Understandably the work of documentary filmmakers faces new epistemological challenges in the wake of generative AI. There’s a good read: “Can you believe the documentary you’re watching?” by NY Times film critic Alissa Wilkinson that provides great insight into this situation (at least it did for me). One of my colleagues recently pointed me toward Eli Horwatt’s Substack article “That celluloid look of AI advertising” that examines OpenAI’s choice to advertise their AI generation services with commercials produced on 35mm film. It’s not simply a gotcha moment but something more subtle that asks us not to take for granted the shift from, as Horwatt writes, “the aggressively anodyne look visualized through AI” to the visual world of these commercials that is simply juxtaposed in the content.

These are serious concerns about our capacity to speak truthfully to one another. The epistemic crisis is a “morbid symptom” of the digital interregnum (to give Gramsci his due here). My contribution comes from a different position, though not one that fails to recognize this matter of concern. I’ll add two considerations to what is emphasized in these articles.

First… We’ve been through this before in some ways. That’s not to disregard it but to help understand it. Plato was worried what writing would do to the author and the embodied truth of the voice. And he wasn’t wrong to be concerned, but we’ve seem to have managed to struggle on. Sousa opposed the gramophone and we managed to struggle on.

But now text and sound are suspect, as are images and videos.

With documentary we have an adoption loop that is familiar in genre media studies. The first documentaries aren’t documentaries per se because the genre doesn’t exist (e.g. “Workers leaving a factory.”). The propaganda of the 30s. In the 60s its cinéma vérité. By the 90s, documentaries are regular PBS and cable channel fare.

Then the digital starts to weaken the signal. The digitization of media allows for editing of the archive in new aesthetic ways. Heck, even I could use the “Ken Burns” effect in iMovie. In the late 90s and early 2000s this was CGI, color grading, all the digital stuff. The cost of that stuff started to create a separate documentary genre. The infrastructural capacity to stream video exploded the genre. What is YouTube? What do we make of videographers/vloggers like Casey Neistat and his 12M+ subscribers? The netflixization of documentary in the 2010s resulted in algorithmic storytelling: narratives crafted for a new audience and genre governed by computation.

Now we have generative AI and it too is a cost separator, at least currently. We can say gen AI makes image generation easier and it does. But when we say that I think we say it in the way we might say that ketchup makes everything taste better. In the hands of experts and with access to the best tools and the necessary compute, we are getting something altogether different. Something better than this (which is just a brief Sora 2 demo but the guy flipping on the surfboard is interesting.)

Second… (I did say two things, right?). We’re soaking in it. When we pick up a contemporary digital camera (e.g. from our smartphones to the most expensive professional gizmos), AI is built into the firmware and hardware. The camera doesn’t operate a different way. The image is AI-generated. And we know that contemporary post-production software is AI-driven. That which was once merely algorithmic is now something more/other.

A subtle example. A contemporary film editor can help you find scenes in your content by text. E.g. “find the scene with a man eating a sandwich.” Harmless enough, but I’d say this might radically expand ones cognitive capacity to edit in a way that is not dissimilar to how “find/replace” or “cut/paste” changed text production in the 80s. It is still “just editing,” but editing used to be suspect… until we normalized. Now we have normalized editing but we’ve altered the mechanisms of editing beneath it. And we tend to glide past that.

Until now, when those gaps can be filled with AI content. Just take that ugly guy out of my shot. No, the other ugly guy. Oh whatever.

There is a clear material difference between taking a photo of a sandwich, no matter how AI-fancy the camera is, and asking ChatGPT to generate a “photographic” image of a sandwich. But it isn’t the difference between truth and fiction or the real and the fake. It’s something else.

Turning back to historical contexts, editing was once a vanguard political method. It was a propaganda tool. We went through years of minimally edited films to establish the truth value of film. As Richard Lanham put it, we shifted from “looking at” the film’s construction to “looking through” to the world it represents to us. That was/is more than the willing suspension of disbelief we do for fiction films.

It took time for us to believe in the truth of films. And maybe we really shouldn’t. But that’s another discussion. Films/video still put people in jail. It is a recognizable feature of human life that we start to believe things and then after time we think of them as true. Then we can’t even remember a time when that idea went from an active belief to an accepted truth in our minds.

It’s called energy conservation.

The claim that documentary films can communicate the truth of others (or even its own truth) is a belief. It doesn’t have to be naive. We mean ethical documentarians are committed to presenting truth and have the technical means to do so. The first part doesn’t need to change for the second part to break. The second part isn’t simply technical. It is technoculture. Does the technoculture provide the means?