Although it would be nice if more folk were interested in what birds were being seen in Dundee, this one particular blog-post probably needs a much wider audience than my usual figures, for reasons that will become apparent when you read it. If you agree that the more people who are made aware of what this post is highlighting, the better, then please do share this far and wide. It may save you or someone else a lot of time, confusion and/or frustration at some point in the near future.
---------------------------------------------------------------
Last Sunday a photo appeared on Facebook of 2 Swans in flight taken at Loch of Kinnordy that day. The question was being asked as to whether or not these might be Bewick's Swans or the more likely Whooper Swans. Something about the yellow on the bill didn't looks right to me for Bewick's. It appeared to be the wrong shape for Bewick's, but it was also not quite right for Whoopers, (but appeared more Whooper-shaped to my eyes). I messaged the photographer via Facebook messenger to offer my opinion that I didn't think they were Bewick's, and explained my reasoning why. The photographer had also posted the photo to a Bird ID group and there were a few folk on there saying 'Bewick's', though there was one comment from an experienced birder who I'm aware of (but don't think I've ever met), pointing out that the shape of the yellow was more Whooper-like than Bewick's and that Whooper was also far more likely than Bewick's (which more or less mirrored my own thoughts but was explained in much more detail than I had managed).
 |
| The Swans photo as posted on Facebook |
Not long after I'd spoken to the photographer of the Swans, a WhatsApp message appeared on the local bird club group from the local area Bird Recorder saying that 2 Bewick's Swans had been reported from Loch of Kinnordy. I decided to stick my head above the parapet and said that having seen the photo on Facebook, I didn't think they were Bewick's. Fast forward to Monday morning and there was a new post on WhatsApp saying they looked like Bewick's. At this point, I was at work and not really able to participate in the discussion fully. The previous evening I had been rather fixated on the shape of the yellow on the bill and I hadn't paid too much attention to the head shape. Looking again on Monday morning, the head shape did indeed appear to favour Bewick's more - though the shape of the yellow on the bill still niggled. I quickly posted a reply where I admitted that I may have been wrong - given the head-shape and that I was less convinced about them not being Bewick's than I had been the night before.
A comment was added by someone else saying they had shown the photo to some 'decent birders' and that the birds pictured were "100% Bewick's". A very experienced local birder then asked to see the photo which hadn't been posted to the chat. As the photographer isn't a member of the WhatsApp group I shared a copy of the photo privately with them and said that I was 'not wholly convinced either way'. However the photo was then posted to the group by another participant. I thought it a bit odd that there was no immediate statement of "the swans are (whichever)" from the very experienced local birder. However, much later in the day, I received a reply to my message to him where he queried the possibility of AI being involved. This was very much a lightbulb moment for me. I had recently seen another photo online where features of a bird had been inadvertantly changed quite considerably by an 'enhanced editing' option in the software used and hadn't been noticed/queried by the poster of the photo (the AI possibility was spotted by a very experienced birder).
I messaged the swan photographer to ask if they had the original unedited photo and if so would they mind sending me a copy so I could try to solve the mystery of what species they actually were. They were able to send me a copy of the unedited photo via Facebook messenger on a phone - which did reduce the quality a bit. However, I lightened the photo (without any AI intervention, using the Snapseed app on my phone). I then overlaid 'my' lightened version over a phone photo of the original 'edited' version, and the actual 'truth' was revealed. Despite the slight block-iness of the unedited pic (due to being shared via phone), it was easy to see that AI involvement in the editing process had actually changed the shape of the yellow on the bill, as well as the overall shape of the head, resulting in a false picture of what had actually been photographed at the time of the sighting.
 |
| The unedited photo (feel free to save and lighten/crop this photo yourself) |
 |
| Lightened and cropped edit from unedited original |
 |
| Overlay of FB posted edit & my edited version of photo to show features together. Non-AI enhanced version is top in each pair. |
"100% Bewick's" the swans most definitely were not. However, superficially there was a resemblance, just as there was superficially still a resemblance to what they actually were. 2 Whooper Swans. The resultant AI enhanced photo were clearly identifiable as 'winter' Swans, but didn't look fully like either of this migrant species pair, though it is perhaps understandable why such 'definite' IDs were offered up, as we all expect photos to be a 100% true representation of what was actually seen by the camera, when the photo was taken. My basic edit (cropped and lightened only - no AI) showed that the birds photographed were Whooper Swans.
-------------------------------------------------------------
It appears that increasingly we can no longer trust the 'reality' of photos posted online. Although Photoshop etc has made it possible to edit photos extensively for many years, it is much more time-consuming/labour intensive to create similar results to that which AI can do in seconds with a single click of a mouse button. AI should get better in time, but as things currently stand, it creates a huge problem with regards to photos as evidence of bird sightings. In cases such as these photos, the end results have been 'created' inadvertantly, just by trying to lighten underexposed photos. You wouldn't expect anything to change within the photo except the overall brightness, but it clearly had. By way of explanation, it appears that at some point, 'AI enhanced edits' (or something similar) have been OK'd and the software in use is now defaulting to doing what it thinks is 'correct' rather than just the required very basic edits.
Not everyone is particularly 'tech-savvy' so it is easy to see how easily this can, and will likely, happen, and when it does, the 'AI edits' may be blatantly obvious - changing major ID features, or they may be as subtle as the Swans' bill/head shape seen here. Obviously, not everyone will realise what AI-enhancements can and will do to their photos, and may make the argument that if it saves a bit of time, then why not use it? As can be seen above these sorts of 'AI enhanced' photos can and will likely make local recorders/rarities committees/eBird & Birdtrack assessors tasks harder and more time-consuming than they already are.
It has also been suggested that AI can be 'influenced' by the file name, so if the photo being edited is titled Great Tit but is actually a Blue Tit then AI may change some features to make it look more like a Great Tit (I have seen AI models being 'eager to please' when particular opinions are being sought, and so can see how this might arise, but as of yet, I have not come across this personally with regards photo edits).
If your photo-editing software offers 'enhanced' editing, or 'AI enhanced' editing, or 'smart' editing it may be wise to steer well clear of that option, if at all possible.
I don't know what the overall solution to this particular problem is, but I'm posting this particular blog-post as a 'heads-up' of something that every birder/bird-photographer probably needs to be aware of. There will be, (and already have been) purposely AI-generated 'fake' photos of rare birds in places where they shouldn't ordinarily be. There are also false positives generated by the Merlin app (and similar) now all muddying the waters of what is and isn't a 'genuine' photo of a 'genuine' sighting (or in Merlin's case the identification of a calling/singing bird). Accidentally AI-enhanced photos being introduced into the mix is most definitely not something that is needed. Unfortunately, chances are things are going to get a lot worse before they get better...but the more people who are aware of AI messing things up, the better - so that it either doesn't happen in the first place - though if it does, then someone may flag up the possibility.