Lots to unpack in the Facebook @OversightBoard decisions today: the OB is throwing some elbows to define its own role & scope, context is crucial in all of these decisions, and the OB’s policy recommendations reflect what human rights orgs have been saying for years.

The Instagram/breast cancer case is v. interesting. https://www.oversightboard.com/decision/IG-7THR3SI1 It’s an opportunity for the OB to define its role, since Facebook restored the post at issue after the OB took the case. FB argued the case was thus moot, but OB says nope, there’s still lots to address.

For one, if FB could take cases off the OB’s docket simply by restoring the content, that would give FB too much control over the OB. And the user whose post was wrongfully removed ought to get the full explanation of what happened, which the OB opinion can provide.

The breast cancer post at issue was automatically removed with no human review. Shockingly, FB apparently argued that the role of automation wasn’t relevant in this case:

Fortunately, the OB rejected this and explained that lack of human review is a crucial factor & is inconsistent with int’l HR standards: “Automated content moderation without necessary safeguards is not a proportionate way for Facebook to address violating forms of adult nudity.”

(For more on automation in moderation through a human rights lens, a shameless plug: https://journals.sagepub.com/doi/full/10.1177/2053951720920686)

The OB had a variety of policy recs for FB in this case, including ones that many orgs have called for, e.g. in http://santaclaraprinciples.org . Basic stuff like telling users which rule they violated & whether automation was used in the takedown, and giving them the chance to appeal.

Some recs do raise implementation questions. The rec to “Improve the automated detection of images with text-overlay” is good/obvious, but probably not super-simple to do. Parsing overlay text in an image to understand broader context is a tricky machine-learning task.

But overall the OB makes clear that it has “a broader concern at Facebook’s lack of transparency on its use of automated enforcement, and circumstances where internal appeal might not be available.” Automation will be imperfect and remedy/appeal/human review is essential!