Anarcho-Capitalist Kurosawa Summer
Kurosawa's influences and an LLM analogy
Austin affords very few pleasures in August. Last year I attempted to go for runs early in the morning all through summer. This year I gave up. I stick to air-conditioned spaces like the Ninja Turtles stick to tunnels. One of the air-conditioned spaces I've frequented a lot this year is Austin Film Society, which showed about 10 Kurosawa movies in July and August, and I watched six of them. I did not really intend to watch that many, but I got mildly addicted to the cathartic feeling I got after watching the first one. After every movie—regardless of whether it was a tragedy like Ran or a comedy like Yojimbo—I felt like my mind had expanded. Unfortunately the feeling was fleeting, so I had to go watch another one every couple of days.
Watching Kurosawa movies back to back is discovering the source material for screen-first media that you grew up with. For example, Hidden Fortress bears a close resemblance to the original Star Wars, particularly in the framing of the story from the perspective of the weakest links in the empire—Hidden Fortress from the perspective of peasants, and Star Wars from the perspective of R2D2 and C3PO. Here is a side by side comparison. Sergio Leone ripped off (not derogatory) Yojimbo to create the spaghetti western genre, beginning with Fistful of Dollars. Seven Samurai has probably been remade in every region that makes movies. The Hindi remake, Sholay, was the longest-running movie in India for about 25 years. Kurosawa himself was influenced by American westerns and noir pretty early on.
It’s easy to look at Kurosawa’s chain of influences in cinema and make the analogy to LLMs as a similar medium, where words written using an LLM are referencing works by others, except instead of a finite and select few, they are referencing a large corpus of texts that exist on the internet. The oft-cited criticism against this is that LLMs are plagiarizing and profiting off other people’s writing, but homages and influences are more respectful and innocent somehow.
I don’t think this is true. In a lot of cases, 20th-century filmmakers and musicians in the West and elsewhere gatekept their influences. George Lucas was not going around crediting Hidden Fortress until Kurosawa’s movies became popular in the West. Fela Kuti’s influence on David Byrne and Talking Heads was not acknowledged until ten years after the acclaimed albums came out. Even more interesting is the case of Yojimbo, directed by Kurosawa, originally inspired by Dashiell Hammett’s American noir novels. When Sergio Leone adapted Yojimbo into Fistful of Dollars he did not credit Kurosawa until he was sued, after which Kurosawa got 15% of worldwide earnings, which was more money than Yojimbo ever made for him. Dashiell Hammett did not receive a single dollar from either of the productions.
The generous reading here would be that the language of movies and writing is a commons culture that anyone pursuing either taps into, sometimes consciously as homage and more often unconsciously because the ideas have percolated in the world around you. More importantly, after a medium attains a certain level of maturity, it is hard to distinguish where influence ends and originality begins. As J.G. Ballard, one of the greatest prose writers of the 20th century, once remarked when asked about how he writes: “One simply pulls the cork out of the bottle, waits three minutes, and two thousand or more years of Scottish craftsmanship does the rest.”
When an artist produces a work they are also contributing to the common knowledge of the medium that made the artwork possible in the first place. In the case of LLMs, the writer’s work is used in the creation of a new medium, one that would potentially lead to new types of media. I think here it is also useful to connect this to copyright laws that currently protect written works and films.
In his paper Twilight of Legality, John Gardner argues that corporations preferring arbitration over public courts is akin to asking for free favors from the government, because corporations want the benefits of legality (enforceable contracts, legal protection) while resisting its burdens (accountability through courts and public process). Copyright, as it functions today, does something similar: works are locked away for the maximum possible term — sometimes effectively never entering the commons (for books by individual authors, copyright is life + 70 years) — which means they never reciprocate into the cultural commons of the medium. The legal system exists on the premise that, eventually, individual ownership yields to public domain, because it is the public taxes that make the legal infrastructure of copyright possible. But, current laws have stretched that horizon so far that for most practical purposes, the commons is denied.
Seen in this light, LLMs are an anarcho-capitalist work around for referencing culture in the same way artists always have. It is our legal system and ideas of ownership that have to catch up. Instead of entering a healthy cycle of protected work → commons → new creation, we have protection without reciprocity.
There is, of course, an asymmetry here between a multi-billion-dollar company referencing your work vs. an individual filmmaker referencing your work. But the creation of a new medium almost always requires enormous capital. Most films after the silent era were produced within the Hollywood studio system, and only in the 1950s and 60s did smaller, independent filmmakers begin to find backing. This isn’t unique to cinema. The printing press was first the domain of wealthy patrons before pamphleteers democratized it; early photography demanded costly equipment and chemicals before Kodak made it portable; radio and television began with massive networks before local stations and public access opened the door to smaller voices; video games were once the sole province of console giants before cheap engines and digital distribution enabled indie developers. LLMs are still in their studio-system phase—costly to build, requiring corporate scale—but if the pattern holds, we may eventually get a vibrant commons medium that anyone can tap into.
