Technology and birding, photography and generative AI

There is a need, in any debate about changing culture or emerging technology, to watch ourselves carefully for signs that we are entrenched in a position simply because we don't like change.  It's easy, especially as we age, to dislike change because we dislike change, and there are times if we're honest with ourselves where our subjective preferences are informing our response to new things more than the objective truth about them.  

I'm a middle-aged man.  I do not rely on my phone to be the primary source of information about everything in my world, and I have never knowingly asked AI to do anything for me.  I eschew AI-based searches online and I have not got an account with any LLM (Large Language Model - such as Gemini from Google or GPT4 from ChatGPT).  However, many people younger than me treat LLMs as something entrenched in their lives.  A person I know well works to train LLM programs by evaluating and rating their responses based on real, live interactions with the AI from the general public.  

As a side note, I wonder how many people realise that the AI retains everything they put into it, and, legally, owns it?  I wonder how many people know that there are a small army of data engineers assessing both the prompt and the AI response in the weeks and months after they used it?  Would people still ask it to diagnose symptoms, assess their finances, to write their erotic texts, to be their unjudging companion knowing that strangers will paw through their correspondence later as quality control in order to make AI seem more human?  An interesting exercise in behavioural anthropology, perhaps.  

The main point is this: there is a whole generation of people who are learning to depend on AI to make decisions and to know things, and to do things.  Indeed I know a number of birders who use AI to generate their entire scripts for their online content (and honestly, you can tell), a lack of talent no longer an obstacle to minor celebrity in an obscure and unfashionable field.

So, throwing our hands up and decrying those who use AI in its wider definition isn't really an option.  It's so widespread, so deeply woven into the fabric of thought of an entire generation that our tutting and fretting from the past (as it appears to those invested in AI) is not going to have any impact at all on its use.  The bottom line is: people are going to use AI no matter how flawed or damaging it is simply because it's a shortcut and they can.  If people can see the damage mass agriculture does and still eat beef flown half way round the world, if people can see the damage that fossil fuel emissions causes and still fly on holiday, if people can see the cultural damage done by social media and still join in, then why would we imagine that the downsides of AI will stop the vast majority from using it?  

AI is bad in a number of ways, and smarter people than me have described the vast and utterly horrifying environmental impacts of LLMs with their use of water, their consumption of fossil-fuel generated electricity and the sheer size of the data centres grabbing at land that is at a premium for wildlife.  I won't go into this, though I am aware of it.  

But AI is not one-dimensional.  Most of us don't mean Merlin or Obsidentify or the programs that identify nocmig calls through machines listening on our behalf when we speak negatively of AI, though all these apps are AI-driven using pattern recognition programs powered by deep convolutional neural networks (that specialize in processing structured data—like images, audio, and video—by using small filters that slide across data to automatically extract patterns and features).  Many naturalists use these apps to help them ID birds and insects and broadly, though there are a few notable exceptions, these uses are treated with respect.  There isn't an automatic shunning of people checking their worn moth pictures out on ObsID; though as I've written elsewhere there may be some looking down on those who depend on Merlin to do their birding for them.

Pattern-recognition technology has some benefit for society, like the quick detection of cancers or genetic illnesses in hugely complex datasets that can be analysed incredibly rapidly by AI, or the counting of birds in a colony based on drone images.  Days and days of repetitive work for scientists can be done in less than a second by a pattern-recognising AI.  This is valuable work and in this regard, AI as a tool is useful and positive.  

Where many birders and naturalists draw the line is at generative AI - the kind of use of AI that makes something for us, whether that be drafting an email, making a holiday itinerary, or creating "art" very cheaply (at the point of use) for media purposes.  At heart, the key principle of AI generation is that everything the AI does draws on everything it has consumed as it learns, and given that it has almost unrestricted access to the internet it can draw on unlimited sources of information in order to generate.  In fact, almost everything anybody has ever uploaded to the net is at its disposal.  It's often poor at the moment, but it improves rapidly, drawing on the same pattern-recognition technology that we looked at earlier.  Why does AI draw people with 6 fingers, or feet instead of hands, or a smile of a hundred teeth?  It's following patterns and repeating them too much.  One day it will stop doing that, unless we ask it to.  The ethics of how and where and who owns the information it accesses are deeply debated, but for my part, the stolen intellectual property aspect of generative AI is enough to prevent me from ever using it for myself.  

