In reviewing observations for a project I help manage, I encountered an image of a ladybug that appeared to be AI-generated (or at least, some kind of photoshop job). I believe Forum policy is not to link to individual observations so Iām just attaching the photo itself below.
My question is: have others encountered AI-generated images on iNat? Based on the location/date of the ladybug image and the fact that itās the only observation by the observer, Iām pretty sure the observation was made by an undergraduate enrolled in a course where they were required to upload an observation to our project to get extra credit or something.
I imagine the student decided to use AI (or perhaps Photoshop) to create a fake image, in order to save the āhassleā of finding a real bug to photograph. I am familiar with cases in the past of students uploading photos of organisms found on the internet to iNaturalist, resulting in obvious cases of data falsification (the species they uploaded was way out of range, wrong plumage for the season, etc.). The perils of āforcingā students to participate in iNat is its own conversation, but I think AI might augment those perils.
The Citizen Science journal currently has an open call for papers on AI and citizen science. If this particular ladybug observation is not just a one-off, and others have noticed AI-generated images on iNat, I think it could make for an interesting case study on some of the problems AI presents when it comes to publicly-generated data collection.
I do remember seeing an account where half their observations were of AI-generated images. I flagged them for copyright and it started a discussion about how to handle this situation. At the time, there didnāt seem to be any consensus on what to do with these observations. The flag remains unresolved.
I think some updates to the guidelines are needed to direct users on how to deal with AI images. Iād recommend heading down to the DQA and marking ānoā for āevidence of organismā. That way, they donāt reach Research Grade.
If you zoom in on the ladybug, the sharpness and coloring and shadow look at odds with the rest of the photo. Thereās a cartoonish quality thatās hard to explain exactly. @That_Bug_Guy also noted the photo seemed off in the original observation.
Iām confident itās an AI-generated ladybug edited over a real window frame. I had written a very in-depth explanation but Iām just now noticing the ladybug is missing a reflection (despite the window-frame having a clear one)
Searching for āladybug anatomyā on Google provides in the 1st response a photo very similar to that ladybug. I donāt know if itās a common species, but [in France] I never saw it.
I was concerned about the pronotum shape, about the legs, and about the black dots of very different sizes.
This is a very sad issue. Sad to think there are identifiers that just āspamā, and now observers as well. As in any field of activity, a personās reputation can be an important factor to consider. I donāt know what to say, but I feel concerned.
Maybe look again a bit closer? Thatās the outside sill you see through the window, not a reflection. There is nothing from any of the detail on the inside sill visible as a reflection in that image.
Iām not expert enough on ladybugs to comment on whether thatās a real animal - but it looks like a pretty normal ācrap photo taken in poor light with poor focus, most probably with a phone that likes to oversaturate colours (because that makes iPeople give more likes), then resized by inat because the original was too largeā to me.
If itās a cut and paste job they did excellent work on the shadow detail, so occamās razor says āitās probably realā to me - and this is probably an excellent example of how quickly a real concern about a real problem can turn into people thinking they see witches lurking everywhere! ā¦ just sayinā :D
That probably wonāt last long. The current big thing in video game graphics is realistic lighting, mostly ray tracing. I donāt think it would be too much of a stretch to get ai to make use of that. Of course, I just play the games, not make them. Take a small mountain of salt with that opinion.
The height of the window sealant hides where the reflection of the beetle might have occurred. If you look at the shadow under the beetle and the other shadows in the image they match up very well. I see nothing out of the ordinary in this image.
Photoshopping is different than AI. You can sharpen all or a portion of an image in photoshop to the point that it looks fake, although the original subject organism was real.
I meant the reflection onto the surface below the ladybug, sorry if I wasnāt clear.
Another concern is that if the flash was on, the ladybug would have no visible shadow, which isnāt the case. Looks like someone just used a drop shadow onto a cut-out of a ladybug. The shadow doesnāt match the supposed light source.
Maybe not AI, but a photoshop. The flash appears to be on, meaning the shadow would be behind the ladybug, not to the front of it. The angle also seems off and the ladybug seems to be too sharp in comparison to the rest of the photo
Thatās a real image. Not AI and not photoshopped. I downloaded the pic and zoomed in using an image viewer software to see the individual pixels. Nothing suspicious. AI ruins everything in two ways: we have fake stuff that looks real, and real stuff will be seen as fake.
I regularly take photos of moths with a simple fill flash. They all have shadows. Flashes create shadows. One of the reasons good macro photographers recommend flash diffusers is to reduce the harshness of shadows.
