In the last few months there has been a proliferation of articles about data-driven algorithms and how they are ruling and governing our lives. Fashion is actually one of the main industries where algorithms can be successfully employed: the purchasing behaviour of global shoppers, the items they search, like, recommend or pick, are indeed valuable data that, once screened and processed, can help defining trends and drive sales.
Yet, while Artificial Intelligence may be ruling the e-commerce side of things, robots may find it hard defining style and quality. The proof of this statement? Google's Project Muze.
Designed in partnership with European e-commerce company Zalando and with U.K.-based digital design studio Stinkdigital, this experiment - recently presented at Bread & Butter - employs Google's network-powered DeepDream computer vision program, is based on its open-source platform TensorFlow and consists of an algorithm modeled on the human brain.
The Tensor Flow system draws indeed information from a neural network trained on various design preferences including colour, style and texture as suggested by over 600 fashion trendsetters, by data included in the Google Fashion Trend Report and by styles that have trended on Zalando.
Want to play at being a fashion designer? Answer the questions the system suggests you regarding your gender, age, favourite music and current mood, and then the computer will attempt to generated creative decisions and produce a virtual outfit.
The main problem with the machine learning system is that, no matter your choices, this complex search generates more or less the main shapes and silhouettes over and over again, all of them rather cringing.
Two designs prevail: most unisex/men choices seem to produce a loose top and pants; most unisex/women choices will get you a mountainous bell-shaped construction with no holes for arms and some giant ruffles added around the hemline. Quite often the items are also covered by a disturbingly giant manta ray-like cape-cum-poncho floating in space, inspired by smoke (according to the site...).
There are occasional variations with mini-dresses covered in protruding spikes or tops and pants wrapped in bizarre ropes (mind you, the latter may turn useful in our urban jungles…), while patterns are simply crazy and could be filed under "animation" rather than textiles, even though they don't even look half as good as the ones you see in the clothes of characters in current animated feature films or in some experimental videos showing the latest wonders of computer graphics.
Surely this is not Haute Tech, but Haute Mess, the stuff of your worst acid-induced nightmares being conjured up on your computer screen (though these outfits may prove a success if you're trying to create Comedy of Manners costumes for an audience high on hallucinogenic substances…)..
You seriously wonder what's the point of wasting resources on this when there are designers out there that are successfully developing a path towards the future of fashion by finding a way to combine hand-made and machine-made, in the same way as architecture has introduced innovative shapes by combining the human touch with parametric design.
As it stands Project Muze proves that, by searching through our favourite things and moods, "deep learning" algorithms may be able to analyse our collective buying trends and maybe even make general predictions, but they will not be able to generate the next big thing design-wise because they can't inject emotions and a much-needed human touch in fashion.
In a nutshell, if this is the result of the potential of algorithms in clothing design, fashion creators out there can sleep sweet dreams, as machines will never steal their jobs (well, there's vapid celebrities and their "collabs" with fashion houses/brands doing that already anyway…).
Yet, while designing clothes is not for algorithms, there is a place for them in the fashion industry: Condé Nast has just turned to IBM's Watson to help building social influencer campaigns for brands advertising with publications such as Vogue, Vanity Fair and The New Yorker.
Brands will able to use data to understand which social media celebrity is their match made in marketing heaven and targeting in this way their audiences. Hopefully, the results for them will be slightly better than when IBM's Watson was used to help designing the Marchesa gown for the Met Gala...