) e.g. âto ID Species A, look for Morphology B, and note that it is typically found in Habitat C,â that could be useful, I wonât lie. I just REALLY worry about accuracy. My cynicism remains, but Iâm trying not to be too reactionary here.
Some of this sounds like this feature request to make accessible ID tips on taxon pages? Maybe AI could be useful for searching for posts with ID tips and gathering them, but using AI to process them and generate the tips section automatically could lead to mistakes and confusion (especially with identification terms that arenât used in everyday language).
The ID guide page has been requested multiple times and would be a good addition to the site, but it should be editable by anyone (like taxon images) and not generated by AI.