What is this - iNaturalist and generative AI?

General
168

Thank you for your reply, it is good to see some contrasting arguments on the environmental costs of generative AI, as I admittedly don’t see many arguments towards environmental benefits of AI. I read through the paper you linked, but I can’t say that it has swayed my opinion. I found its quantification and comparison of human and AI models to be quite odd, in fact.

For example, part of the study quantifies water usage between an LLM and a human to write 500 words. It concludes that in writing 500 words, humans consume more water than an LLM. I honestly cannot tell how it is quantifying that- whether it means water to cool our computers, or water to generate electricity at a power station, or literally that writing 500 words requires energy expenditure and we need to get up and drink water to sustain ourselves afterwards. Either way, in this case I don’t think that using an LLM instead of human labor to write 500 words would actually reduce water usage, because that human would frankly still exist and would still need to consume water. The LLM is not replacing the human’s existence, and if the human is not spending its time writing those 500 words, it’s probably just writing a different 500 words. Instead the water consumption from that LLM is additive, such that water is now consumed for both the human and the LLM.

I believe my argument can be generalized to their analyses of carbon emissions and energy consumption as well. If an AI performs a task instead of a human, that human will still exist and will still be using energy and emitting carbon. If iNaturalist chooses to implement this feature (and I decide not to leave the platform), then I will probably still be online vetting the results of that AI tool. Now I am using the same amount of energy and emitting the same amount of carbon as I would have done otherwise, but there is also an AI tool using additional energy and emitting additional carbon.

I’m sure this comment is more than 500 words, so I apologize sincerely to the environment and my peers for the amount of energy I have used, the carbon I have emitted, and the water I have used. In fact, even more water than others, as I am currently sick and drinking much more water than usual. In my defense, I would have been on my computer anyway, likely using the same amount of energy, emitting the same amount of carbon, and drinking (or using?) the same amount of water.

169

My little pet peeve with LLMs everywhere is that, when it generates, there is no identifiable individual behind the words, no matter how great or useful they are. You have a beating heart, and so do I, and that connects us in ways that AI cannot do. Plastic flowers are beautiful and don’t die, but they’ll never replace plants in my garden.

I am happy to see how many humans are here agreeing and disagreeing on the same forum, caring about this thing. I am inspired by reading many of these comments. All full of heart! Look at the love for nature and education @ryan84 , @graysquirrel, @natev and so many others put in the comments.

I believe we have reasons to be optimistic. iNat won’t stop being cutting edge in tech, and will keep caring for humans and nature for the foreseeable futures. Tech updates (e.g. a nicer CV algorithm or a sleek iPhone app) are great, and the team may make some wrong turns in the way. Maybe big tech partnerships are needed for the mission, like the Google Maps, NVIDIA and AWS ones seem to be, but always making sure decisions are made toward empowering and respecting humans.

The iNat team is sure to be following closely these discussions, and I am sure they are very carefully planning responses to all of this. I would say it is good that there is no detailed response yet, it may very well be a sign they are taking adequate time, so everything is clear and well explained.

(oh, @spidercat commented! I am really enjoying their comments, full of haemolymph-filled heart and wittiness.)

It would be great if the team ensured that a good deal of this Google grant is towards funding a community-driven GenAI policy for iNaturalist, getting feedback, listening to people, running Q&As, AMAs, and the like. Running far-reaching surveys, community calls, and fostering spaces for diverse voices to be heard. Seems like we would not have a shortage of participants.

170

Hi everyone, many have suggested a community-driven wiki summarizing identification characters, keys etc. That would be certainly one way to go.

Maybe the iNat team could look into alternatives, e.g. modern approaches how character tables can be created by (non-expert) communities, which are then used to guide an identification process. There’s an app that does exactly that, though the interface is in German

However, there are a bunch of papers explaining the approach (and testing the success from an educational standpoint, which is what ultimately matters here):
http://dx.doi.org/10.12691/wjce-8-1-3
https://doi.org/10.3389/feduc.2023.1190462
https://www.researchgate.net/publication/327816613

The app “ID-Logics” has its home here: id-logics.com – Du kannst Natur.

Rather than engaging in another questionable partnership with a Tech Giant, I’d love to see iNat exploring other ways of engaging with the community - be they experts or newcomers, and finding ways that are centred around humans curious and eager to learn.

My 4 cents…

171

Thanks @arboretum_amy, that was my mistake and I’ll change it back to “grant”, which is accurate according to our Head of Development. I know changing it back and forth doesn’t engender any more trust here, but I want to be transparent about the mistake, which is all I can do.

