AI: True Threats | Articles | Ant Smith

When looking at 'The Great AI Debate' in photography we see a number of 'threats' promulgated, which are not threats at all but rather opportunity to disrupt the status-quo (see my previous article). And let's face it, in this world the status quo is always worth disrupting.

But behind this hand-wringing smokescreen of 'threat!' lies the true dangers we are facing; lesser discussed concerns because they more greatly impact our common good.

The Lowest Common Denominator

Wow, just look at how much I can improve my image of Corvids on Steeple with the help of Adobe®™'s zany senior creative director 'Doc Brown' (I dunno if he is a doctor but he seems to style himself thataways, and in the spirit of actual doctor Doctor Karl I'll allow it, even if it is a conceit...)

Of course, I actually prefer the original (left) image with it's pastel sky set against the golden glow of the steeple in the late afternoon. It seems more delicate, subtle even (but then I did write a whole article on How to take boring photographs).

Doc Brown's sky image is called Spectacular004 (does he really have as many as 999 spectacular skies?) - but when does 'spectacular' become mundane?

Certainly when one sky image is used and reused mundanity prevails, no matter how 'spectacular' it may have looked in the first place. But actually as creative individuals we won't do that, jump on the first 'wow' sky image we find and whack it into all our pictures. Much more likely we'll help fuel a whole new micro-industry trading in skies. Or perhaps even make our own library of great sky pictures for shoe-horning into future photos...

...but the true threat here lies in the idea that real isn't good enough; that there is only one 'ideal' to aim for. That there is a perfect sky, so much better than the one above your head. Perfect blue skies with white fluffy clouds, deep red sunsets, Rainbows, The AURORA BOREALIS - splattered here, there and everywhere across our cultural landscape; because no one's gonna buy and sell the drab grey skies of an English winter's afternoon.

Sky replacement though, it's pretty obvious. Easy to attack, easy to defend. Whatever. More insidious is this idea that there is one common good. That somehow we can get to the best, the very best result by looking at billions of images to help make each pixel the best they can be. Samsung's idea that there can be one perfect photograph of the moon. It seems Samsung 'enhances the detail' in moon photos even when the detail isn't actually there.

[Ref: reddit from instagram user @ibreakphotos]

The claim is that the Samsung AI is so very smart that it has learnt how to create moon-like textures on moon-like objects - not that it is wholesale shipping pixels into the image... but does that distinction matter? The point is, the results are based on what has gone before. The AI is only as good as what it has been trained on; so at its very best it can only match, not exceed human endeavour.

But, replicating the best of what went before isn't the definition of excellence - being excellent means to exceed what has gone before. Taking one strict, calculated view of what is 'right' (or was 'right') doesn't help us to advance. Doing the same over and over only leads to mundanity (however good that same thing was, in the past). And once the AI training materials are themselves AI informed images? Will there not be a downward spiral of increasing mediocrity? A negative feedback loop? A locked groove? Hmm, probs.

Our understanding of 'perfection' is so limited that chasing after it leads us to a tiny pot of fool's gold. We don't need to artificially create 'perfect images' - we just need to make the most of what is right in front of us. We don't even need to go chasing off around the world trying to find the perfect location, we just need to expose the beauty (or horror) of the world at our feet.

Anything more than 500 yards from the car just isn't photogenic.

Edward Weston

Spectacle, beautiful sunsets, vibrant colours, ultrasharp definition - everything that defines 'great' photographic images in our zeitgeist - are trappings, but one way to see. Vital if you want to work for the advertising industry; but still just one way of seeing. The one way we have trained our AIs.

AI seems amazing, until you realise that all it can help us to do is create yet more mediocrity. Brilliant, beautiful, stunning mediocrity. To see one way. To think one way.

Accelerated Obsolescence

My current PC (Orac-1) benchmarks at around 16k (score is relative to performance of the original IBM PC), placing it mid-table in the premier league when it comes to CPU performance; and still above average all round. I built it 4 years ago (based on an old HP Z820 workstation) at a performance level that I hoped would serve me well for up to 10 years. And it's doing really well on that journey. I'd used the same strategy for my previous PC (HAL) - buying high-end components that were a number of years old. This makes for a relatively cheap yet still good performing machine. HAL kept up with software 'progress' for a decade, so I wanted Orac-1 to do the same.

