3 months ago
- In 2018, Mark Zuckerberg disclosed that Facebook used AI to remove 99% of ISIS and Al Qaeda content before human review, raising early ethical concerns.
- A 2025 UN joint declaration highlighted risks of AI content moderation, including over-removal, discrimination, and censorship due to biased datasets and opaque processes.
- Reports show AI content moderation disproportionately impacts vulnerable groups and low-resource languages, with failures in nuance leading to suppression of legitimate content like LGBTQ expression.
- Automated moderation can benefit human moderators by sparing them from traumatic content, though AI training involves similar risks, and potential for improved accuracy exists but is unproven.
- Transparency and accountability safeguards remain lacking, making it urgent to define conditions for AI use in moderation as it becomes central to platform governance.