There was a time, not so long ago, when a photograph of a rocket launch could travel across the internet without being mistaken for something unsuitable for children. That time, it seems, has passed.

Michael O'Herlihy, a Twitter veteran who recently returned to the platform now known as X to serve as head of safety, acknowledged as much at the Brand & AI Safety Summit in New York on Tuesday. The company's artificial intelligence moderation system, he noted, had once flagged an image of a SpaceX rocket ascending into the Florida sky as "adult content." Neither the author of the post nor anyone else could determine why the algorithm had reached this conclusion. (The rocket, for its part, offered no comment.)

Mr. O'Herlihy previewed what he suggested would be a remedy of sorts: labels that would appear beneath flagged content on X, explaining what the company's AI system believed it had detected, describing how the post's distribution had been curtailed, and offering users a path to appeal the decision. It is, in effect, a system designed to make the machine show its work.

The move comes as platforms continue to grapple with the fundamental opacity of automated moderation — systems that operate at enormous scale but whose individual judgments often appear, to the humans subject to them, as mysterious as the workings of the Delphic oracle, if the oracle had been trained on several billion posts and occasionally confused aerospace engineering with adult entertainment.

Whether the new labels will satisfy users who have long complained about unexplained restrictions remains to be seen. But for those who have spent years shouting into the algorithmic void, the prospect of receiving an answer — any answer — may itself constitute progress.

Original story published in adweek.com: "X Previews Labels That Tell Users Why Its AI System Flagged Their Posts"