Elon Musk says AI will make a 'historically accurate' Odyssey – here's why that's impossible

In July 2026, Elon Musk posted a three-minute AI-generated scene on X and promised that his AI tool Grok Imagine would produce a full-length film of The Odyssey within the year, claiming it would be "historically accurate and faithful to Homer's art." This announcement came days after the global release of Christopher Nolan's The Odyssey film—which grossed over $640 million worldwide—and Musk had spent months attacking the film as "woke."

Musk's promise directly challenges the narrative authority of Nolan's film, but its core claim—"historically accurate"—faces fundamental technical skepticism.

The Odyssey is a 2,800-year-old epic containing supernatural elements like witches turning men into pigs and cyclopes. But setting those aside, a simpler question arises: accurate to which history? There are no photographs of ancient Ithaca, no video of the Trojan coast—only stories, poems, pottery, paintings, and a century of cinematic imaginings of the ancient world. That is all the generative model can see.

When Grok Imagine renders an ancient scene, it does not reach back in time; it finds the center of all existing related images—the most likely ancient warrior is an average of Gladiator, Troy, 300, and a century of sword-and-sandal films. Thus, AI-generated images will not look like the past, but like the average of other movies about the past.

More critically, a machine that seeks the most likely image will not correct our stereotypes about antiquity but will concentrate them. For example, asking for a classical statue yields gleaming white marble—because that is how we imagine the ancient world. But ancient Greek and Roman sculptures were originally painted in strong colors; the pure white marble in our minds is a later idealization, not reality. AI models trained on our images of the past will confidently replicate our errors and label them truth. This is the opposite of accuracy—it is an amplification of our own mistakes.

"Historically accurate" is essentially not a script problem but an image problem. On-screen accuracy is conveyed through light, lenses, skin, texture, fabric, and grain—hundreds of decisions a cinematographer makes per frame. A machine also makes these decisions, but it does so by consensus, defaulting to the most probable option.

Research has found that generative AI makes individual works feel more creative but homogenizes the entire field, with everything trending toward the same mean. This regression to the mean flattens not only prose but also visuals. You get neither an accurate past nor an exciting lie; you get the average—and there is nothing memorable in the average.

The appeal of cinema lies precisely in "enhancement"—a choice about what to exaggerate—and choice is something the average cannot make. A machine living in the mean can smooth but cannot enhance. Thus, an "exciting" AI blockbuster loses the very quality that excitement depends on from the start.

Musk is not merely demonstrating technical feasibility. If he only wanted to prove technology, he would not have chosen the word "accurate"—especially for a poem, where accuracy is meaningless. Using "accurate" is a bid for authority: the right to define what is true.

Musk has publicly opposed Nolan's casting of a Black actress as Helen and a transgender actor as a Greek warrior. He operates within a broader political project about who has the right to tell the nation's stories. Now, he claims to possess a tool that can faithfully depict the past. Following this logic: if his machine determines what is historically accurate, then he decides what the past looks like. And those trusted as the "authors" of the past are also in a favorable position to define the future.

Therefore, Musk's promise should be met with skepticism—not because AI cannot make images (it clearly can), but because "historically accurate" is a claim about truth. And truth is precisely what a machine trained on the average of everything cannot provide. In this case, truth is also what its owner most wants to define.