Would you give up your photo editing ability?

Like others here, I think losing the ability to crop and adjust exposure would be too much of a compromise.

I’m hoping that iNat is in a good place to weather this AI-image storm, as there are good flagging and moderation tools and there are several ways other than a submitted photo to verify an observation, including how many times that observer has observed that species before, how many times they have been the first to correctly identify that species for others, and how likely that species is to be found at that place and time based on all our other observations.

Same, I generally would not give up cropping, and almost all observations aren’t cropped anyway

This question leans more towards ethics and integrity rather than the technology itself. Nature observation is largely an honour system: people say they have observed a certain organism on a certain date at a certain location, and almost all people are honest about this. If someone lies about a sighting, which is actually quite rare, they may lose their reputation and credibility forever. ‘Swallowgate’ (the famous faked record of a Violet-green Swallow in North Carolina) is still a great example case of this honour system. Same thing applies to photography, AI and every other aspect of life; there will be manipulation by a few but luckily the honest, benevolent people constitute the vast majority.

Just because some people have misused artificial intelligence to fake a photo of a Red-winged Blackbird, it does not mean that now suddenly everyone should either have to be able to prove the authenticity of their edited photos or give up photo editing as a whole. Basic photo editing is essential to showing the identifiable characteristics of an individual: uncropped photos in poor light are not the most useful for identification, so cropping and basic lighting adjustments (brightness, shadows, highlights, contrast) make the sighting more valuable. Even some more sophisticated editing, such as vignetting, dehazing and noise reduction, if not overdone, could make a photo more pleasing to the eye.

Adjusting sharpness and colour (temperature, tone and saturation) are a bit riskier, because the whole idea of reporting an organism to a nature database is to stay true to the original sighting and to show the real colours of the organism, but then a camera lens is simply not a human eye and will have always distorted the authentic experience in one way or another. To add to this, for some organisms, a great deal of reports are made with black-and-white camera traps at night, leaving the organisms completely colourless on the image and not actually observed by any human being in person. This could also be considered artificial intelligence but is widely accepted.

So I do not think that, based on the malpractice of a few, from now on photo editing should be limited to everyone in order to prove that the photo is not AI-manipulated. Instead, the rare cases of dishonesty with the use of AI should be consistently detected, reported and removed by the community, manipulative accounts should be banned and good ethics should be reinforced, so that the influence of wrongdoers remains insignificant for the database. Lastly, the more knowledge and experience we have of photo editing, the easier it is to spot when something is off, to suspect the use of artificial intelligence and to flag a suspicious sighting.

I think the best way to combat fake records is to remove the incentive to create them. Whether it’s a fake description on eBird, a photoshopped image on iNat, an AI-generated picture, a real picture taken from the Internet or a friend and uploaded with fake data, it’s all equally problematic. Generative AI makes faking stuff a bit easier, but it’s not a new problem by any means; this is just the current “newest way to fake stuff”. In past centuries, making fakes like the Coventry Fairies or the Cardiff Giant took a lot more time and effort than making fake discoveries today, but it’s inevitable that ease of recording data comes with the side-effect of ease of faking of data using related technologies.

There are sociopaths who just want to break things and will fake stuff simply to mess with people and get attention because it brings them joy, and there’s not much we can do to prevent that from happening. But I’d bet the average “fake observation” is from someone who is either being forced to use iNat against their will (as part of a class requirement) or from someone who is trying to win some form of competition (i.e. a BioBlitz that bills itself as a competition to see who can make the most observations and top the leaderboard). Working to eliminate those uses of iNat will take away the motivation of a lot of the fakers, and I’d say that will have more impact than trying to validate every single photo in some automated way. If anything, we might be catching more of the fake observations now that they’re being made with AI, as some AI images are easy to spot, while photos stolen from some obscure corner of the Internet are virtually impossible to notice.

It’s necessary for me to edit my images, for both quality and file format. The most important are cropping, brightness, contrast, and file format. Other less critical changes include noise, sharpening, and several others than I sometimes use.

There is a huge difference between normal photo editing and AI generation/modification. We should not ban photo editing to solve a separate problem.

There is editing and clarifying. Sharpening an image and cropping seem like must haves or we’ll lose a lot. Other that that, I don’t use or need anything else.

Thanks everyone for the interesting responses. I can’t respond to every comment individually, but they’ve raised some points that I hadn’t considered when I started the thread.

