top of page

Photographic Truth at the Advent of AI

Artistic representation of the world has always been akin to magic—a conjured illusion facilitated through talent and practice. But art is artifice, and the evidence of its manufacture can’t be entirely masked. Nor does it need to be—aside from the considerable pleasure of color and shape alone, it’s often the illusion that makes art remarkable. It is the attempted deceit itself that delights.

Photographs are special in that they have adjacency to an actuality—a real thing, place, or person but in a different space and a previous time. The power of photography grows from this connection to the real world. A photograph is a moment in time captured forever, even when enhanced by the airbrush or manipulated in the darkroom. But we tend to believe in images even when we shouldn’t, and even as technology, year upon year, makes greater folly of this instinct.

Digital manipulation of images began in earnest decades ago, and with it came the possibility of undetectable fakery. It has been a source of consternation ever since. I’m thinking of this in relation to the current freakout over artificial intelligence. There’s enormous hype about ChatGPT at the moment, and the corresponding image-generating AI, DALL-E 2. Some of that hype is deserved, but most isn’t. I examined the results of a single prompt entered into DALL-E 2 to generate an image.

A freeze-frame photograph of a horse galloping on a racetrack ridden by a black jockey

I selected these words with a particular historical series of photographs in mind, and specified a black jockey for no other reason than being a fan of Jordan Peele.

First, I tried the phrase in a Google search and it returned the image I was thinking of, which is what Google is supposed to do.

Eadweard Muybridge's Horse In Motion photographs
Eadweard Muybridge's Horse In Motion photographs

Here's DALL-E 2. It's not terrifying because it's so good, it's just terrifying. :)

DALL-E image generation from textual input
DALL-E image generation from textual input

It’s fair to say that if you want to fake an image in 2023, a digital photo editor like Photoshop is still the tool of choice. On the other hand, if you aim to conjure a scene that has never existed but is indistinguishable from a photograph, you might have to put in some significant effort or just wait until 2024. Or maybe even longer, but you won’t have to wait forever. In this not-distant future, we may sometimes mount great resistance to being fooled by convincing but mendacious imagery. In other cases, we will be complicit in our own deception. It all depends on whether the message reinforces our existing beliefs, or if we sympathize with the messenger. We’ll choose to believe in the lie, or we won’t. In this respect, not even perfect AI image generation will change our behavior all that much. The truth will still be accessible if we want it. And besides, we’d all benefit by being a little more skeptical of what our eyes tell us anyway.


I searched for evidence to validate this opinion and found a good article. It’s from two years ago, but to my knowledge, human nature hasn’t changed in the interim.

And here’s an excellent primer on AI image generation.


bottom of page