It's manipulation all the way down: AI images and the epistemology of the real
Adobe Photoshop’s Sensei, Open AI’s Sora, Stable Diffusion, Midjourney, and DALL-E 3 all allow amateurs and professionals alike to generate and edit images with the support of artificial intelligence. But image manipulation is far from a new phenomenon. We need only look at the analogue photo editing of Stalinist Russia or the monumental cinema of Cleopatra (1963) that used matte painting effects to see a long tradition of altering, constructing, and editing images for both entertainment and political ends. For many, the fear is that these modern AI tools might be able to construct images, audio, or video footage of such surpassing finesse that they would be indistinguishable from reality. That we might be duped, by software, into believing that which is not true. In the hands of those with a desire to undermine democracy, extort, blackmail, or intimidate this could well be a frightening state of affairs.
I think, however, that whilst such a fear is well grounded it might not be where the real locus of concern lies. As an example, we can think of the recently released, and retracted, image of the Duchess of Cambridge and her children. Some mystery and conspiracy swirl around her current situation, where she has been, and what might be happening, and then an image is released that purports to be a casual family snap for Mother’s Day. Very quickly it is shown that this image has been manipulated. It doesn’t seem as if it has been AI-generated, which might be a bit of a stretch for now for even the most cutting-edge image-generation models, but good old Photoshop has played a role in tweaking this photo. What this says to me is that when people place their attention in the right place we can figure out if images have been digitally manipulated - whether through AI generation or photo editing. Maybe the casual observer might not pick it up, but certainly, people with expertise will. We have to imagine that the skill and sophistication of these images will increase - maybe even rapidly - but the images with the biggest ‘claims’ will always draw the greatest scrutiny. This leads me, then, to what I think is the deeper problem.
What happens with all of the images we’re not paying attention to?
If bad actors decide to release AI-generated video or images of Joe Biden doing something that would sink his chances at a second presidential term then the world’s attention will fall upon them and every pixel will be minutely dissected by the best in the field. If images circulate purporting to show microbes in the soil of Mars, they will be examined forensically by the world’s experts. In both cases, I suspect the deception would be uncovered. But what happens when it’s not the headline-grabbing images that are manipulated, invented, or created by AI? We’re taking in hundreds, even thousands of images a day - the vast majority of which we’re giving no second thought to. It’s here that we’re most vulnerable to this kind of manipulation.
The epistemic question is not whether we can be fooled - we almost certainly can. But rather when are we being fooled?
Far more dangerous than AI-generated photos of Biden punching a homeless person, or bacteria in red soil would be those images that it doesn’t even occur to us to think critically about. A 30-second ad break, a grainy picture in a newspaper, an Instagram post. Anything that reinforces our ideology will fly right past our defences. A photo of Trump inappropriately touching a waitress? Well, of course, I’ll believe that. It fits my picture of the world perfectly so why would I bother to question it? Worse still, will be those images that don’t even seem to hold significance but fill our everyday lives. Images in advertising that subtly reinforce stereotypes, prejudices, or forms of discrimination. Images that confirm my sense of identity. Images that depict the world in the way I already imagine it to be. There’s just as much distortion happening here, and it may well be far more insidious than the grander forms of manipulation we might be more sensitive to.
This epistemic problem taps into many existing, and well-studied, issues in psychology and philosophy. Implicit and unconscious biases, confirmation biases, the halo effect, and more. What marks this new problem out, however, is that AI image manipulation and generation can occur at a scale we’ve never yet seen, and can iterate, improve, and target us in ways far beyond anything that has come before. Much of the discourse around AI images will inevitably focus on fears that some great conspiracy may falsify the truth in some seismic, dangerous way. Of course, that’s possible, but that’s precisely where the greatest expertise can and will be leveraged to expose the fraud. Far more toxic will be the flooding of our visual world with manipulated images in precisely the places we are most susceptible to influence.
Every time we spot one of those grand gestures of manipulation, a Kate Middleton with the family photo for example, we will congratulate ourselves on having spotted the trick. On not being caught out. Like a magician, these gestures can be made precisely to cover up the real sleight of hand. We trick ourselves into thinking we know where the deception is and then leave the door flung wide open to all of the rest of the lies and deceit. In Sumlacra and Simulation, Jean Baudrillard makes the same argument about Disneyland,
“Disneyland is presented as imaginary in order to make us believe that the rest is real, whereas all of Los Angeles and the America that surrounds it are no longer real.”
We mustn’t make the same mistake here, we mustn’t come to believe that because some images are palpably fake the rest are real. And here lies the true epistemic knot of AI imagery - no longer is it the case that images are distortions of reality but that the distortion does not even have a reality behind it. In the case of the Kate Middleton photo, there was, we must assume, a real photo - however manipulated - and so a real truth was being obscured. Of course, even if the photo had been pristine - it was still manipulated. It had been posed, primed, styled, and released through sanctioned channels. With AI imagery, though, there is no underlying truth. It is simulation all the way down.
“Simulation is no longer that of a territory, a referential being or substance. It is a generation by models of a real without origin or reality: a hyperreal.”
How do we overcome this? For Baudrillard the only solution is to return to the local - to ground our understanding and existence of the world in the locality we find ourselves in. To connect truth to being, and to connect meaning to presence.
“Meaning, truth, the real cannot appear except locally, in a restricted horizon, they are partial objects, partial effects of the mirror and of equivalence. All doubling, all generalization, all passage to the limit, all holographic extension (the fancy of exhaustively taking account of this universe) makes them surface in their mockery.”
For a fundamentally globalised medium such as the internet, that may prove challenging. But not impossible. The internet, in its early days, was far more local. Message boards, chat rooms, and physical networks all connected users in localities - digital or physical - in a way that they no longer are. Though Baudrillard was far less optimistic about our chances.
“The apocalypse is finished, today it is the precession of the neutral, of forms of the neutral and of indifference. …All that remains, is the fascination for desertlike and indifferent forms, for the very operation of the system that annihilates us. Now, fascination… is a nihilistic passion par excellence, it is the passion proper to the mode of disappearance. We are fascinated by all forms of disappearance, of our disappearance. Melancholic and fascinated, such is our general situation in an era of involuntary transparency.”
The film that was, perhaps, most explicitly inspired by Baudrillard’s work was, of course, The Matrix. A film in which the protagonist can ‘wake up’ and finally see the real, on one hand, and the virtual, on the other. Neo’s awakening was not without difficulty, but he eventually triumphs over the world of images much like Plato’s prisoner finally escaping his cave. If we think about that last quote of Baudrillard’s, however, we find no such triumph. Instead, he sees a melancholic disappearance and an involuntary transparency. To me, this reads far more like Truman in the Truman Show than it does a latex-clad ninja fighting his way out of a computer. What Truman realised is that there’s no truth lying at some deeper level of reality beneath that false veneer we’ve uncovered.
There are no blue pills or red pills. Instead, there’s only taking oneself out of the game.
Let’s, then, absent ourselves from these structures that allow for our manipulation. When we give up our attempts to expose the ‘truth’ and instead focus ourselves on interpretation and analysis, we can immunise ourselves from these manipulations. The question stops being, what truth are the Royal family trying to hide with this photo, and instead, what’s the narrative being presented? What story am I being sold? Who benefits? There never was any truth behind these pictures, that isn’t how truth works, so let’s not think that we’re only now entering an age of distortion. Instead, we need to recognise that truth emerges from practical concerns and activities in the world and doesn’t sit, crystalline and perfect, waiting for us to uncover it. Once we relinquish that myth, the debilitating venom of AI manipulation can do us far less harm.