The company pulled the offending item, which first surfaced at a Trump rally, almost immediately after an advocacy group complained. Teespring, the third-party seller that posted the shirt on Walmart’s marketplace, did the same.
“As soon as we were alerted to this content promoting violence against journalists, we removed the content, added this content to our automated scanning systems, and kicked off a human sweep of the site to find and remove any similar content,” Teespring told the Associated Press.
That’s a fine idea, but this isn’t the first time Teespring has run into trouble. In August, the apparel platform offered T-shirts decorated with swastikas alongside words such as “love,” “peace” and “zen” on rainbow backgrounds. This was part of an effort by a designer to “share the beauty of this symbol detached from the Hatred associated with it.”
The “master plan,” as the company described it, did not go over well. Teespring yanked the shirts, only to catch more criticism the same week for garments featuring slogans such as “Hitler did nothing wrong ever” and “We’re all Hitler now.” Teespring removed those, too.
Then there were the “Keep Calm and Rape A Lot” T-shirts that appeared on Amazon back in 2013. (Amazon’s founder, Jeffrey P. Bezos, also owns The Washington Post.) Wholesale clothing company Solid Gold Bomb blamed a “scripted computer process” designed to riff on the classic “Keep Calm and Carry On” World War II posters. The rape advocacy shirt, along with its companions “Keep Calm and Hit Her” and “Keep Calm and Knife Her,” slipped through the algorithmic cracks. And Amazon missed them in turn.
Teespring’s statement after the latest mistake points to the trouble that caused all these PR crises. Retailers that rely on third-party sellers, or sites that rely on third-party distributors of any kind, need some method of monitoring the content that makes it onto their platforms. Automated scanning systems, apparently, aren’t up to the task on their own — at least not as they’re built today.
That’s a narrow threat when only apparel is at stake. The rainbow swastika T-shirt was at worst cruelly disrespectful to Holocaust victims. The “Keep Calm and Rape A Lot” slogan may normalize sexualize violence, but it’s unlikely it actually inspires men to go on an assault spree. And although the rope-tree-journalist shtick is similarly inflammatory (it’s not just anti-reporter; it’s also racist), the concrete danger to newspaper staff seems slim.
But the algorithm question goes beyond Teespring or Walmart and Amazon. Social media platforms also run on content that doesn’t come from their employees. These platforms have all missed things, too, and do every day: Russian propaganda, beheading and burning videos by terrorist organizations, old-fashioned harassment, and hate speech. Facebook’s and Google’s representatives even blamed algorithms for some of the nefarious ad targeting on their sites during the 2016 election.
With so much to sort through, companies can’t possibly pay the number of people they need to distinguish the acceptable from the awful. That’s why, even as today’s algorithms fail, tomorrow’s are heralded as our future saviors — and someday, they probably will be. But at this point, it’s not responsible to throw human input entirely to the wayside and rely on automation alone. The recurring problem of the outrageous T-shirt is a small-scale example of a larger truth: For now at least, humans are going to have to keep coming to their own rescue.