What Machines Cannot Remember

Dario Amodei, documentary filmmaking,

and the irreplaceable act of choosing what matters

In January 2026, Dario Amodei — the CEO of Anthropic, the company that built the AI assistant I sometimes use to check my own work — published a twenty-thousand-word essay arguing that humanity is sleepwalking into a crisis. Not the cinematic crisis of sentient machines. The quieter, more corrosive one: that artificial intelligence will become better than humans at most intellectual work faster than our labor markets, our institutions, and our political systems can adapt. He called it "unusually painful." He predicted that up to half of all entry-level white-collar jobs could vanish within five years. He used a phrase that has stayed with me: AI is not a substitute for specific human jobs but "a general labor substitute for humans."

I have been making documentary films for more than thirty years. I read that sentence the way you read a diagnosis you were half-expecting.

Dario Amorei | Photo-Illustration by TIME (Courtesy Photo)

The entry-level problem hits documentary filmmaking with particular precision. The researchers who spend months in archives. The assistants who log footage at two in the morning. The editors who build rough assemblies that no one will ever see but that teach them everything about structure. The writers who draft treatments that get rejected and rewritten and rejected again. These are the roles AI is absorbing first — not because the work is unimportant, but because it is, in Amodei's framing, classifiable. It can be described, broken into tasks, automated.

What concerns me is not my own position in this landscape. After thirty years, a filmmaker either has a point of view or doesn't. What concerns me is the pipeline. Documentary filmmaking, like most crafts that involve judgment, transmits itself through apprenticeship. You learn to see by being in a room with someone who sees differently than you do. You learn what a film is about by spending a year on something that turns out to be about something else entirely. If AI hollows out the entry-level infrastructure of the field, it doesn't just eliminate jobs. It breaks the chain of transmission.

My own practice has always been built around what I think of as the counter-archive: the recovery of figures and moments that institutional memory has chosen, consciously or not, to forget.

George Frederick Bristow, the nineteenth-century American composer written out of the canon. The Piccirilli Factor, about the Italian immigrant sculptors who carved the Lincoln Memorial. Eduardo Bradley, who crossed the Andes by balloon in 1916 and was celebrated for a moment and then filed away. In each case, the work begins with a question that no algorithm generates: why does this absence exist, and what does it cost us?

That question is ethical before it is archival. It requires the filmmaker to have a theory of what cultural amnesia looks like, where it comes from, and who it serves. AI can search every digitized newspaper from 1916. It cannot decide that the search is worth conducting. It cannot feel the weight of a name that should be known and isn't.

This is not a romantic defense of human exceptionalism. It is a practical observation about where the work actually lives. The labor AI is absorbing — logging, transcribing, aggregating, summarizing — is real labor, and its loss will hurt real people. But the labor it cannot yet touch is the interpretive act at the center of documentary filmmaking: the decision that this story, and not another, is the one that needs to exist.

Amodei's deeper worry, beneath the labor displacement numbers, is about concentration. That AI accelerates inequality — that it pools power and wealth at the top while eroding the economic stability that makes independent thought possible.

In documentary terms, this is not a distant scenario. It describes a future in which large platforms deploy AI to produce vast quantities of algorithmically optimized content — historically shallow, emotionally calibrated, efficiently distributed — while the infrastructure that supports independent documentary work contracts around it. The libraries, the foundations, the distributors, the programmers who believe that a film about a forgotten composer or a balloon crossing of the Andes is worth putting in front of an audience.

The threat is not that my films become obsolete. It is that the ecosystem that makes them possible — and that makes them findable — gets disrupted faster than it can reorganize.

Joaquin Sorolla (another subject of my interest) spent years trying to paint what Spain actually looked like — not the Spain of postcards or propaganda, but the Spain of light on water, of working fishermen, of a country that contained more than its reputation. He was doing, in paint, what documentary filmmakers do with a camera and thirty years of accumulated obsession: insisting that the record be more complete than the official version.

Amodei is right that we are entering a passage. But documentary filmmakers have always worked against the compression of meaning — against the tendency of the present to flatten what came before into usefulness. We slow down. We stay with a subject. We ask questions that take years to answer.

That is not inefficiency. That is the work.

What machines are becoming better at, faster than our institutions can process, is the production of plausible knowledge — summaries, assemblies, narratives that hold together without necessarily being true to anything in particular. What they are not becoming better at is the prior act: deciding what is worth knowing, and why the forgetting happened, and what it would mean to remember it correctly.

That act is still ours. For now, the question is whether we are building the conditions — institutional, economic, educational — that will allow the next generation of filmmakers to inherit it.