Beyond True and False — How AI Rewrites the Language Games of Images

Here we are: another newsletter about images—after the interview with the photographer Rhiannon Adam and a piece on how news photos depict heat waves.

It’s odd, for me, to have this interest in images: I express my thoughts in words, not pictures. Of course I take photos, as everyone does, but mainly for "personal use" (happy memories in the form of mediocre images stored on my phone) and my drawing skills are rather poor. Still, I'm interested in how public knowledge is made, and images are a core part of it. So, I think it is important to understand the place of pictures in our shared knowledge, what role they can play in the great game of public discourse. And whether AI-generated images can serve the same role, work differently, or completely overturn the existing rules.

The cornerstone of this topic is that AI-generated images are at least changing the level of the game. A terrific, and perhaps terrifying, example is the AI propaganda used in particular by far-right movements. President Trump has posted AI-generated content dozens of times on his Truth Social account: Stuart A. Thompson has analyzed this AI bonanza: cosplaying as the pope in one fake image, watching agents arrest Barack Obama in another, standing atop a mountain having "conquered" Canada in a third (observing a mountain that strikingly recalls the iconic Matterhorn, a peak that has nothing to do with Canada or America). Political experts note that even the most anodyne uses of AI by the president normalize these tools as a new type of political propaganda. "It's designed to go viral, it's clearly fake, it's got this absurdist kind of tone to it", says Henry Ajder, who runs an AI consultancy. "But there's often still some kind of messaging in there". It redefines – or in some cases discards – the idea of being "presidential".

But in this newsletter I’m interested in a more general reasoning: trying to understand if, and why, AI-generated images—or “images generated using AI”, because AI is an instrument in the hand of a human being that has decided to make or share them—are false and what count as a “true” image.

What We Talk About When We Talk About Image Truth

So, the starting point is the truth of images. But facing this question is like opening two Pandora's boxes full of philosophical troubles.

The first box concerns the concept of truth itself: what makes some of the things we say or think true and others not? Long story short, you'll find theories of correspondence (a statement is true if it corresponds to reality or facts), epistemic theories (which study how we obtain or justify statements we consider true) or deflationist ones (according to which truth isn't a real property of propositions, but just a linguistic way to emphasize what we're asserting). Very important, and interesting, are also all the analyses on lying and how it's possible to deceive even by saying something true.

The second Pandora’s box concerns images: even ignoring the question of the nature of these objects—and also setting aside the complicated status of mental images that a significant part of the population can't even form—there remains the question of whether images can be bearers of truth. In other words: are images one of those "things" that can legitimately be true or false? Ernst Gombrich, the great art historian, was certainly right when he said that images aren't propositions and therefore can't be true or false, just as a statement can't be blue or green (he wrote that in an essay contained in Art and Illusion). Traditionally, being true or false concerns declarative (apophantic) statements and their propositions, excluding linguistic expressions like commands or questions; it would be strange to include images instead.

However, we commonly speak of true images and false images and it's important to understand why (and whether it's a legitimate use or one of the many "imperfections" of our natural language).