What is this - iNaturalist and generative AI?

[dhobern]

@tiwane Sorry if I’ve missed something. Back in June, many of us signed up for a webinar/information session that was being planned to present the next steps and how this was expected to pan out. I registered, but I don’t think I received anything more. Did I miss it, or is it still planned? Thanks.

[stockslager]

if this is the blog post, why is anything that is said in it bad?

i’m interested in many of the things I observe. By “interested” I mean that I’m a passionate hobbyist. I often click on the ID’er to read their profile. If they seem to have a lot of professional expertise, I’ll often wonder how they were able to ID what they ID’d… So I’ll Google something like “palpada albifrons vs. vinetorum”. Google usually provides an “AI Overview” that shares the diagnostic traits distinguishing one vs. the other. I have no idea how Google assembles this data, but it allows me to consider how the expert might have been able to ID to species. Sometimes I’ve begun looking at the AI overview, deciding if the diagnostic traits push me toward the “most likely” species returned by the algorithm. Ironically, I’ll be less apt to take the algorithm’s suggestion if this is implemented. This is because it’ll be one technological source corroborating the ID rather than two. Since Google and iNat are now separate, I look at Google’s AI overview matching what the algorithm returns as two reasons to take the suggested ID (if I can see what they’re saying distinguishes the two species).

I don’t see how sharing information that many users already look at directly would hurt anything.

[anfra1969]

@stockslager

I don’t see how sharing information that many users already look at directly would hurt anything.

As stated earlier, there is a difference between doing this for the 1st place ID proposal or using it for the 2nd step confirmation, pushing the ID to research grade. Today, no AI is considered a subject matter expert for multiple and often very systemic reasons which were already explained before. Therefore, it cannot be taken for a reliable source of information. And if you need to use AI, you are no expert either, so please don’t do confirm the ID.

Also, even if today some people already may apply AI as unreflected information source for ID confirmation, it hurts, because it’s degrading ID quality. It would hurt the more if gen AI was officially included and advertised within the platform as you propose.

@stockslager

I have no idea how Google assembles this data, but it allows me to consider how the expert might have been able to ID to species.

I am pretty sure also Google cannot answer the “how” in relation to a specific case. And that’s a large part of the problem. Biologist experts have developed dichotomous keys to distinct between species of same genus and that’s not what AI uses. You find them often in internet in expert forums when searching for a specific ID proposal. It’s hard work reading them and in consequence it seems mostly beyond horizon for today’s chatXYZ engins. The usual answer I get from AI when asking for a species is a list of features. It does not tell me that there are three other species which are undistinguishable from the first without dissection of genitals. If i ask more pushing, it always tells me then not to ID beyond genus, which sometimes is justified but sometimes is not. All i see is a big IT gap, not an opportunity for the user community.

@stockslager

Since Google and iNat are now separate, I look at Google’s AI overview matching what the algorithm returns as two reasons to take the suggested ID

Your hypothesis of independence might be true today for different models like CV and LLM. But for how long, and who will inform you when it changes? If it’s two LLM systems, who tells you they don’t feed each other even today with data, creating new mermaid and gargoile like creatures which are as much real to the AI as lions and bears?

And is two times a similar and independent electronic dream a true reason to believe? Some species are rarely reported only due to a lack of experts in numbers, being able to do a dissection and provide a positive ID compared to another very similar species which is GENERALLY BELIEVED to be more common.

[stockslager]

@anfra1969

Your hypothesis of independence might be true today for different models like CV and LLM. But for how long, and who will inform you when it changes?

This is a fair point but my belief is that iNat sits in a better position to understand the independence (or lack of independence) of counter parties than I do. They’ll also sit in a position where they could (if they chose to) give a broad indication of how their counter parties derive their information being hosted on iNat. In my view, the more counter parties the better. Assuming (as you point out) that each is trained on a unique dataset that was itself assembled from a unique community of experts. I’d love it if there were a way for different providers to plug-in to iNat and share what species they each think is most likely. This would place the value where it rightfully belongs… on the humans here whose discernment would sit atop all the technological providers. It doesn’t bother me that they want to integrate with Google. I hope they integrate with others too.

