A.I. algorithms had to change when COVID-19 changed consumer behavior

But once the COVID-19 pandemic hit, A.I. systems had to be overhauled. The software, it turns out, had never been trained with data reflecting a drastic upheaval in consumer behavior caused by the virus and therefore couldn’t adjust to the new reality.

“A.I. is actually a living and breathing engine,” Lan Guan, managing director of applied intelligence at Accenture, said on Wednesday during an online discussion about A.I. organized by Fortune.

Many companies using A.I. these days are learning the hard way. It’s just one of the wrinkles of using a technology that is only as good as the data fed into it.

Clothing retailer Stitch Fix saw huge shifts in the preferences of its customers, many of whom were forced to work from home during the pandemic, said Tatsiana Maskalevich, the company’s director of data science. In just one example of that shift last quarter: Sales of athleisure like yoga pants, tights, and leggings surged 350%.

At Stitch Fix, A.I. helps recommend clothing to thousands of human stylists who ultimately decide which items to send to customers. The dramatic shift in what shoppers wanted due to the pandemic meant that those stylists had to better anticipate customers’ taste in clothing because the algorithms were out of the loop.

“Having humans there and having that personal connection helped us to really bring our clients along,” Maskalevich explained during Fortune’s event.

At Facebook, A.I. systems that translate posts into different languages were suddenly hit by unfamiliar terminology during the pandemic, Joelle Pineau, co-managing director of A.I. research at the social media company and a computer science professor at McGill University, said.

“The language of COVID is a new language for many of us,” Pineau said. “A machine-translation system did not have these terms before, so we really had to stretch what our A.I. systems were able to do to adapt very quickly to this new realty.”