"The issue of recommendations arises in this case because the complaint alleges the defendants were recommending ISIS terrorist recruiting videos, which under certain circumstances could give rise to liability under the Anti-Terrorist Act,"

It seems to me there is an issue with the notion of “recommendation” here. There needs to be a difference between an intent to promote specific content, vs. a content-agnostic algorithm that just lists what “other users who liked what you liked also liked”. If (say) some Reddit moderators promote terrorist content, they should certainly be liable (or the editorial rules that require them to do so, if any). But when Reddit or YouTube shows you stuff based purely on statistical correlations of what you and other users viewed/liked in the past, then that mechanism in itself shouldn’t be an issue.