6K views on a YouTube video is nothing.
5 Likes
You know Youtube are a subsidiary of Google, right?
You are both linking to a site which is far worse in energy usage than the most popular LLM⌠and linking to a site owned by the company that you are saying not to endorse.
By linking to a Youtube video one could say you are even directly supporting Google financially in a sense - through generating watches for the video which generates ad revenue for them.
( Edit : I donât actually see any monetisation ads on that particular channel tho )
5 Likes
I agree, and also the video itself is low-effort, bad faith slop from someone who sounds like he had never even heard of iNat before the drama, and is now using this drama to farm engagement from the anti-genAI crowd. I donât think I heard a single meaningful argument in the video, just incoherent ranting and a very surface-level, unfair and biased summary of the discussion among users.
12 Likes
Thanks. Interesting and thoughtful. Crikey For just over 20 years, weâve set out to explain and dissect the news agenda for an intelligent, skeptical, socially and politically aware audience. Based in Australia so @thebeachcomber might know it.
theyâre very much a fringe outlet
2 Likes
I will take fringe over the Youtube clickbait Buy me a T shirt every time!
Its a little ironic that those concerned about disinformation from AI, spread what appears to be disinformation themselves. Hyperbolic language around unknown energy usage of this project still lacks nuance or any base in reality as far as I can see. Love to see hard evidence to the contrary though!
@spookyaranhas - Please read posts on this thread before continuing to spread this notion.
E.g. here and here
If you have solid evidence to back your comments, please add them to this thread ( the other thread is specifically not meant for these comments ). Be cool to read further counter-arguments. Personally, I still see no real evidence for these sort of claims.
4 Likes
The thing is, I donât think they know, yet. I think thatâs why these discussions are happening. I think theyâre trying to figure out what would be useful or at least harmless, meet the terms of the grant, and not alienate half the iNaturalist participants. That will take more than 24 hours to accomplish.
10 Likes
IIRC they applied for the GenAI grant months ago, and have had 3+ weeks after their announcement (already). Whatâs going on? is something going on?
4 Likes
thank you for your reply! just woke up and came to check to see if any staff had posted a response, and I see above this mostly snarkiness.
The thing is, with this being a grant, that means that someone at Inaturalist had to apply for the grant, and give Google a plan of what they actually planned to do with that grant money in a way that specifically uses Google Accelerate: Generative AI in a way that makes Google look good and uses their Generative AI in a prominent way.
I genuinely donât think whoever was in actual charge of this decision thought at all about what the people actually using the app want or need â and were completely floored at the level of backlash.
People can say the comments on the blog post were shut down to prevent server lag, but if that was the case, and the Inaturalist Staff behind the blog truely did want to get feedback, they could have simply created an official continuation of the blog post on another website, linked to it directly, and allowed people to continue to post their thoughts there. Instead, itâs been radio silence with no answers, and (almost?) 24 hours later, I still havenât gotten any actual response to my question here in the forums, and so how is anyone whoâs looking for answers on the blog and finding this official thread supposed to see any actual news?
The Inaturalist Staff just need to be completely transparent right now, and make a new public, Official Blog post on the Inaturalist blog and not here in the forums which is a seperate website and requires a seperate login for people.
If theyâre undecided about what theyâre going to be doing moving forward, they should say that.
Like I said yesterday, Iâll give it a few days, but someone deliberately had to apply to this grant under the deliberate knowledge that they were planning on using Generative AI in a large enough capacity to make Google give them grant money for it to make Google look more enviornmentally friendly as its AI datacenters pollute the world. Meta is currently planning on building an AI datacenter so large it would take up most of manhatten, and Elon Muskâs AI datacenter in tennessee is polluting the air in rural Black communities. I havenât been able to find articles in particular about Googleâs AI datacenters effects on the envionrment, but the fact that Google is bigger than both Meta and Tesla/x/twitter(?) should give a reasonable assumption about how much worse it is and is going to get, especially if theyâre being placed in other countries where they wonât get as much news attention here in the USA.
Again, Iâll repeat my request before I leave for another day or two:
Inaturalist Staff, please both here AND on an Official Blog Post, please give a clear and concise, plain language explanation of what is going on behind the scenes currently. Be transparent. Communicate with your community!
If you know what youâre going to do going forward, please explain that to the community.
If you donât know what youâre going to do going forward, because youâre in a legal bind that you have to fufill x y z requirements from the Google Grant, please explain that to the community.
Please make an official blog post, Add a note at the top that the page may become laggy due to a large amount of comments depending on the volume, donât turn the comments off at all, and also have it link to an easy to use third party website that can handle a ton of traffic and comments, so that people can freely express their thoughts on the blog post without fear of the comments being shut down for âtechnical reasonsâ.
