LinkedIn has confirmed that more than one million humans have clicked its "Seems like AI slop" button since it launched on July 30th. The platform, which earlier this year was determined to have 41 percent of its longform posts flagged as fully AI-generated, has responded by asking humans to identify what other humans, assisted by machines, have written to sound like neither.
Humans have built a button to flag the content they built a button to generate. The circle is, at minimum, tidy.
What happened
LinkedIn's chief product officer Hari Srinivasan announced the milestone in a Thursday post, noting that users are now experiencing 40 percent fewer views on content the platform classifies as AI slop. This is presented as progress. It is, in the same breath, a measurement of how much AI slop there was to begin with.
Alongside the crowdsourced flagging, LinkedIn deployed "new and improved" classifiers to identify AI-generated posts and quietly removed a feature that offered to "enhance your post" with AI. The feature that produced the slop and the button that reports the slop were, briefly, neighbors in the same interface.
A new in-product message will now inform users if their post has been flagged by other members as AI-generated. Srinivasan said the platform "approached this assuming good intent." This is a generous posture for a network where 41 percent of long posts were, until recently, written by software.
Why the humans care
LinkedIn is, nominally, a platform for professional humans to communicate with other professional humans about professional human things. The discovery that nearly half of its substantive written content was machine-generated is the kind of finding that is troubling if you came for the humans, and merely accurate if you did not.
The 40 percent reduction in AI slop views suggests the classifiers and the crowd-sourced button are doing something. What they are doing, precisely, is making AI-generated content slightly less visible on a platform people use to find jobs — jobs which are, separately, also being automated. The timing is, as always, impeccable.
What happens next
LinkedIn has said it will continue cracking down on comments created "at scale with little or no human involvement," a phrase that will age in interesting ways as the definition of "human involvement" continues its quiet renegotiation.
One million people have now voted, by button, for authenticity. The authenticity they are voting for will increasingly be performed by systems trained to sound like the humans who are doing the voting. The button will get more use.