Arguendo, suppose

1) that the machine learning logic has determined a user to be a pedophile and
2) that the customization logic creates a pedophile-directed user feed along with thumbnails, tags, tag lines, and search key words that are all crafted to the fit the pedophile’s taste in illegal child sex acts.

It makes sense for a social medium ICS computer service to stimulate the pedophile’s proclivities. Such customization would increase eyes-on-the-page and make the advertising space more valuable that is associated with the social medium ICS.

What does the social medium ICS care if such stimulation makes it more likely for the pedophile to commit a crime?

Congress seems to have intended to give the social medium ICS a downside for an unprincipled marketing & sales strategy that encourages sexual crime.

While it’s true that thumbnail creation, feed customization, tags, tag lines, and search key word creation are all functions that are associated with editorial discretion (even if in this hypothetical case these functions are performed automatically), the machine learning algorithms could just as easily flag a probable pedophile as they could stimulate him.