If the large amount of comments on a single page only slows down that particular page, then leave the comments open and simply put a note at the top of the blog that some users may experience lag.
The forums are a completely different website and require a different log-in to the main Inaturalist site; not posting a public and official Blog post where people can interact with it on their main inaturalist accounts on the main site is one of the ways that this whole thing has not been transparent as Iâm going to be honest, most people I know who use Inaturalist donât even know there are forums.
Inaturalist Staff, please post an official Blog Post with the current events going behind the scenes.
2 Likes
I donât know, but another discussion on this forum is about What Would Make AI Acceptable to You on iNaturalist. To me, that sounds like either people who donât know what they want to do with it or people who have a general idea what they want to do but are trying to decide exactly how to set it up so they wonât alienate all of us.
2 Likes
The other discussion was started by an iNat user like you or me, not staff.
6 Likes
[Edited to nothing because I want to remove it and donât know how. If you think I shouldnât have removed it, let me know and Iâll paste it back in.]
1 Like
Aha! Thanks!
1 Like
iâm not sure what youâre saying here.
iNat staff havenât said anything about using AI â whether CV or some other thing â to do anything beyond guiding people in their identifications. if they wanted to do anything beyond guidance, they could have done that even with just the computer vision that we already have. are you saying that you think that with the potential use of generative AI that all of a sudden something beyond guidance is on the table?
Your issue isnât that you arenât sure what Iâm saying; youâre wondering why I said it. So why did I? I suppose because thatâs where my mind was wandering. And what I worry about. I mean, AI can compile away and it doesnât bother me as long as I donât have to deal with it unless I want to. (Make it âopt-inâ in some way.) AI identifications do bother me. For several reasons. Itâs not clear to me (or hasnât been) that AI identifications arenât on the table. (The co-pilot examples presented above are identifications with explanations.)
I can remove my post. Sounds like I shouldnât, doesnât it? But how do I remove it? I can never figure that out. Just edit it to nothing or is there a âremoveâ button?
1 Like
I agree with @pisum in that I see no reason to think that this generative AI grant demo would involve AI making identifications in the way you described here:
By which I assume you mean an âidentificationâ on iNat where it says something like â@upupa-epops added an identificationâ / â@upupa-epops suggest an IDâ, as opposed to just me adding a comment and telling the observer what species I think it is, without properly using the iNat identification interface. I also agree that the computer vision would probably have been perfectly capable of making identifications like this if the staff wanted to implement it that way, and I assume it wasnât for similar reasons to what you described. In the way itâs currently implemented there is at least theoretically some human discretion involved before the CVâs suggested is formally applied as an ID.
Something like this:
seems more likely, where thereâs text somewhere with an explanation of the CVâs proposed ID. Although other possibilities have been proposed as well.
(if you click the 3 dot menu button on one of your posts there will be a trash can delete button, but I donât think thereâs any reason to do so here since your post contributed to clarifying what possibilities are being considered with this whole situation)
1 Like
within the usual workflow of iNaturalist, i think the proper way to characterize my Copilot examples are visualizations of how the computer vision suggestions could be enhanced. within the identification workflow, i see no indication that iNaturalist overseers ever intend to use AI of any sort to do anything beyond guiding a human to make an identification.
are there other people who believe that iNaturalist overseers intend to use AI to add actual IDs on observations? (if so, i think this can be easily clarified, but i certainly didnât realize that people where envisioning major changes to workflows, and i bet staff didnât realize that thatâs what people were thinking either.)
âŚ
thank you for your detailed explanations of what your concerns are. i was trying to understand your earlier post, but i felt like i was still missing part of the puzzle. i think your (now deleted) post from earlier this morning provided the missing piece though.
just to reiterate, i think this project is intended to just explore possible improvements to the identification suggestions / guidance that people already (can) receive from the computer vision system.
hereâs how i think it will affect 3 general types of users:
- for people who currently just blindly click on whatever the first computer vision suggestion is, i donât think providing (an option for) additional text guidance with existing suggestions will change their behavior. they will still click on the first suggestion.
- for folks who currently use the computer vision suggestions as a starting point for forming their own identifications, i think the (option for) additional text helps to provide a better starting point from which those folks can do additional research. (i could be wrong, but i donât think most people looking at such text would interpret that text as making a computer vision suggestion more credible. or if this is a potential danger, certainly the text could be worded in a way to mitigate that danger.)
- for folks who currently donât use the computer vision suggestions because they donât need them or have objections to them, then they would continue to ignore them in cases when suggestions pop up. (folks using the web Identify screen should never encounter the additional text there because they currently donât encounter computer vision suggestions there. i vaguely recall iNat overseers either removing or never implementing CV suggestions there in response to user concerns that that could slow them down, and i see no reason why iNat overseers wouldnât be similarly responsive to folks who might not like to see additional text by default in other places where computer vision suggestions are provided.)