I’m beginning to think that the issue is considerably more complicated than simply ‘edited vs unedited’.

The C2PA approach sounds promising in that the provenance can potentially follow an image through its editing history. So perhaps photographers wouldn’t have to give up things like cropping, exposure adjustments or even AI noise reduction. Instead, those changes could become part of the recorded history of the image. But as pointed out already, that doesn’t establish whether the file itself is actually genuine. For instance, someone could photograph an elephant in a zoo and claim they’d taken it somewhere on the African plains. So C2PA seems capable of strengthening the evidence for a photograph without actually establishing that the observation itself is true.

And it raises another question, also pointed out here: who would actually pay for this technology, and how complete could that provenance chain actually be?

It would presumably require the camera to create the original credential, the editing software to preserve and extend it, and the various other programs and platforms in the workflow to do the same. What happens when one link in that chain doesn’t support C2PA, or when the credentials are stripped somewhere along the way? The Nikon Z6 III example mentioned above seems particularly interesting in this respect.

And finally, I’m wondering about posterity.

Suppose authenticated photographs become commonplace in 2036. Would this new form of evidence retroactively weaken the old kind, or would there be two kinds of evidence in play?

As I understand it, genealogists ran into a different version of the same problem. Once DNA testing became mainstream, parish registers, certificates and service records had been sufficient proof for generations, yet a well documented paper trail alone started being treated as merely provisional once DNA became the expected standard, regardless of whether the original documentary case was ever actually weak. Perhaps the solution would be to run separate frameworks side by side evaluating older material on the kind of evidence it can actually offer, rather than the kind it can’t.

For iNaturalist, given the enormous historical value of the observations that have already been accumulated, I would hope that ‘no provenance information’ could never become synonymous with ‘untrustworthy’.

With the pace of this technology it might be worth pondering that if it was ever adopted, then what exactly such authentication would establish, what it wouldn’t establish, and how we would treat the enormous body of legitimate photography that exists outside the system.

I’m the same. I do basic edits but not interested in using AI to change images. Crop, lighten, sharpen, remove spots, that’s it.

I do mostly macro photography, which requires good lighting often not present in the field, and is prone to noise issues. Prohibiting editing would be a major downgrade to the quality of insect and arachnid photography on iNat, and will lead to missed range extensions when the identifying feature is in a dark part of the image

So no, banning editing would be horrible for iNat

And what about cropping? Banning that would make the site almost unusable

I also don’t think AI main cause of range errors. Stolen images and location falsification have long been known as things to watch for before AI

Cropping a photo should be a total non-issue even for purists. Narrowing a photo down to eliminate surroundings and focus in on the subject does not alter the image, except for some loss in resolution.

Even if this

were to happen, photos taken from 2024 and earlier (when AI wasn’t really good enough to fake things well) would probably still be considered “reliable”.

Interestingly, though, I see a bit of the reverse pattern when thinking about scientific data and citizen science databases. Data from natural history collections are often treated as a gold standard/sacrosanct, when in reality there are often errors (some rather large), and locational precision was much lower as these data were collected prior to the widespread availability of GPS technology. The geospatial coordinates found on pre-GPS museum records have largely been created well after data collection from textual location descriptions which are anywhere from TRS based descriptions to the nearest town or just a county. The average observation location from iNaturalist is likely approaching an order of magnitude or more precise than these older records, but they are often discounted as being lower quality. Of course, in some cases, this is true, but the technological advance has lead to a massive, systematic increase in data quality which (in my opinion) is often largely overlooked.

I had a museum curator question the location data associated with the photo voucher of a specimen that was found dead in the field but not collected. It was a really good new location for that species. The curator seemed to think that if the specimen had been collected, that was somehow a more valid voucher than the photo when it comes to location. It’s just as easy to fake a location for a physical specimen as for a photo … in fact the photo might have location data imbedded in it whereas the specimen would not of course. (There is a lingering bias against photo vouchers among many curators of physical collections.)

It puzzles me why, when there are people doing things that are wrong, we focus instead on the behaviour of the people doing it right. No one using iNat who is not breaking their rules or harming the site, should be expected to change anything. They have nothing to change as they are not the problem.

C2PA is utterly impractical.

Its implementation relies on every step of a workflow being compliant which is completely unrealistic. Hardware support is patchy and software support is essentially non-existent.

