Prompted by @gcwarbler to start a new topic about this from discussions in this thread – sorry for seeing this late, Chuck!
My original comment:
There is also the issue of phone cameras embedding AI in the photo processing. Most users don’t know that the phone is automatically “processing” the photos at all. For small inverts I’ve had my phone totally change the leg stripes and other markings to “enhance” the photo without my permission, and it’s an opt-out feature rather than an opt-in on most smartphones nowadays. I’m not sure how to address this beyond asking users whether the markings in a suspicious photo match what they saw in real life, and then directing them to a how-to disable this feature?
Some additional context:
This BBC article explains it better than I can. I’ve noticed this particularly with small subjects which have important details that read as “noise” such as setae or trichomes. There is a smoothing or sharpening effect that would get applied to my smartphone photos which ended up sometimes altering the appearance of, or even erasing, small identifying details. To turn this off, I had to go into my camera settings and turn off all optional settings. I have a Samsung, it’s probably different on other phones.
(If this is already being discussed elsewhere, happy to merge!)
I have noticed this most egregiously on Samsung Galaxies (Galaxys?) based on the photometadata. To be fair, I really only ID anoles, but the only anoles I’ve seen that have obviously been AI-processed to something unnatural have come from Galaxies.
I’ve left comments to this effect on some observations and downvoted the DQA flag. On one of the observations I left a comment on, the user objected and thought I was accusing them of intentionally manipulating the photo with AI. I clarified that I thought that the issue was likely default behavior on their Galaxy, and they were surprised and said that they would look for a way to turn it off. I briefly searched online, and it seems like the methods for turning off the default processing have changed over time and some features may not be possible to turn off on some phones.
Some of the effects may even be traditional algorithms and not AI. Gaussian blur and subsequent edge sharpening. At one observations of a marten jumping high in the trees I had to use frames from a video, because they were less distorted than the photos and the details to actually see that it is probably a marten were only visible in the video frames. It probably was not an issue of too large exposure time or similar.
What to do? If they are truly AI additions that distort the reality and invent new features, I think they had to be downvoted in the DQA. And the photographer should be informed.
The more basic filters are annoying, but there is often no way to turn them off, unfortunately. At least they do not invent new stuff on the larger scale.
When I did a little research about this a few months ago, it was only the most recent Samsung models, Galaxy S25 and S26, I believe, but things changes quickly. Which Samsung model do you use? Specific model information is helpful.
I agree, just cause it looks a bit over processed doesn’t mean generative AI was used. Phone cameras have been doing all kinds of processing for many years now.
Agreed, I don’t think the line should be “was the editing done with AI or another method”, but rather “did the editing substantially change the appearance of the subject”. There are both AI tools and traditional tools which alter images in unintentionally deceptive ways, and I don’t think the method of processing should be the deciding factor as to whether to vote No on “accurately depicts scene”.
The AI optimization settings in the Samsung cameras are spread across various settings–and some cannot be turned off (and your pics may look like crap if you manage to turn them off). The work-a-round is to not use the Samsung camera app at all by installing and using the Open Camera app. It does not use any AI at all. It does apply standard (old fashioned) optimization, and even that can be turned off in the “processing settings”.
I did a side-by-side test between the Samsung Galaxy S24 camera app (default settings; with AI optimization) and the Open Camera app (default settings; no AI optimization), and I don’t see anything that alarms me (in terms of fear that AI has altered the image to make it no longer represent reality). I can’t speak for what the newer S series (including Ultra) does. But anyone with a concern can replicate this side-by-side test against Open Camera.
You’d have to reject the photo upload based on the metadata. The metadata could be scrubbed if someone cared enough but I doubt the majority of people would.
On my Galaxy s23 (which I’ve just sent in for a part exchange on a Fairphone 6+), the problem isn’t with close-ups. It’s with a blurry bird in a tree that you then crop - and it fills in the pixels to make it look more birdlike. It would be interesting to see a comparison side-by-side with the s25
Arguably, the stock app loses detail by getting rid of the noise more aggressively. Some sharpening after the denoise also appears to have taken place. Nothing seems to be invented, just the smallest scales are somehow distorted. Nothing that would lead to rejection of the photo.
It looks like Fairphone isn’t entirely free of AI:
“It’s also worth mentioning that our camera systems in both the Fairphone 4 and 5 use AI algorithms to optimize auto settings and for post-processing, especially in low-light conditions.”
But I’ve read that “Fairphone’s automation focuses strictly on computational photography—optimizing true-to-life colors and ensuring sharp focus rather than heavily editing the scene.”
I really hope this isn’t the world we’re going to have to live in. There are a huge number of people who can only take photos on their phones. Imagine suddenly being shut out of iNat because your brand-new $1000 phone added AI photo processing the week after you purchased it.
If you take a photo within the iNat app that should avoid the extra AI processing right, since it’s a different camera app?
In that case the solution would be to just check the metadata if photos are imported from the gallery and give an error or pop-up flag if certain phone models are detected (or even better, check both the phone model and whether the photo was taken by the default camera app or a different camera app). Then the developers can decide to either a) force the observer to retake them within the iNat app, or for a softer approach b) ask the observer to review the photos for any inaccuracies. If (b) is chosen then observers using flagged phone models and having to deal with that flag every time they upload photos from their gallery will probably decide it’s just easier to use the iNat app to take their photos.
That would be very unfortunate. Firstly, the app’s photo features are very rudimentary and secondly, you cannot store the photos to deal with the upload later. I never upload when I am outdoors.
I agree and I do the same, but the question is whether it’s worth the tradeoff to avoid getting a lot of photos from observers who are naive about the AI effects they’re introducing onto iNat.