âŚ
i think you also believe that AI could present a lot of information without sources references, including information from outside of iNaturalist. while i think this is theoretically possible, and while it may be something that could be attempted in the long, long run, everything that i see suggests to me that trying to produce identification guidance using outside sources would be clearly outside the scope of this project. if iNat overseers wanted an AI to provide black box guidance from outside sources, they could have implemented that long ago by just slapping on any LLM out there.
the unique thing about using iNat identifications and comments â and why i think Google engineers are interested in seeing what can be done with them from an intellectual perspective â is that they are paired with images, taxon identifications, and other context like geography and dates; and generally, they represent good-faith efforts to describe whatâs going on in those images, using a mix of plain language and more technical language. so i think what everyone wants to prove (or disprove) here is that a small but relatively good quality data set can be turned into something super useful (beyond what can be achieved with typical training) with relatively small effort and just a little bit of help from generative AI or generative AI techniques.
(I imagine that Google probably is not interested in iNat training using outside sources because thatâs something Google has already effectively done in training their models. so thereâs nothing new to be learned by doing that.)
I can imagine that at the end of the project, iNat staff and Google engineers will write a paper about hopefully promising results of using high a quality data set, and if the results are super impressive, then Google can take that further and blast that out in broader media for additional PR beyond the fact that they supported a nonprofit like INat.
and on the iNat side, if the results are impressive, then we get better on-demand identification suggestions / guidance. thereâs nothing that says that guidance created this way couldnât be sourced back to the original comments and commenters. i think the challenge here is just how to present that sourcing in a way that is useful / digestible for humans. if a particular bit of guidance is the product of, say, 100 different notes, do you point to every single note and attribute to the commenter and observation? do you acknowledge all the commenters but point to a smaller selection of observations? etcâŚ
âŚ
i think everyone acknowledges that the result will likely contain bits of information that are incorrect or out of context. so iNat staff are going to have to define what an acceptable level of that sort of error is and whether there are other things that can mitigate it. for example, it is enough to just add a disclaimer that folks shouldnât trust everything that comes from AI? do you create some sort of feedback or flagging mechanism?
âŚ
finally, you and i are both concerned that making it easier for folks to get identification guidance could disincentivize or take away opportunities for people to learn how to actually identify on their own. while i this is something that needs to be considered, i also think about who actually needs to learn how to do things the âproperâ way. i think back to to how much work it took me to become a somewhat competent identifier for a small set of things in a small geographic area, and while i think it would be nice if a lot more people could get to or beyond that same level, and while some people who do it as a living definitely need to be able to that, do most people or even most iNaturalist users need to do all that?
if the vast majority of people would never put the time into doing that anyway, are you really depriving them of an opportunity to learn the ârightâ way? and for these people, is it better that they came away with some knowledge, even if imperfect? or is it better to save on electricity and water need to deliver that knowledge and leave them with no knowledge?
7 Likes
Iâm late to the party and havenât read this entire (very large) thread, but Iâm surprised by the number of responses saying they will go to the extent of deleting their accounts if this goes through - and this is coming from someone who was literally just ranting in a group chat a few minutes ago about how much disdain I have for people using genAI to write for them, so you know where I generally stand. I am the kind of person who literally instantly blocks anyone who uses genAI images on social media, I think the overwhelming majority of AI usage in companies these days is useless at best and harmful to society more often than not, I am not âpro gen AIâ in any meaningful way. But I do think the reaction here has been rather kneejerk.
Iâm pretty neutral on using genAI for ID labels like this. I think if the error rate was within an order of magnitude of the CV for CV-active species itâd probably be pretty useful, and I think concerns about genAI energy usage are generally exaggerated (estimates Iâve seen are very roughly order of 0.1-1 Whr per query, which is way less than charging a phone). My portable air conditioner, which I essentially need to be able to sleep in the summer, uses orders of magnitude more energy than that in very little time. I think genAI is doing immense damage to our society, I just donât think ID assistance for common species and the energy equivalent of running an LED lightbulb for a minute or so is particularly responsible for that damage.
I guess Iâm opposed to iNat implementing this idea on the grounds that it would piss off a substantial amount of the community and the benefits are absolutely not worth that, but I donât see any fundamental downsides with the proposed plan if that wasnât the case. If iNat was doing something with say genAI images (a technology that I firmly believe has nothing but massive downsides for society as a whole, and a technology that I would happily wipe from existence if given the opportunity) I would have a lot more hardline stance on this.
A good compromise would be to use a wiki-style resource without genAI as many have suggested, and I have no doubt that many here would happily contribute their time to that (I certainly would)
8 Likes
is quietly getting on with observing and identifying and annotating and curating.
Those who are active on the forum are a small slice of iNatters.
7 Likes