It’s a good example, though, of why I or other staff haven’t responded again with more details. We want to make sure that what we say is accurate when it’s posted. We’ve been doing a lot of listening to the community and having discussions on our end, and we’re working on a response that should answer many of the major questions people have and provide more clarity. I again apologize that this was not ready and available when the announcement was made.

I want to thank everyone who’s shared their opinions and thoughts, I’ve been reading them as they come up and they’re very thoughtful and honest. I really appreciate them.

172

Thank you. Even just this message lets us know that you and the other staff do indeed care.

173

I gotta say, I do not envy you right now. Thanks for doing what you can and good luck over the coming days. Hopefully cooler heads will prevail on both sides of this, and the end result is something that drives away a minimal amount of users (preferably none, but I think no matter what the outcome is, a handful of users are gone for good, unfortunately) and proves to be helpful to users both casual AND “power” (powerful? I like the sound of that. muahaha!).

Also, unrelated to much of anything, but just wanna say: appreciate you, @tiagolubiana! It’s nice to feel seen. :grinning_cat_with_smiling_eyes:

174

I have over 800 videos and 2000 photos, which will equal somewhere in the ballpark of 500+ observations, sitting on my computer waiting to be processed and uploaded to iNat. I’m not going to bother going through them until you change your stance and clarify what is going on. It’s your move, iNat. You’re making me withhold valuable data because I no longer feel comfortable sharing my observations with y’all.

175

It’s interesting to see how different the response to the subject of generative AI is from different people. There seem to be information silos, where for example my impression is that people on Bluesky as a group tend to be pretty closed to the concept whereas on Twitter people tend to be more open to exploring their potential. You can see that here with conflicting beliefs about the environmental impacts (and assuming that everyone shares their belief), although maybe less differences on the ethics.

I appreciate the clarification here; I was confused about people concerned that their images and comments might be used for training, as I had kind of taken for granted that iNat content was probably already being used for AI training of some sort, somewhere. Other photo-storing websites like Flickr and eBird/Macaulay Library prevent downloading of users’ images, and other social media platforms like Instagram and Twitter prevent you from browsing user content without logging in. iNaturalist has no restrictions like that, so anyone on the internet can easily find anything on iNat and download it. There have been a least a couple examples of scam websites claiming to be selling photos stolen from iNat. But I suppose that’s a different subject than explicitly giving someone permission to train on content.

177

It would have been wise to write an extremely detailed, careful, and considerate blog post from the start, but I can wait a few days for it. Even if you manage to do good damage control, and it turns out this deal isn’t as catastrophic as it initially seems, the first blog post was grossly tone-deaf to the social consensus of iNat users towards LLMs.

I understand that using an algorithm to collate comments on a taxon would be highly useful – but to take the next step of allowing a black box to then summarize it? no way.

My suggestion is: instead of worrying about keeping up as iNat gets bigger, stop getting bigger. You are doing fine. You’re falling into the trap of needing more, and more – infinite expansion is an illusion and always fatal.

178

No doubt it is challenging. It’s why I said in an earlier post that I do NOT envy iNat staff right now, especially public-facing staff, who have to take the brunt of the public response. I cannot imagine the stress. :sweat_smile:

Your linked research is interesting, but I am scratching my head about your bringing up the gender gap. I’m not sure how relevant that is here, since a lot of us are anonymous/gender unknown (don’t assume anything just based off emoji choice! we contain multitudes! :wink:). But maybe you meant it as a “huh, interesting” aside, which, as a nerdy person, I suppose I can understand, lol. Still though! A bit irrelevant IMO.

I stand firm in my position that this is - among other things - a communication issue, and that I hope iNat learns from this.

179
  1. “AI expert” is a bit of a wastebin taxon. AI can mean anything from data analysis algorithms to the LLM/genAI mentioned here. I’m not sure we should take the word of an article contributing to that exact oversimplification. In addition, people who work with a technology are usually going to be more in favor of it because they have a financial incentive in it being utilized more.

EDIT: I took another look at the article you linked and the methodolgy for choosing rhe experts was “people who presented at AI conferences”. I have even less faith in it than before. People who aren’t optimistic about AI wouldn’t go present at an AI conference…

  1. The implied sexist rhetoric isn’t helping your arguement. Why would opposition matter less if the opposition was mainly women? Can they not have opinions on technology?
180

The gap between men and women?

I see a gap between USA dealing with No Kings Day on Saturday - and venting at iNat. So many iNatters on this thread who don’t usually bother to comment on forum posts.

Can we not wait and see how this plays out? iNat and this forum are already public. The example in the blog post with the frogs looks promising.

