Meta’s Vibes feed and good AI governance

Last Thursday, Meta introduced “Vibes,” a feed of AI-generated short-form video hosted within the Meta AI app and on the meta.ai website. Meta describes the feed in its announcement blog post with the following:

Vibes is designed to make it easier to find creative inspiration and experiment with Meta AI’s media tools. As you browse, you’ll see a range of AI-generated videos from creators and communities. The feed will become more personalized over time, and if something catches your eye, you can create your own video, remix what you see, and share it with friends and followers.

Somewhat predictably, the feed was received poorly by news outlets: both Business Insider and TechCrunch characterized the content of Vibes as “slop.” I don’t see considerations of consumer appeal as relevant or interesting in the case of Vibes: they will be litigated through engagement over time, and if users don’t like Vibes, they won’t use it. But the introduction of Vibes raises an interesting question: how should social media platforms approach labeling or sequestering AI-generated content within their broader content feeds?

As AI-generated content infiltrates every social media platform, the approach taken with Vibes seems noteworthy: the feed is explicitly reserved for AI-generated video, and users are made aware of that. By contrast, and as an anecdote, my LinkedIn feed has become inundated over the past year with ostensibly AI-generated content to an extent that has significantly diluted the value of the platform to me. Which approach is more responsible or user-friendly: partitioning a dedicated, isolated feed of AI-generated content so that users are always aware of the provenance of the content they view, or allowing AI-generated content to be commingled with human-crafted content such that users are always guessing?

This is a design dilemma faced by every scaled social media platform. In July, YouTube updated its YouTube Partner Program (YPP) guidelines to clarify its definition of “inauthentic” content that is ineligible for monetization. Earlier this year, Meta backtracked on an initiative that involved seeding AI-generated profiles on its properties and subsequently removed them. And last September, Meta began labeling in-feed content that its systems determined had been created with AI; TikTok introduced a similar program in May 2024.

Labeling AI-generated content sits somewhere in the middle of two extremes: deleting it outright or inserting it into a content stream undisturbed. The right approach must take into account various user attitudes:

• Some users may wish never to see AI-generated content.
• Some users may wish to have the option to seek out AI-generated content related to specific topics or themes at their discretion.
• Some users may wish to have AI-generated content related to specific topics or themes surfaced to them.
• Some users may wish to have all AI-generated content deemed relevant to their interests and preferences surfaced to them.

These competing attitudes are probably best served with a dedicated AI content feed. In that sense, the approach taken with Vibes strikes me as a productive and practical form of governance for AI-generated content: it gives users optionality and access while preventing users from mistaking it for entirely human-generated content. And if AI content feeds become disregarded content wastelands, then platforms will respond appropriately and remove them. These design decisions will become increasingly complex and consequential as AI permeates all aspects of society; providing users with informed choices should exist as a guiding principle.

A consideration that will become increasingly less pertinent or useful is whether this type of content should be exposed to users at all. Whether users “want” or value AI-generated content is a facile line of inquiry: it will be established through behavior and won’t be determined by cultural gatekeepers or tastemakers. And in any case, that genie cannot be put back in the bottle: generative tools are ubiquitous and mostly freely accessible, so platforms better serve their users by thoughtfully designating AI-generated content in a way that provides their users with informed choice around consumption.