[anfra1969]

Two (or more) blind robots were sent to the wild, expecting they can compensate each other’s weakness and provide lessons the deaf.

It sounds to me like a joke. Maybe a starting point for a Scify novel.

P.S.: Read Isaak Asimov. He had a PhD in Biochemistry, a general interest in mathematics and AI and is one of the big three of US American Scifi literature. Most of its novels are about AI. For me an evergreen best entrance point to ideate about human-machine interaction and its consequences.

[mo0nsgreenthumb]

Thank you for this, will have to read through it.

[zoology123]

Yes thanks for bringing this up, seems like radio silence? I also haven’t heard anything about it since it was announced.

[tiwane]

@dhobern

Back in June, many of us signed up for a webinar/information session that was being planned to present the next steps and how this was expected to pan out.

We still plan to have one when we have something concrete to share and we’ll be in touch when a date and time have been determined.

[anfra1969]

An important question regarding the possibility of qualifying / revisioning gen AI output is a non-technical one: How willing would curators and subject matter experts actually be to do that ?

My impression is that it’s already hard work to keep the human output in this platform at a tolerable level. Not sure that training gen. AI to expert level was ever before seen a reason for them to engage.

[anfra1969]

I found the following article on the NABU page, a German sister association.

It seems they have already invested a lot of thinking on the differentiation between “validation” from “plausibilisation”. As far as i understand, our quest here is about the interference between gen AI with the latter , how trustworthy it is and how limits and thrash holds can be defined.

It’s a lengthy article but maybe worth reading to sort things in a good way.

You need let it translate…

https://nabu-naturgucker.de/meldeportal/plausibilisieren-von-naturbeobachtungsdaten/

[joegiulian]

I’ve been following this thread since day one, upon the “alarm’s” sounding.

A number of comments I wrote entire responses to only to delete before posting; before replying, I seriously consider whether those I disagree with are 1) worth engaging with based on their tone and 2) capable of compromise, anyways.

I wanted to reply directly to this one because I hope others will critically evaluate how realistic, mature, fair, and patient their perspectives are in this matter. After all, what is owed to (what I hope all pragmatic persons can see) an extremely benevolent, well-intentioned citizen science organization, with limited staff and volunteers? I know my answer.

I could shed a tear for the National Museum of Brazil fire in 2018; countless biodiversity data lost forever; it too was probably full of uncurated materials, just as iNaturalist and every other major natural history collection. Who could even fathom what was lost?

Now, could I shed a tear for iNaturalist users who delete their accounts over an announcement portending the possible expansion of AI in ways they are unwilling to compromise on? I cannot. I surmise their contributions are just as careful as their activism. For every almond that takes 1 gallon of precious Californian water to grow, does anyone shed a single tear any more? I don’t like how almonds taste, I would like them deleted from this phase of the Anthropocene, perhaps someone will help with that.

[dhobern]

Thanks - I appreciate the reply.

[jenn1199]

My first thought is what would the environmental impact be if iNaturalist did use AI?

[LuBu]

It would be pretty swell if the people over at team iNat could get (it) together and put out a statement that said either:

“Despite the community pushback and concerns brought against our announcement to impliment generative AI into our platform, we will be going forward with our plan”

or

“We have listened to the concerns of our userbase and found that to move forward with Generative AI in our database would compromize the validity of our data, as well as go against our values as a company, and therefore will not be moving forward with the plan to make Generative AI a part of our site”.

Instead there’s an ongoing dance around the question of “Is this gonna happen or not?”. Instead of changing the goalposts from what was initally announced, pick a side and stick with it. Make your final decision clear to your users. You owe that much, considering how you have taken so much in donations, publicity, and content, and labor from the community. No more being wishy-washy. Show us where your values lie.

[natev]

I respect the iNaturalist Team’s decision to not brashly answer the concerns of a vocal but likely non-representative subset of their community. Taking some time to think things through beats making a nice sounding PR statement that they may later have to recant.

