Option not to train the AI?

why are you trying to avoid training the AI? attempting to forestall the singularity? :robot:

6 Likes

Some observations can be identified and confirmed without the image representing the species to the automatic eye…it requires more info and an understanding of the context or some other factor to interpret it…eg a leaf peeping out from under other foliage

1 Like

I don’t know why exactly @questagame asked… But I guess this is a good question if the concern is that the photo is poor enough to mess up what the AI might have learned well, even if the observation is easily capable of reaching RG. For example a well-known and readily identifiable creature, but well-hidden in the uploaded photo.

2 Likes

I can see the use for this on a per photo basis. I often upload spectrograms alongside images and sounds of birds. I’m not sure if the AI training on the spectrogram image is helpful without standardized scales for time and frequency. Disabling training on those spectrogram images but allowing it on the regular photos would allow the AI to get training, without adding garbage data.

Thanks for the replies. Sounds like the answer is “no”? I think the question then becomes “why not”? Or “why wouldn’t you have it?”

It’s easy to add, no? QGame has had it in its “settings” since day one (screenshot below). Best practice, as you know, is generally to give users’ agency, awareness and data control - or am I missing something? If it turns out to be superfluous, and no one uses the setting, well that’s great, but why not let users decide?

1 Like

i suppose every system is allowed to decide what’s important to them and define their own terms of service. for example, iNaturalist seems to give quite a bit of thought to copyright and letting users define their own copyright terms, whereas it seems QuestaGame simply says that everything submitted through their system will get a CC-BY-NC license (although strangely, it creates data in iNaturalist with CC-BY licenses on the observation and CC-BY-NC-ND on the photos, though that’s probably off-topic).

it doesn’t seem like iNaturalist specifically says anything about AI in their Terms of Service, but it doesn’t seem like it’s doing anything with its AI that appears to be nefarious or unethical or even controversial. it’s not like the case of IBM’s facial recognition AI (see https://abovethelaw.com/2019/04/in-your-face-how-facial-recognition-databases-see-copyright-law-but-not-your-privacy/), where there was a clear privacy angle to consider. (i suppose, though, there’s an argument that could be made that there are greater considerations than privacy: https://www.vice.com/en_us/article/59ydmx/copyright-law-artificial-intelligence-bias.)

so personally, i don’t see why there needs to be policy around opting in to training of the iNaturalist AI. (but maybe i’m missing some angle?)

that said, there does seem to me to be a little bit of a gray area regarding whether iNaturalist should use photos / observations to train the AI where the observer has reserved all rights. the TOS does grant iNaturalist certain rights regardless of license to do certain things, but it may be a bit of a stretch to include AI training into that list of allowed things… but that may be starting to stray off topic…

a couple of other wandering thoughts… the questagame_bee user that’s the curator in the QuestaGame project has a really creepy looking profile picture (in my opinion). you may want to change it. i assume it has access to see unobscured coordinates, and i assume you’re using it somehow to feed data to the QuestaGame Bio-Expertise Engine. i didn’t check either way, but you may just want to make sure obscured coordinates aren’t accidentally revealed somewhere in this chain, if things are working the way i think they might be working… oh, and this post also seems to talk about some sketchy practices: https://forum.inaturalist.org/t/reducing-sightings-to-questagame-group-account-and-update-on-sync-account-system/2432.

5 Likes

ok i edited my post because arguing with questgame guy is probably not a good use of time but i think it’s iportant to consider this guy’s posts in context of how Questagame has used the iNat community. Makes it a bit off, if nothing else. But we’ve all gone through this before in the Google group

2 Likes

Thanks @pisum for this. Yes, it make sense to add those copyright options in the settings in QG. It’s a good idea and we’ll see if we can implement in our next release.

Re the questagame_bee user, where are you seeing this? Could you point where to look so we can address it? (Maybe in a different or private thread?).

Not sure I follow your point @Charlie, but QGame is posting a subset of its sightings on iNat as a “Group Account.” We understand there are iNat users who didn’t like this (you?) and that it can run counter to the “social networking” culture of iNat. (We were not fully aware of this culture when we first started using the APIs, but quickly learned).

But we also understand there are other people who DO like it; we get a lot of good feedback, and the data exchange has shown real value.

Recently iNat proposed what seemed like a good solution, which we tried to implement, but it did not work from our side.

We’ve recommended (and specialists in the field have suggested the same) that iNat remove “Group Accounts” from its leaderboard, and allow its users to decide if they want to filter out “Group Accounts” - e.g. “don’t show me Group Accounts.”

Would this solve the problem, or is there something we’re missing?

When it comes to AI, wonder if the same solution seems logical? Some iNat users (like yourself?) may not think iNaturalist should accept observations from users who don’t want to share their images with AI training systems, correct? So would it also make sense to allow iNat users to decide, for example, “show me all observations” or “only show me observations from users who allow AI training?”