Until and unless such support is essentially ubiquitous then even talking about implementing it is worthless.

i think the original premise may be based on a misunderstanding that new mechanisms (C2PA) to identify AI-generated/-edited images would not be compatible with subsequent editing. this is not correct. subsequent edits can be done in a way that preserves the provenance of the image, as long as the editing tool(s) adds to the provenance chain.

having an unbroken provenance chain that shows no AI tools were used from capture to final version would be good evidence that an image was made without AI.

having a broken chain or no chain just puts you back in the current state of affairs, where you just need to rely on judgement or certain technologies (ex. watermarking) to identify AI, although this can result in lots of false positives and things slipping through the cracks. people who are intentionally trying to hide their AI use will always try to hide their work here.

so in a future world where more hardware and software will have the ability to record provenance, people will start to look more for the provenance chain in cases where the provenance really matters. but right now, iNaturalist doesn’t handle it, most cameras and smartphones in use don’t capture that information, and not all editing tools support it. so it’s a little early to nudge regular iNaturalist users to change their behavior, although maybe professional photographers or researchers might want to start adopting cameras and editing tools that support C2PA for their own reasons.

science is bolstered by repeatability, right? so if you have, say, 1 image by 1 observer of a rare plant at location X, and no one else is able to subsequently document it, then even with a photo that has the full provenance chain, folks might still reasonably question the observation. even GPS coordinates captured by the camera may not always be accurate, nor may the timestamps always be accurate, even if we know that the image was captured by a particular camera. (that said, an image captured with provenance could still be preferred over one without provenance, all other thins being equal.)

on the other hand, if you have 15 images by 15 observers of the same rare plant at location X, then even if none of them have provenance chains captured, there’s likely little reason to question that the rare plant exists at location X, right?

There are quite a few security experts who disagree very strongly with this. In particular, Neal Krawetz has exposed many of the serious flaws in the design of C2PA, and written numerous detailed blog posts about them. A general overview of his conclusions can be found here: C2PA’s Worst Case Scenario. And for an entertaining recent example, see here: C2PA and Pixel Glitter Milk. (NB: as this is a fast-moving industry, some of the details may have been overtaken by more recent events - but all of the core issues remain live).

For those who find these articles too technical and/or too long, here is one of the closing summaries:

We live in an era of deep skepticism, where public trust in visual media is at an all-time low. Proponents of C2PA argue that cryptographic signing solves this problem: if an official photo carries a valid, hardware-backed C2PA signature, the public can trust it. But the truth is that the C2PA signature carries no weight for providing any type of reliable authentication, validation, or provenance. Instead, it turns every device into a powerful tool for laundering disinformation as fact, which is worse than doing nothing.

What I find most unsettling about C2PA is that it’s clearly a product of the corporate world, rather than being developed purely for the public good. Every time I see a company uncritically pushing C2PA in its marketing, it leads me to distrust the whole idea of it a little bit more.

Only really use cropping and brightening (I keep my shutter speed at like 1/1000 or whatever cuz ants are fast) I don’t really think anything else is necessary. Except if you want to add watermarks but I think that isn’t really editing

Would do far more harm than good. I would have personally had to upload about a TB of images if I were uploading the RAW images. Cropping, light enhancing, even exporting as a JPEG would all lose that signature. Enforcing it would also mean every single person on iNat would need to buy a new camera.

that’s incorrect. if you use, say, Lightroom to process your photo and export a JPEG, the JPEG file will have a C2PA manifest from Adobe that can include information about the original RAW file manifest (if one exists), plus information about what was done in Lightroom.

If your original RAW doesn’t have a C2PA manifest, then that’s fine, too. you just don’t have a complete C2PA chain, which is the case for most images today. no big deal. no need to change anything unless you want to.

I hadn’t heard of C2PA before this thread. However, having done a little light reading, I’m not personally convinced that it would solve the issue of people uploading AI Generated content.

I can see why it might work as a way of checking the provenance of pictures for a news website or corporate website. Because if a photographer, journalist or agency supplied a picture that wasn’t theirs, there’d be the possibility of professional/commercial embarrassment.

However, for a citizen science website where the aim is to source a mass of content through the actions of unpaid, often unknown, ordinary people, things are rather different. If some of them want to fool around and post images that they didn’t take, there’s absolutely no professional jeopardy for them in getting found out. Ultimately of course, as unpaid volunteers, there’s little reason why iNat posters would want to spent their own money on kit that conforms to C2PA standards.