Many of us edit photographs, whether cropping (zooming in) or brightening, or changing the balance of shading in a photo where the shutter speed or ISO were set wrongly.  It's common for us to adjust an image in terms of its colour, and this can also be misleading.  Not a problem if your picture of a Goldfinch needs a bit of a lighter tone, since it's a common bird and causes no controversy.  With a rarity, however, it can be misleading and most County Recorders specify that images must be unaltered and raw when submitted.

People using AI to de-noise their photographs takes us a step deeper.  The heart of the issue comes from this: using AI to "improve" pictures relies on the LLM either taking patterns from within the picture and grafting them into place to replace other colours or objects, or using previous photographs of that species to replace blurry parts of the picture, using the patterns it has learned to adjust the image to be clearer.  It isn't magically redrawing our low quality photograph of a Wood Sandpiper like an artist with a pen; no, it's using somebody else's photo to improve yours.  A photo that neither it, nor you, own.  Effectively, morally, stealing.  

Questions of the quality of this process aside which is often (currently) poor - and there are some serious issues with it creating impossible to identify images and suggesting details that don't exist in real life, which causes all manner of problems for the ornithologists assessing the occurrence of species in the UK - it's a moral issue that the photograph being presented for people to show your sighting is not really real.  It's a composite of images.  It's a best guess from a program rather than being what you actually saw.  I've seen images recently of waders that were correctly identified in the field but when presented online are so subtly badly edited by AI that I would have assumed they were a totally different species (in this case, a Pectoral Sandpiper that had been given features of a White-rumped Sandpiper).  There wasn't an upside down wing, or odd legs, or Escher-type logic problems with the image, but it just wasn't quite right and it stood out as wrong.

It feels like a question of authenticity, or integrity, or sheer basic truthfulness to me.  Is your image your image, or is it a patchwork of stolen pieces of other people's photographs?  Is it what you saw, or is it a misleading and potentially scientifically damaging image?  This, by the way, is quite aside from those people who are simply dishonest about their images.  Those who post fakes and string sightings to generate clicks and collect likes online.  Grifters gonna grift, I suppose, and it doesn't take long to root out those irritating liars who have always seen something incredible in a place where it shouldn't be - if nothing else their consistently unbelievable sightings will draw them some scrutiny because nobody sees all the good stuff all the time.  These people, once identified, can be safely ignored, their reputation for lying meaning that nobody trusts them anymore.  Their grifts are short-lived.  

The real damage is done when well-meaning people use AI to just pep their pictures up as if it's a harmless little edit rather than the creation of a misleading image from misappropriated source material.  It's this that is causing a culture of distrust to grow.  I've got into the habit of waiting for a rare bird sighting to be proven with a photograph before racing off to see it since there is simply too much rapid reporting of rare birds in error to fully trust the apps; but what if I can't even trust the photos?  What if the camera has been caught in a fib?

Can we ever trust any information that comes to us online without it being a verified source?  That's the long-term impact of the use of generative AI in cleaning up bird pictures: an erosion of trust and the frustrating creation of a necessarily closed community of trusted and proven names as the only authority on what's real.  

The use of generative AI to enhance or edit photographs opens up a can of worms to the detriment of all except in the short-term the person posting the image, who gets a few extra engagements online from the quality of their picture.  It's rare for me to be against change altogether in any given setting, but the use of and reliance on generative AI to adjust, edit, and enhance bird photos is a poor practice and any responsible birder should avoid using it.  At the very least it should be standard practice to be very clear and up front about its use, offering comparisons to the original image so people can judge for themselves what has been seen, but even this ignores the use of an environmentally disastrous and morally ugly source of prettifying a picture for the sake of online popularity.