To my knowledge, there is not a mechanism for the proposed project to “compromise the validity of our data” and iNaturalist has made it pretty clear that generative AI/LLMs is actually in line with the core values and mission of their nonprofit. Keep in mind, they pioneered the computer vision tool for general identification of living organisms, something that has been critiqued heavily through the years and has driven knowledgeable users off of the platform. And what they are proposing is extremely unlikely to be environmentally harmful according to peer reviewed research shared multiple times in this long thread, would not compromise iNaturalist data, would not give Google any power, would not necessarily produce AI slop, and has a lot of upsides too, which are being woefully neglected in this discussion.

I have read this entire thread and have found that every objection to this project fails on logical grounds, mostly non-sequitors and other types of logical fallacies. That isn’t to downplay personal perceptions, opinions, and conclusions, I’m just observing that the dissenters have given iNaturalist staff no reason to cancel the project other than threats. And, with a few exceptions, those threats are due to lack of knowledge of the science around the environmental impact of LLMs and AI hallucinations. I genuinely hope that I do not come across as rude here, that’s not my intent, I just feel the need to defend the iNaturalist team from unjust allegations based on the misunderstandings of well-meaning iNatters with whom I hope to reason.

[jdmore]

There has been no

@LuBu

announcement to impliment generative AI into our platform

There is no

@LuBu

plan to make Generative AI a part of our site

So neither of your proposed announcements would be relevant. Please read what Staff have actually said in this topic and in the original Blog post.

The only announcement has been that they will be exploring ways to use AI (in some form) to surface and aggregate existing helpful ID information, and producing a demo or prototype of the results. Any incorporation of that prototype into the production iNaturalist site would be a separate and later decision.

@LuBu

Show us where your values lie.

What values are you looking for? Disregard an opportunity that might help support and improve the site, and instead submit to the immediate outrage of a small subset of site users? Make an announcement about some imagined plan when said plan doesn’t yet exist?

I could see room for skepticism about “values” if iNaturalist was a for-profit company trying to maximize financial gain for its staff or shareholders. Knowing that the people who founded the site in 2011 are the same people who still lead the organization, I have no doubt that the “values” manifested in their ultimate decision about whether to incorporate their AI prototype in some way (or not) will continue to reflect the original mission and vision of the organization.

@LuBu

You owe that much, considering how you have taken so much in donations, publicity, and content, and labor from the community.

Participation in the iNaturalist community always has been, and no doubt always will be, completely voluntary, with no expectations or strings attached.

[DianaStuder]

Secondly, you could read the previous comments about this

@tiwane

I’m not an AI expert by any means, but my understanding when it comes to energy use is that enormous LLMs like ChatGPT or Gemini take a lot of energy to train and to use. They train on datasets that are immensely more vast than just the comments on iNat, requiring a lot more energy than an iNat model would need for training. Also, they use a lot of energy when generating responses to queries. For the scenario I outlined above, the model would not be generating new responses for user queries (there would be no text-based query feature like with ChatGPT). Like the geomodel, it would generate pages each month that would be a stable resource people would access for each taxon, cutting out the energy required to generate a new response for each use.

[adrien_ferrand]

I’m back to this topic after a some wonderful hikes in the few weeks we call summer here. Nature is beautiful out there, and some great encounters!

I’ve found some interesting posts in the Forum that shed light on unusually cute creatures I’m now thinking of all the time. As a result, I’ve been reading lately about diptera and seeing with fresh new eyes something so easily deemed banal. This is what I love about the iNat forum, bringing together a collective of people passionate about nature. Before that I had an interesting exchange about sponges from a diving observation, I’ll influence my next dives!

I thought I might remind some of you what else is out there. iNat is a community but not a Facebook rant place about opinions.

I’ve finally convinced some new people who love nature to give it a try, and they love it. A 70y old new user is using ChatGPT to help her simplify the taxonomy and not get lost in the terms. I found that a great use of this tool.