Almost every day I see a new generative art project creating text, images, audio or video.

There are some which seem useful (InteriorAI, for example, can create alternative themes for styling a room), and of course writing tools like Jasper have raised a lot of money to fund development and marketing.

I remember, almost forty years ago, trying a rudimentary music generator written in BBC BASIC. Of course there was no machine learning involved. As this ancient article explains, the software worked on the principle that there are some simple rules of music which can be applied to avoid mistakes that make music sound bad. However, even though it followed the rules, the generated music wasn’t really interesting.

By now we can train computers using massive rulesets to generate content and it is well-formed according to the rules. There are no grammatical errors, pictures are sharp and detailed, and video is smooth. But is the content interesting? Does it bring a perspective that provides insights?

Certainly products such as InteriorAI can help to speed up a human creative process, just as PhotoShop provided a huge win by allowing people to try many different image effects very quickly. Computational photography has clearly raised the production values for everyday photos taken on phones; the average photo looks better with fewer blurred faces, closed eyes, and so on.

I’m sure that most people building generative tools do not think they’re creating a substitute for hand-crafted art, just the same as fast food companies know they’re not producing the most creative or original food.