We don’t have all the answers, but are looking for solutions - hence our post about the “whys” and “why nots” so we can better understand the reasoning here.

Thanks!

You can’t stop publicly posted photos from being “viewed” by AIs. You might be able to convince people not to share data with a species id AI but the actual threat - the big scary googles and Microsofts and tyrannical governments… can see it. Might as well use it for something valuable too.

In terms of QG you were using the inat community to generate IDs for your (maybe for profit?) project without being forthright with the community about it. That’s an ethical issue. The algorithm isn’t.

1 Like

@questagame – what exactly is the point of this thread? why not simply discuss and work with iNaturalist staff on this? the way i read this, you’ve made a promise to your users that you realized you can’t keep, and now you’re trying to convince us (without evidence / support) that there’s a problem so that we in turn will pressure the iNaturalist staff to make changes, hoping that that somehow will save you from having to go back and tell people that you overpromised.

speaking of promises, each time you’ve promoted your product here in this forum, you’ve talked about how you want to fix the way you’ve integrated(?) your system with iNaturalist, and yet months later, all we keep hearing are regrets that you still haven’t been able to fix it but that you are still committed to addressing past mistakes and bad practices. at what point will this promise become something else?

6 Likes

I’m more-or-less dedicating myself to staying out of the discussion but I will say that QG does not appear to be “for-profit”. They only earn money from grants.

AI use like this is new, so I guess the “privacy” discussion of whether you can decide not to let AI use your images is a developing issue/topic that QG wants to inquire about here. Perhaps not pertinent now, legal-wise, but maybe down the road we will actually have to consider this. I hope not though.

@pisum - really keen to address the issue you raised about the questagame_bee account; can you explain where to look? Not sure we understand what you’re referring to.

I don’t know if there’s a formal process in iNat for requesting new features - is there? ( have discussed with staff) - or should we just move this to “feature request” channel? (Thought it was in that channel originally, actually).

No, nothing to do with trying to convince you there’s a problem, or any promises. Easy for us to not share sightings that have “use for AI” switched off.

Great you have an opinion on this @charlie People have lots of different opinions. Which is a good thing. Settings options allow people to have different opinions and make decisions for themselves.

Potential downsides, from what I can tell, is (a) don’t want to paralyse users with too much choice, especially about trivial things. I’m not sure if the AI option would do that or not, or (b) it could be that the resource required to implement the feature is prohibitive. Again, not sure about this.

hey questagame guy, if you’re not able or willing to address valid community concerns that keep being brought up without dismissive language that says nothing, you aren’t ever gonna get buy-in from the community on your… whatever it is… you are trying to latch onto iNat with.

this is how i feel about this whole thread.
:man_shrugging::woman_shrugging:

if any of the things in this list are false, then there’s nothing to worry about. otherwise, you might want to fix something.

  1. questagame_bee can view the true coordinates of obscured observations in the QuestGame project because it is a curator in that project
  2. you’re using questagame_bee to feed data over to QuestaGame / BEE
  3. one of the data points you’re feeding is location
  4. the location you’re feeding is the true location (as opposed to the obscured location)
  5. you’re not doing anything else on your side to obscure the true coordinates before displaying them in QuestaGame / BEE
1 Like

he doesn’t like the AI so he is trying to continue doing what he is doing here but not contribute to AI training with questagame observations, i think.

You’re welcome to create feature requests like any of us through the forum’s feature request section. They need to be submitted for approval by moderators/staff and worded as requests not questions or debate proposals.

3 Likes

Thanks @mira_l_b - appreciate this. Okay, will do.

Hi again @mira_l_b - thanks again, just reading through the notes on “Feature Requests” - it says:

“We are likely to not implement any large changes that are not already on that agenda until next year, so it would be best if you started a brainstorming topic in General rather than make a Feature Request.”

Seems there are two issues coming out of this thread:

  1. Settings option to “not use for AI”

  2. Settings option to “filter out Group Accounts” (and to remove group accounts from leaderboard)

Would these be considered “large” issues? Or easy to implement?

Do you think it’s best to brainstorm here first?

Thanks again.

not sure why you think this should be a topic of conversation. unless something has fundamentally changed since the post below, seems like you really should be working directly with iNaturalist staff to troubleshoot and address the real issue. until you make some headway on that, seems like everything else is just a distraction.

8 Likes

Yes @pisum had several conversations with Scott and Carrie about this and we’ve done our best - and have made it clear the issue is closed from our point of view unless or until iNat - the staff? the community? - wants to discuss other options (iNat can always delete the account if it wishes).

Not sure, however, what any of this has to do with the original thread - but anyway, will try to conduct a more positive discussion elsewhere; and if that doesn’t work, will just shut up about anything that Charlie might disagree with. :-)