Shonen Jump owner Shueisha’s recent survey on AI translations once again got people discussing the use of AI-powered machine translations for manga, bringing up all the valid concerns around noise and fuel pollution, job loss, privacy, mental health, and theft of intellectual property. But beyond all of that, the simple truth is that AI large language models are fundamentally ill-suited to translation and localization work in general.
Before any AI evangelists interject, no, this is not a case of it needing more training data or time to work out the kinks. It is a problem with the process of how machine translation works at a base level and its incompatibility with the task at hand. Despite the repeated claims by tech company CEO’s, AI is not conscious or aware. It is intelligent only artificially. It looks at a word and replaces it with another word marked in its database as being similar, and any meaning that is maintained (or injected) is purely accidental. An AI translator cannot judge whether the word or sentence is a suitable replacement or how well it carries the connotations. It cannot judge how well the meaning will be conveyed because it cannot actually judge meaning at all. Which is a problem, as the meaning is the entire point.
Screenshot by Siliconera
Think for a second about how informationally dense language is. All the tone, themes, meanings, references, and wider context all affect each other. It is pretty much impossible to perfectly translate something as complex as a story without leaving something behind. A big part of a translator’s job is essentially triage. Deciding what meaning is most important to preserve and what might need to be left behind. Translator Sarah Moon illustrated these complexities and trade-offs in her video on the Chihayafuru anime, analyzing some of the contentious choices the translators and localizers had to make in the name of legibility. An AI translating manga, or any medium, does not have the capability or judgement to make such decisions.
Go Ad-Free With Siliconera+View the site with no ads, be able to provide direct feedback about what we cover, and be automatically entered into our monthly PC game giveaway.Join Siliconera+
But limitation can also spark innovation, and human translators can use lateral thinking to bridge the gaps between cultures in creative ways. While an AI translating manga can only use what it was trained on, humans draw on their own lived experience or create entirely new ideas to help convey very specific ideas across cultures. For another example, let’s look from manga to video games. Specifically, Dragon Quest VIII and the character Yangus. In the original Japanese, Yangus speaks like a yakuza thug and refers to the protagonist as “aniki,” or brother. In the English localization, he speaks with cockney or estuary accent that not only communicates his criminal past (through association with London gangsters), but also his simple nature (association with the poor or working class) and his role as comic relief (British accents are just inherently funny). It’s an inspired choice, honestly. Imagine how much worse off we would be if he had been given a Chicago gangster or Italian mafia accent instead.
These are the kind of nuances we’d lose out on if Shueisha turned to AI translation for manga for series found in Shonen Jump. Even if non-AI machine translation methods will likely remain widespread for “quick and dirty” translations of very small and simple things, AI models have far too many wide-reaching drawbacks to be tenable. For anything more complex than a tweet (and even that can be pushing it), human translators are a necessity, not a luxury. To paraphrase from an article by Jost Zetzsche for the American Translator’s Association, “Machines translate words, but humans translate meaning.” And if you aren’t translating for the meaning, what is even the point?
The Shuiesha Manga Plus survey concerning AI translations is open until September 27, 2026.