Meta’s AI is summarizing some bizarre Facebook comment sections

If you’ve been on the Facebook app lately, you might’ve seen Meta’s AI inject itself into the comment section with summaries of what people say. Given how wild Facebook comment sections often become, it’s not hard to imagine how ridiculous some of these summaries turn out. (This isn’t the first time Meta’s AI has appeared in the comment section, by the way: 404 Media spotted it pretending to be a parent in a Facebook group.)

After seeing screenshots of the feature shared on Threads and Reddit, I decided to check the comment sections on my Facebook app. I found the AI summaries popping up on many of the posts I checked — unhinged responses and all. One AI summary on a post about a store closure said, “Some commenters attribute the closure to the store ‘going woke’ or having poor selection, while others point to the rise of online shopping.”

Another Facebook post from Vice about Mexican street wrestlers prompted a comment section summary that said some people were “less impressed” with the performance and referred to it as a “moronic way of panhandling.” The AI also picked up some of the more lighthearted jokes people made about a bobcat sighting in a Florida town. “Some admired the sighting, with one commenter hoping the bobcat remembered sunscreen.”

It’s still not clear how Meta chooses which posts to display comment summaries on, and the company didn’t immediately respond to The Verge’s request for comment.

Either way, the summaries really don’t include anything that I found useful (unless you love vague notions about what random people have to say) — but it could help you identify posts where the comment section has gotten too toxic to bother scrolling.

The AI summaries have also prompted privacy concerns, as Meta is feeding user comments into its AI system to generate them. Over the past week or so, many Facebook and Instagram users in the European Union and the UK received a notification informing them that Meta will train its AI on their content. (Data protection laws in both regions require Meta to disclose this information.) Although Meta will let these users object to having their data used to train AI, the process isn’t that simple, and the company has rejected some users’ requests.

Here in the US, Meta’s privacy policy page says the company uses “information shared on Meta’s Products and services” to train AI, including posts, photos, and captions. Meta lets you submit a request to correct or delete personal information used to train its AI models, but it only applies to information from a third party. Everything else seems to be fair game.