I wonder if iNat plans to roll this out in test batches - as they did with that - checking the quality and accuracy of IDs experiment. You are invited to … tell us if this is an effective way to distinguish between Frog A and Frog B.

181

I’m also baffled by the inclusion of the bit about gender. I don’t really see how it’s relevant so it seems odd to bring up, but at the same time I have to assume women are less enthusiastic about AI due to some of the more unethical uses I’ve seen that have targeted women.

That being said, I’d also like to note that the article’s definition of experts was people who work with AI. In other words, people who are already invested in the technology and more likely to have a reason to want it to be used more, given that they’re the ones developing it. To me saying “experts are more enthusiastic about AI” when what’s actually happening is that the people developing AI models are more enthusiastic about it seems misleading at best.

182

Exactly what I was thinking about the energy usage thing. It’s a data point that makes no sense considering people are still going to be here fixing the inevitable mistakes the gen AI are making. We’re not being replaced.

And another thing not brought up is that gen AI is adding to our existing energy costs. New power plants and data centers are being built, which obviously is not great for the environment and is certainly more costly and damaging than building an apartment complex for 500 people.

Me going on iNat to help ID stuff is no more energy intensive than me deciding to watch TV or play a video game. But adding AI servers and buildings for it is additive, it’s on top of us existing already.

Frankly the whole “more efficient than 500 people” argument is moot when you have to use considerable energy to build for it. It’s similar to the argument of buying a brand new EV vs buying a used hybrid car. It takes YEARS to offset the various costs (not just monetary) of building a new EV, vs buying a used hybrid that’s already built and is able to immediately save energy.

183

I think one thing worth noting when it comes to using generative AI to give an explanation of why the CV thinks something is a particular species is that the CV doesn’t actually use the same features identifiers do.

Analytical AI like the CV has analysed a dataset and figured out that certain coloured pixels in certain places correlates with something being labelled as a particular species. It doesn’t “know” what the identifying features are the way a human identifier would. It “knows” statistics and correlations.

Think about how many times the CV has suggested that something is a particular bird or insect when the photo is of a plant with no animals visible, because it usually sees those species in trees or on other plants, so it’s learnt that the presence of those green pixels often correlates with that animal being there.

Using generative AI to have it say “I think this is this species because it has these features” in no way reflects the actual logic it’s using. It tells you nothing about how it actually came to that conclusion. It’s just put words together in a particular order because it’s learnt that when that species is mentioned, these words tend to be said together.

Maybe this isn’t a big deal compared to the other issues being brought up, but I keep seeing people saying they’d welcome a feature that says how the CV come to the conclusion it did, and I think it’s important to recognise that this would not at all be doing that

184

I would additionally suggest two things:

  1. please be proactive in flagging inflammatory comments, it makes moderating such a fast-moving and heated thread easier.

  2. you can use word search in-browser to find out if people have already said what you were gonna say. again, to keep the thread from becoming too unmanageable.

185

I’ve spent the last 3 days walking around in the heat hanging flyers to raise awareness for the protests. And just now I discovered that somebody has ripped down half of them already. I’m at the very ragged end of my emotional rope, it feels like the entire world is burning down, and now the only refuge I have left - iNat - appears to be preparing to align itself with something I find absolutely abhorrent.

I’m just so tired.

186

But that is a good example of something it would be useful to spell out in so many words - this caterpillar is often found on this milkweed. It remains up to the identifier to decide, no caterpillar here, but, it is that milkweed.

What can I say? I stand with you across the Atlantic Ocean. Take care.

187

Different people read things differently. I saw this as conceptual, not as implementation detail.

I think this answers the question. They want to “use GenAI.” They didn’t say they want to turn the site over to GenAI. And they started hinting at one detail of exactly how they would “use” it. I agree with you and everyone who has said that more detail is needed. And that is exactly the reason why I am pushing back against the doomsayers and the account deleters.

Maybe I am guilty of only reading into the blogpost what I want to believe. If that is the case, then I am feeling very cozy as a member of this community right now…

188

I see what you’re saying, but LLMs work by learning what words often go together in certain contexts. In this case, it would most likely learn that when identifying that caterpillar, people say “I can tell it’s this because it has these markings” so that’s the information it’s going to give. Not at many people are saying “I know it’s this because it’s on milkweed” so that’s not likely to be given as an explanation.

Once again, the AI doesn’t work using the same logic as us. It may be basing the ID on “there are green pixels here” but it doesn’t know why those green pixels mean that species might be there. It’s just pairing common combinations of words with an ID with no connection between how it came up with that ID and how the comments were generated