Amazon has an algorithm that designs clothes by replicating human creativity

Chanel’s runway robots feel a little less outlandish every day.
Chanel’s runway robots feel a little less outlandish every day.
Image: Reuters/Charles Platiau
We may earn a commission from links on this page.

At its core, Amazon’s goal is to remove as much friction as possible from a customer finding what they want, ordering it, and having it land on their doorstep. Its ambitions started with books, but have since extended to just about every other category, including clothes.

While it works to achieve a frictionless future in fashion, the company is testing a way to design new clothes without human designers.

MIT Technology Review reports that a team at the e-commerce company, working at a San Francisco research center, has devised an algorithm that analyzes images to learn about specific styles of clothing. It then creates similar new items from scratch. The technology is still in early stages and not ready for Amazon to put into use, but it shows what the company might have in store for the future.

To do it, the team uses an AI technique called generative adversarial network (GAN), originally developed by a Google researcher. It works by having two deep neural networks look at examples of existing fashion, and then go back and forth with each other, like a designer-critic team, to create something new that matches certain criteria. A shirt, for instance, must have two sleeves. It learns from past human creativity in fashion and, in a sense, replicates it.

It’s not entirely clear how Amazon would put this technology to use. Theoretically, it could analyze what people are buying and see where there are holes in the market, using the new AI to generate designs for its many private fashion labels that could fill those gaps. Stitch Fix, a subscription clothing service, is currently using AI for that purpose and says the sales have been great.

Stitch Fix does, however, use human designers who turn the AI’s suggestions into an actual garment. Whether Amazon’s technology would require that kind of guidance is uncertain, but if not, it raises the question: Could the algorithm tell if a look it generates is a dud? An Amazon spokesperson said the company has no additional information it can provide.

It’s perhaps telling that Amazon has also been working on AI that can judge whether an outfit is stylish. It recently announced Echo Look, which lets users take pictures of their outfits using a camera powered by Alexa, the company’s interactive voice service. Among its features is the option for users to get a second opinion on an outfit they’ve photographed, using machine-learning algorithms combined with advice from fashion specialists at the company.

MIT Technology Review also notes that a group of Amazon researchers in Israel have machine learning that can analyze a few labels attached to images, in an environment like Instagram for example, and determine if a look is stylish.

Amazon’s ambitions don’t end at clothing design. Earlier this year, it received a patent for an on-demand clothing factory that would let it manufacture clothes quickly, and only after they had been ordered, saving the sunk costs of producing up front.

Imagine a future where Amazon allows you to shop a selection of AI-designed clothing that it only produces after someone buys it. The scenario might never come to pass, and if it does, it’s still a long way off. But don’t put it past Amazon to consider it.