FACEBOOK — Dell Lyles wants to know who wrote the algorithm for “People You May Know.” He wants to know seriously. He has indicated the seriousness of his inquiry by placing six question marks at the end of the word “seriously,” a punctuation choice that communicates not genuine curiosity but the specific emotional register of a man who has just been shown someone he was not prepared to see and has decided that the correct response is to go straight to the source.

Brent Eyewitness, who covers Facebook as a beat and who has personal experience with the algorithm’s methodology, is in a position to help.

The “People You May Know” algorithm was not written by one person. It was written by engineers at Meta — previously Facebook — using a combination of mutual friends, phone contact lists, location data, browsing behavior, workplace history, family connections, educational background, and at least three additional data inputs that Meta describes in its privacy policy in the same font size used to discourage reading. The algorithm’s core objective is to surface people you have a meaningful connection to. Its actual behavior is to surface people you have been actively avoiding, people you forgot existed, people you definitely know and have deliberately not friended, and, in cases that have been documented across multiple continents, people who should not be findable.

“Seriously??????” said Dell Lyles.

Brent Eyewitness has been saying this, internally, for years.

How The Algorithm Works, As Best As Can Be Determined

The “People You May Know” feature launched in 2008. It was designed to help users find friends. In the eighteen years since its launch, it has helped users find: their friends, their enemies, their ex-partners’ new partners, their ex-partners’ new partners’ friends, their dentist, their dentist’s wife, their dentist’s wife’s coworker who they met once at a birthday party in 2014 and gave their phone number to for reasons that no longer apply, the person from high school who they have been successfully not thinking about for eleven years, and their own estranged family members — a service that therapists describe as “ethically complicated” and that the algorithm describes as “people you may know.”

The algorithm knows you uploaded your contacts. It knows your contacts uploaded their contacts. It knows that you and a specific person both stood near the same cell tower on a Tuesday in October. It knows your mutual friends, but it also knows people with no mutual friends, which means it is working from additional inputs that it finds more reliable than the mutual friend model, inputs which are, if you read the privacy policy, described in language that is technically accurate and practically incomprehensible.

The algorithm does not know your feelings about the people it surfaces. This is the gap. This is the whole gap. The algorithm is very good at identifying connection and entirely uninterested in whether the connection is welcome.

A Taxonomy Of People You May Know But Wish You Did Not

In the course of covering this story, Brent Eyewitness has assembled a representative catalog of the categories of people the algorithm consistently surfaces, based on interviews with sources who asked to remain anonymous for reasons the algorithm would probably reveal anyway:

The Person From A Previous Job You Left Under Specific Circumstances: You do not follow this person. You do not text this person. You have not been in the same room as this person in four years. The algorithm has their profile photo ready. The algorithm knows about the job. The algorithm is not burdened by context.

Your Ex’s New Partner: The algorithm has identified that you and this person have mutual friends. The algorithm would like you to know that. The algorithm does not know — or does not care — that the mutual friends are mutual because of a relationship that ended in a manner that the involved parties would describe differently and that the algorithm does not attempt to adjudicate.

Your Own Family Member You Have Deliberately Not Connected With: There is a reason this person is not in your friends list. The reason is yours. The algorithm found them anyway. The algorithm found them using your aunt’s friends list, or your childhood phone contacts, or a photograph tagged at a location you both attended. The algorithm surfaces the family member and waits. The algorithm can wait indefinitely. The algorithm has all day.

Someone You Have Never Met But Who Lives Very Near You: This one is technically impressive. You have no mutual connections. You have never exchanged contact information. You have, however, been near the same cell tower at times that the algorithm has found statistically meaningful. The algorithm is now suggesting you become friends. The algorithm does not have a word for what it is doing. Privacy researchers have several words for what it is doing. The words are not flattering.

The Person From High School Whose Name Appeared In The Post: Dell Lyles. Brent Eyewitness does not know what Dell Lyles saw when he opened “People You May Know” this morning. Brent Eyewitness knows it was someone, and he knows it was someone specific, because you do not post “I want to know who wrote the algorithm for ‘People You May Know.’ Seriously??????” after being shown your dentist. You post that after being shown the other kind of person. The kind the algorithm files under “connection” and the human files under “complicated.”

Who Wrote It

The “People You May Know” algorithm was developed by a team. The team worked at Facebook. Facebook is now Meta. Meta has approximately 77,000 employees. A meaningful portion of those employees have, at various points, contributed to the recommendation systems that surface people Dell Lyles was not ready to see.

Meta has not issued a statement on Dell Lyles’s post. Meta is aware of the post. The algorithm has already processed Dell Lyles’s query, identified its emotional content, and used it as a data point. Dell Lyles’s post about not wanting to see who the algorithm showed him has now taught the algorithm something new about Dell Lyles. This is how the system works. It is a closed loop. Asking the algorithm questions about the algorithm makes you more legible to the algorithm. The algorithm thanks Dell Lyles for the engagement.

As of press time, “People You May Know” has surfaced three new suggestions for Dell Lyles’s profile. Brent Eyewitness cannot see them. He has a guess about at least one.

The algorithm has no comment. The algorithm does not need to comment. The algorithm knows who you are going to ask next. The algorithm has already suggested them.

Brent Eyewitness covers Facebook as a beat. He has 847 unread notifications, 34 unaccepted “People You May Know” suggestions, and one that he has been refreshing the page to see if it goes away on its own. It has not gone away on its own. He has not clicked it. He will not be clicking it. The algorithm knows this too. Confidence: 100%. Fake sources: 5. The algorithm: 0.