The recent discussion about AI-generated and AI-enhanced wildlife photographs has got me thinking about where photography and this platform might be heading.
A recent article warned that generative AI poses a threat to biodiversity databases such as iNaturalist. It points to the case of a supposed Red-winged Blackbird in Brazil which turned out to be an image of an epaulet oriole that had been altered using AI. The researchers also note that around 1,400 iNaturalist images had already been flagged for AI use, out of more than 610 million images.
Tony Northrup recently discussed a possible alternative approach. Rather than trying to develop increasingly sophisticated systems to detect AI images, what if cameras instead cryptographically authenticated photographs at the moment they were taken?
In other words, instead of proving that an image is fake, we prove that an image is real.
But there’s a catch.
If an image has to remain cryptographically verifiable as an original photograph, how much photographic editing would we have to give up?
Would you be willing to submit only authenticated, essentially unedited images to iNaturalist if it meant being able to prove that the photograph genuinely came from a camera?
Or would that go too far?
I’m genuinely curious where iNat photographers would draw that line.
I would not give up the ability to crop, adjust lighting and color, sharpen, and delete lens spots. Pretty basic editing and certainly not AI manipulation.
maybe if you had to upload 2 photos, similar to how some cameras can take a raw and a jpeg. you can edit the raw as you would normally and then verify with the no AI used signature on the jpeg.
The only editing I do is cropping, so if it can still be verified / tracked with that -fine. I would not be willing to give up the ability to crop.
I would also wonder about the cost for new cameras with this feature for people. I am not in a position to buy any new cameras, to get this (or any) new feature, so that makes the question more hypothetical for me.
I do a lot of bird photography and often in the early morning so the light is low and to capture birds often I have to use ISO values up to 5000 and sometimes 10000. I find value in Lightroom Classic denoise in some situations and would not want to give that up. This is an important discussion. My Nikon z6iii has a feature to cryptographically sign the raw image in camera, but after the firmware was updated to provide this feature a security researcher found a way to bypass this so Nikon removed the digital certificate signing process until the issue is resolved. Hopefully someone will come up with a reliable way to detect generative AI changes to images that still allows AI noise reduction which has value for species identification in poor lighting situations.
This could be useful in some situations, but in the specific case of iNaturalist, I’m not sure it would actually be feasible. The main problem is that I’d say 95% of images submitted to iNaturalist (maybe even more) have already been edited by the device used to capture them, mostly a smartphone, more infrequently a camera. Every time you take a jpg image, your device is making some choices for you, everything from adjusting brightness to removing noise and modifying colours, to more sophisticated and radical changes employing AI techniques directly in the device. Most people don’t even realise this is happening and even if they did, would have no idea how to stop it. How would such a verification system cope with this scenario? Even if it could be made to work, I’m pretty sure it wouldn’t be effective for long. Just as AI image generation and AI detection systems are both improving rapidly (the classical lion and antelope scenario), I’m pretty sure that it wouldn’t be long before someone found a way of also faking any in-device verification system (see P:S. below). The answer? I really have no idea and as someone who has always had an almost religious respect for photography as a way of documenting reality, it disturbs me greatly.
Speaking purely personally, the simple answer is “no” . The problem is, I’m obsessed with photography almost as much as I am with nature and when I submit observations to iNaturalist, one of the things that gives me great pleasure is to try and provide the CV and the community in general with clear and easily legible images which I hope may help others ID the organism in the future. I shoot in RAW. That means I can push my camera to its limits, knowing that by working on the digital negative I have a good chance of restoring the image to a pretty accurate reproduction of what I saw when I took the picture. I never (repeat never) manipulate the content of the image (I can’t even see the point) and I’m always very careful to check that no processing (such as noise removal) has modified that content without my knowledge. Doing this gives me enormous satisfaction and if that were taken away from me, I’m pretty sure my enthusiasm would wane fast.
