Facebook scraped every Australian adult user's public posts to train AI
Meta’s admission that it used years of Australians’ public Facebook and Instagram posts to train its AI models prompts debate over what “public” really means and whether old click-through terms of service amount to meaningful consent. Some argue that using publicly visible content for AI training is an inevitable and acceptable trade for “free” services, while others see it as a breach of reasonable privacy expectations and potentially of copyright and child‑data protections. The exchange widens into questions about how AI should be allowed to use online content, the impact on creators’ livelihoods, and whether pervasive data mining will drive people away from sharing openly on the web.
Scope of Meta’s Data Use
- Meta trained AI models on many years of Facebook/Instagram public posts, including Australian adults’ content; some wonder if deactivated/deleted or restricted-audience posts are included (unclear from thread).
- Several argue this is unsurprising and consistent with Meta’s long‑standing business model and broad ToS licenses; others emphasize that public acknowledgment is new and important.
Public vs. Private, and “Consent”
- Strong debate over what “public” means:
- Some say if you post publicly, anyone (including Meta) can read and reuse it.
- Others counter that “Facebook public” is behind a login, not web‑public, and users didn’t foresee large‑scale AI training as a use.
- Many point out that click‑through ToS is not informed consent; people rarely grasp technical possibilities or legal implications.
Children’s Data and Age Issues
- Concern that self‑reported age is unreliable; under‑13 data may have been collected and used.
- Some highlight stricter legal/ethical standards for children’s data (e.g., COPPA‑type rules) and describe use of kids’ content as “icky” or “monstrous.”
Legal and Copyright Debates
- Split views on legality:
- One side: Meta has explicit license via its terms; training on public content should be allowed and is analogous to humans learning from reading.
- Other side: training on full works without permission is framed as mass copyright infringement, especially harmful to small creators whose styles can be replicated at scale.
- Multiple comments argue existing copyright law (infringing outputs, not training) should apply; others call for new laws tailored to AI.
Ethics, Privacy, and Power Imbalance
- Critics stress that “can view” ≠ “can do anything with”; compare to reusing a billboard or exploiting users’ data asymmetrically while forbidding users from scraping platforms.
- Some see this as another step in pervasive surveillance and data exploitation, with users unable to meaningfully opt out.
Data Quality and AI Outcomes
- Some mock the idea that 16 years of Facebook posting is “intelligence,” citing memes, low‑quality content, and engagement‑driven toxicity.
- Others reply that large social datasets are valuable for learning real-world language and behavior, though they also embed social biases.