Designed by Data: Fashions With an AI Eye

The machines aren’t coming to fashion design — they’ve already arrived.

And while artificial intelligence isn’t ready to cut the human out of the design equation — yet — the success of Stitch Fix’s “hybrid designs” and percolating interest in AI across fashion highlight the potentially transformative power of mathematics and data crunching in creating style that sells.

Machine-learning technology is already widely used in fashion, particularly in merchandising and inventory allocation, where algorithms help many companies manage backend operations.

Now computers are getting creative — and using Mother Nature as a guide.

Stitch Fix, the Silicon Valley subscription box service, started using genetic algorithms and AI last year to create garments its customers wanted, but that the brands it works with weren’t providing.

The program started with three hybrid blouses last year that led to 15 looks earlier this year and another 18 for this fall, including two in its plus-size business. The styles are not high fashion, but according to the company, highly successful.

Stitch Fix takes data on its users’ style preferences, items kept from each delivery or “fix,” feedback on returned items and social media posts, and then feeds it into its algorithm.

“You take all this rich feedback and you can diagnose which parts of, let’s say a shirt, that [users] are liking and not liking,” said Eric Colson, the Netflix alum who serves as Stitch Fix’s chief algorithms officer. “The machine algorithms are conceiving these styles. The process is very much the same process Mother Nature uses, evolution by natural selection.”

A hybrid look might have three parents, for instance, taking some lacy detail from one, the silhouette from another and the pattern off the third.

The technique borrows from natural processes in several key ways.

• Selection: Successful attributes are chosen when customers give positive feedback.

• Recombination: Attributes are passed on to hybrid designs not directly, but in combination with other successful looks.

• Mutation: The styles subtly change as the process continues.

The end result is overseen by Stitch Fix’s internal design team to make sure the looks are “cogent,” and Colson argued that together, human and data smarts are a more a powerful combination than either one on its own.

“It’s really blending art and science,” he said. “When I say art, I mean intuition; when I say science, I mean empirical decision-making. Either one has value on its own, the two combined are really interesting.”

By way of example, he noted that the world’s best super computer can beat a chess grand master on its own, but not when the player has a little bit of computer help as well, handling some of the raw data crunching and freeing the human to focus on being more creative.

Stitch Fix, which is in the process of patenting its hybrid design technology, is on the leading edge of this kind of work for design.

Others are focusing further down the supply chain.

Justin MacFarlane, Macy’s Inc. chief strategy, analytics and innovation officer, told investors in June that, “There is a lot of noise out there in terms of artificial intelligence, machine learning. And the question we always get is, ‘How do you apply all of this?’”

For Macy’s, he said the answer is to focus on pricing, inventory and the customer.

Despite Stitch Fix’s early success in its hybrid program, AI-backed design is still seen in its infancy in terms of the broader market.

“This is the kind of thing that will start to evolve into something useful,” said Catherine Iger, founder of Fittery, a fit tool that uses algorithms and big data.

“It is really the future of artificial intelligence that is interesting and sometimes controversial — a future where humans create AI that begins to make decisions on its own…and ultimately takes over the world in a dystopian future,” Iger said. “We are many steps removed from that.”

But the way toward a future where designers use AI much more is starting to become more clear — even if the way forward will require a new lexicon and a new thought process.

“It’s all geometry, it’s curves, points and lines,” said Francis Bitonti, chief executive officer of Studio Bitonti, which is working on a new product that will help brands translate their styles into a digital form that computers can read and then apply machine-learning tools.

Bitonti said machine learning can help designers take an “exhaustive look at the space of possibilities:” for instance, coming up with more than 100,000 possible designs to solve a certain problem and then cluster the results into different groups.

“The future role of the designer becomes, how do you frame the problem [for the computer]? Because that’s how you define your solution space,” Bitonti said.

That’s a very different process compared to what designers use today, but perhaps one that will be necessary for a future with more personalization, where companies are not coming up with a composite character to target (i.e., the Polo girl, or the Abercrombie & Fitch guy) but designing for each of their customers as individuals.

“This is going to be a new experience for the customer and it’s going to raise customer expectations,” Bitonti said. “Any brand that gets on top of this design methodology is going to be able to marry those worlds of data perfectly together and your customers are going to demand that.

“If I were to put a mark on this, I would say in three to five years you’re going to have some fundamentally different shopping experiences as a result of these technologies.”

But does this lead to a world without human designers?

The consensus seems to be no.

“Design is really a process — a series of interconnected steps and tasks from concept to consumers’ closets,” said H. James Wilson, managing director, Accenture Research. “Some of the tasks are data-intensive, repetitive and better handled by AI: think about analyzing global trends in fabric costs or Chinese consumer pattern preferences in real-time. AI can take those steps without people. In fact, getting machines to do more of this kind of work can be a good thing for human designers, who now have more time and attention to handle creative work tasks.”

Some of the steps in design are best performed by people, he said, pointing to adjusting how a style looks on a model, which requires “a sense of taste, manual dexterity, the ability to improvise and build social consensus with colleagues around what works and what doesn’t.”

Artificial intelligence can take bits and bytes from everywhere and come up with unexpected design combinations, but it doesn’t have style.

“Smart machines are just as good at the ‘yuck factor’ as they are at factor analysis,” Wilson said. “They can quickly generate a lot of ugly designs, but have no general sense of why they’re ugly.”

“The future of design is about human-machine collaboration,” he said. “And ultimately humans must decide the difference between whether an idea is fashionable — or a Franken-style.”

More From WWD: 

Smaller Deals Seen Fueling Luxury M&A

Revlon Repositions Almay, Ramps Up on Amazon

load comments
blog comments powered by Disqus