Anime lessons in the limits of AI

Another leap forward for ChatGPT's capabilities is, inevitably, accompanied by a wave of people using these capacities to produce knock-offs of work by great artists.

- Stephen Bush

The latest artist to have been the topic of media commentary and much of OpenAI's own promotion – its chief executive Sam Altman has even changed his X avatar accordingly – is Hayao Miyazaki, co-founder of Studio Ghibli.

Social media is awash with images purporting to be in the same style as Miyazaki's work. I use the word "purporting" quite deliberately because most of what is produced "looks like" the output of Studio Ghibli in the same sense that I "look like" Will Smith. The colour and the shape (to a rough approximation) are about right. But take another look with anything resembling care and attention, and what is clear is that the two look nothing alike. Characters do not make eye contact with one another.

Patterns of light and shade, or detailed depictions of the imperfections in wood or stone, are largely absent.

Generative images can make something that is hard to understand concrete and visible.

But they also show us risks that we need to guard against in ourselves.