Artificial intelligence is going to be big in the world of fashion — at least that’s what startups, retailers, and investors are promising. They say the potential for AI to process data about your clothing and come up with new suggestions of what you might like to wear offers big opportunities. But, when it comes to actually turning this digital data into physical garments, it seems fashion and AI aren’t quite a perfect fit.
Take San Francisco startup Original Stitch. The company sells custom dress shirts, letting customers tweak colors and patterns online. As part of a beta service, it also let people send in a picture of their favorite shirt, which the company said it would measure up using machine vision; extracting the sizing information and creating perfect replicas on the cheap.
It sounds fantastic, but the results didn’t match the sales pitch. Reporters from Bloomberg ordered a number of shirts from Original Stitch’s AI tailor, but each time, it delivered disappointing goods. Shirts were too tight, had sleeves that were too long, or could barely be buttoned up at all. So, Original Stitch has shuttered the AI service while it works out the kinks.
It’s a failure, but it’s an instructive one. Although AI is extremely good at processing data in a 2D realm, it’s another matter altogether to apply this information to the 3D world of textiles, which deform and warp in all sorts of difficult-to-predict ways. It’s for this reason that a machine built to fold your laundry took years to develop, costs $16,000, and breaks at the drop of a hat. It’s also why one of China’s biggest textile manufacturers said just this year that it would continue to rely on human workers — not robots — for the foreseeable future. Clothes are hard.
But most likely this will prove to be a temporary setback, with AI systems getting better at handling textiles the more experience they have. The founder of Original Stitch, Jin Koh, told Bloomberg that the company was already improving its machine vision measuring tech, and that a working version could be ready by the end of the year. It will require three photos of each shirt instead of one — more data.
Eventually, said Koh, “technology is going to take over a tailor’s job and do it better.” Let’s see it get the shirts right first.