Skip to main content

People have trained a LinkedIn algorithm to be awful

People have trained a LinkedIn algorithm to be awful

Share this story

Picture this: over happy hour suds in a generic Irish pub, a man wants a cheap laugh to endear himself to his colleagues. He unholsters his iPhone, opens LinkedIn, and types a two-word prompt. The app thinks for a moment, then loads an ordinary profile page belonging to a middle-aged man. In his profile photo, the man sports shaved grey hair and a hoodie that matches. To the right hovers his name: Dick Monster.

Dick Monster lives in the Netherlands and works as a packaging scheduler for Mars — the food company, not the planet. He has proficiencies in Microsoft Excel, project management, and a number of other office skills. Dick Monster also has a name that, to certain American ears, sounds like a laugh line in an Adam Sandler movie. Monster did not respond to a request for comment.

But Dick Monster is not alone on the page. On the far right frame of his profile, Monster is joined by a whole group of people who probably had it rough in elementary school. It’s the "People Also Viewed" box, which ordinarily surfaces colleagues and people with similar skills. Not so for Dick Monster. People who view his LinkedIn profile also tend to seek out Garth Brownsword, Truman Hardon, and Guido Prick.

In an email, Suzi Owens, LinkedIn’s senior manager of marketing and communications, explains that the "module is driven by click/browsing behavior by our 433 million members, and it is often used when people want to search for members with similar expertise."

Of course, that’s not what’s happening in this case. A set of disconnected LinkedIn surfers, connected only by our collective sense of humor, have unwittingly trained the service’s algorithm to package a group based on their names. The algorithm reflects how people perceive the group, rather than who they actually are.

The algorithm reflects how users perceive each other, not who they are

Each profile that branches from Monster’s module spawns, inside their own People Also Viewed bar, another set of individuals with names that could be used by a prank caller. For example, visitors to the profile of Fanny Trouble, a PR Manager at Nike, also viewed the profiles of Hugh Jass and Charles Wankstain.

"In regard to the member profile you mentioned," Owens writes about Dick Monster, "this is a real profile. There is some cultural sensitivity that we need to bear in mind for names in different origins."

Owens is right. But the algorithm exhibits no such sensitivity. The trail that begins at Dick Monster leads to a host of Asian names that are often included together: Yu Arafuka, Hammer Dong, and Poo Poo Pong.

The algorithm exhibits no such sensitivity

Owens wouldn’t say if the company plans to make adjustments to the module. But LinkedIn takes the matter of fake profiles seriously, she says. "We have a number of measures in place to confirm authenticity of profiles and remove those that are fake. We also encourage members to utilize our Help Center to report inaccurate profiles and specific profile content to LinkedIn."

The company does appear to be taking some action. Since we first contacted LinkedIn, numerous profiles have been removed for the fictional Howie Feltersnatch. One of the Feltersnatch profiles was featured prominently on many People Also Viewed modules. (An accompanying picture of the presumably fake Feltersnatch featured a shirtless man, wearing a cowboy hat and riding a tractor through the air.)

But real people on the service continue to be lumped together by their names, not their skills. It’s one more way people misuse a site intended to be a professional network. In The Atlantic in 2015, Sophie Gilbert wrote about the men who wrongly use the platform as a dating site. The People Also Viewed module also sometimes groups very attractive but otherwise disconnected women together, suggesting some people are using the site as a kind of safe-for-work porn site.

The module can be removed

LinkedIn does allow members to remove the module from their profile, though the LinkedIn help page warns that "If one of your goals on LinkedIn is to increase your visibility, this feature can significantly increase the likelihood that your profile will be discovered and viewed by other members."

But even with the module removed from the page, the algorithm behind it will still be learning new ways to reflect the people who shape it — similar to its more verbal cousin, Microsoft’s teen chatbot turned malevolent fuckboy, Tay.

Last week, LinkedIn CEO Jeff Weiner said his platform will be valuable for workers as they lose their jobs to robots. Picture this: after work, three robots meet at the charging station to snicker at the profile of ASS-R 371.