But then Microsoft™ decided not to bother supporting machines from before 2016 after its Win10 end-of-life date (Late 2025); which means I'm only going to get 7 years at best out of Orac, increasing the cost of ownership from a planned £3/wk to over £4/wk!

They've done this because they want to use TPM (trusted platform module - a hardware chip on the motherboard) version 2 which was really only standard after 2015 in new PCs. They tell you "it's all about security" - despite the fact that TPM security is known to be flawed. What they don't shout about quite so much is that its also all about things like Digital Rights Management; so you know, making sure everyone is locked in to paying for content they never really own...

A good PC is a real investment - I spent £1500 on mine so no, actually, I don't want to replace it every couple of years.

Abobe®™ have just done the same thing. As soon as I accepted their latest update (24.5) I got a super ugly 'alert' style message telling me my graphics card wasn't supported, at all. All of a sudden they can't be bothered with supporting DirectX version 11, i.e. any graphics card from before 2016 - because they're 'legacy'. I always thought 'legacy' was a thing to be honoured and respected, not euthanised and sent to landfill. Big Tech clearly thinks differently...

A like-for-like upgrade would have cost me £300. So I decided to sacrifice some features (self-correcting VRAM) and bought something more modern for £150 (GTX 1070ti). Well, it's a lovely card and my whole system feels better for it. It just seems a bit weird though, paying Adobe®™ a fee every month for the priviledge of being forced to spend more money on my PC hardware; thanks guys!

After that little shock-and-awe escapde I can now use Adobe®™'s Denoise AI, which is really very good. With my old card I used to use Topaz Denoise, an AI solution that took a few minutes for each image - which was great as it always meant I could smoke a cigarette while waiting for it. Adobe®™'s Photoshop; though, after petulantly refusing to play with my old video card, trundled away for half an hour, each image! Yeah it's way better, and so I really have to use it. I can't unsee the difference it makes and thereby not use the damned thing. But half an hour! Half an hour on a perfectly respectable machine that's more powerful than most people have available.

Big Tech always have good reasons to force our hand in these matters, good reasons for them. AI must feel like a goldmine to them - to me it just feels like an abusive relationship.

The thing is, it doesn't have to be this way. The reason Adobe®™'s Denoise AI takes 9x longer than the competition isn't just (or even) because it is better. It takes nine times longer because Adobe®™ don't care to invest the effort to make the most of the kit we all already own. Their corporate clients do change hardware every couple of years, whereas we, well we don't matter.

The invention of AI won't lead to us all having to toil away in support of our robot overlords; no, we'll all be toiling away to pay for our software 'subscriptions' and hardware 'upgrades' to the Big Tech FinDoms.

Carbon Hoofprint and the cost of ownership

Image making requires a computer of above average performance. A PC of average performance costs around £1,000 these days. So an average imaging PC likely starts at around £1,500 - what I paid for mine, but at a higher performance level by virtue of respecting legacy.

The Big Tech FinDoms (BTFDs) would have me spend this every 2 years allowing for trade-in values that's perhaps £10/wk, just to own the hardware that gifts us the opportunity to buy the software subscriptions that make us upgrade the hardware again and again and again.

It's quite beautiful, in an obscene way. The market has been engineered to syphon our earnings bit-by-bit with 'text TEN, TWENTY or FORTY' pounds-per-month responsibilities: broadband, streaming services, software subscriptions, charitable causes, engineered cost-of-ownership, insurance!

AI is just another opportunity to fleece us. This is the biggest threat AI poses to society.

And in so doing they are also, fleecing the planet.

Poorly optimised code runs the hardware hotter, draws more power, turns your carbon footprint into a carbon hoofprint; no matter what you might do to be a responsible citizen, you can't expect the same regard from the BTFDs. None of them have 'responsible use of end-user power' as a built in recquirement of their software development. They just don't care to think about it. They just don't care to think about your pocket. They just don't care to think about the planet.

'Our new feature consigns a whole generation of working and capable electronics to landfill - so what?'.

Wrap-up

Instead of wringing our hands over what AI means to our integrity and ownership of our images, we need to look under the hood at the conivings of the BTFDs to really understand the threats that AI poses to society.

Of course, I bought my new graphics card so I suppose that makes me a collaborator; in which case you can simply dismiss all of the foregoing. It was just a dream. A beautiful dream of electric sheep...