That said, the problem is real and serious. If it became obligatory to submit only unedited out-of-the camera (smartphone?) images, I’d understand and certainly not oppose the decision. “The needs of the many outweigh the needs of the few” .
P.S. I’ve just read @ronbinature’s reply… I rest my case:
There are “basic” editors that do not remove the EXIF and location data.
I presently only “Rotate”, “Crop” and “Resize” my pictures with Irfanview
I may exceptionally add a second picture that has been “automatic colour adjusted” to enhance detail.
All the major smart phone companies are enhancing the photos using various AI and machine learning techniques so saying no to AI is saying no to smart phones.
Vast majority of my photos are not with a smart phone and are shot in raw on camera and then adjusted before upload to crop, noise removal, sharpen and adjust lighting as needed (all of which are also done by smart phones automatically as would the camera if shooting in jpeg instead of raw).
All s/w used has generative AI capabilities turned off.
I would not give up the ability to crop. The rest I can do without. I know that because with my crude skills and program (I depend on Microsoft’s Paint), that’s all I do now. The results may not look wonderful, but most of them are identifiable.
I would also say no.
One of the common techniques to capture the details of small organisms is focus stacking, which is where you take pictures at different focus steps and then combine the sharpest portion of each image so that the entire subject is in focus.
Beyond everything else, that seems like something that would take a very long time to actually implement- we’d have to wait for everybody to get a new camera with that feature in it.
If people can’t crop their photos and adjust the lighting of really badly lit photos, identifying is going to become much more unpleasant. A lot more squinting at photos to figure out where that bird is, and once you find it, there’s not enough detail to tell what it is, because…
Is iNat going to allow larger images to go with this? Uncropped images can be a lot larger. And if they’re going to be shrunk on upload, you’re losing a lot of data that way.
Who is paying for everyone to get a camera that can do this?
…so, like, assume at least 10 years. Because some of us are stubborn about hanging onto technology that still can be made to work.
The only way I see something like this even, maybe, happening is on a purely voluntary basis. In addition to the clear, and distinct preferences people have expressed, above. There is also the little detail that iNat is world-wide.
While the question from taylorse mentions one part of the ‘who would pay for this’ -
there would also be the question of who would provide/pay for the infrastructure (internet, cell service, computers for those using regular cameras and uploading…) that might be needed with new cameras, on a world-wide basis.
I live in a rural area in northern California. (Rural = the nearest stop light is 13 miles away; the post office doesn’t deliver, cell service is sketchy at best, etc.). There are plenty of folks I know who do NOT have access to any internet or cell service, even here in California, simply because of where they happen to live. Let alone entire countries where ‘rural’ is the norm and/or there are other barriers.
I think that if this tech is available, then the people who want to use it should by all means do so. For the rest of us, I guess we continue with the ‘risk’ that someone may assume our obs are AI created; and those of us who ID have to continue to keep alert to the AI possibility, and use those tools (existing and new ones that may be available in the future)
As to the OP: no. Putting up bad photos would ruin some of the enjoyment I get from all this silliness. Crop, light/dark, a little contrast, occasional dehaze, noise suppression and fixing the temperature when I forget to switch from flash back to natural light are all I use on iNat photos. None qualify as AI in my mind. I don’t have an issue with using masks, especially in tough lighting situations, but I don’t like to work that hard on a photo only one or two identifiers may ever look at.
. The problem is, I’m obsessed with photography almost as much as I am with nature and when I submit observations to iNaturalist, one of the things that gives me great pleasure is to try and provide the CV and the community in general with clear and easily legible images which I hope may help others ID the organism in the future. I shoot in RAW. That means I can push my camera to its limits, knowing that by working on the digital negative I have a good chance of restoring the image to a pretty accurate reproduction of what I saw when I took the picture. I never (repeat never) manipulate the content of the image (I can’t even see the point) and I’m always very careful to check that no processing (such as noise removal) has modified that content without my knowledge. Doing this gives me enormous satisfaction and if that were taken away from me, I’m pretty sure my enthusiasm would wane fast.
.