Behind the scenes of Instagram (and Facebook): “How AI influences what you see”

Through the words of Nick Clegg, his head of Global Affairs (publish here)Meta has announced a series of updates that should increase transparency on how its AI systems classify and suggest the contents on Facebook and Instagram.

Clegg related these updates to Meta’s approach to solutions Generative AI and to his philosophy of transparency: “This decision (that of explaining how things work, ed) is in line with a broader philosophy of openness, transparency and reliability – wrote Clegg – With the rapid progress of technologies such as AI , it is understandable that people are excited about new possibilities and at the same time concerned about the potential risks of this new technology.

Again: “We believe that the best way to address these concerns is an approach of openness to the development of artificial intelligence. Overall, we believe that as these technologies develop, companies will need to be more transparent about how their systems work and will need to collaborate openly with everyone in this market, governments, and civil society to guarantee that this development takes place in a responsible manner”.

Through “sharing more details about how artificial intelligence systems classify content for News Feed, Reels, Stories and more – Clegg explained – we are making it easier for users to control what is viewed on Facebook and Instagram, making available new tools to support studies of public interest”.

A mix of predictions and information

Intentionally, the goal of the AI ​​would be to make “visualize posts that, ideally, are relevant and interesting” to people. In Clegg’s post, which openly talks about predictive AI, we read precisely that “our artificial intelligence systems predict the interest that a content could generate on a person, so as to prioritize this content over others”. For example, have shared a post (someone’s, evidently) previously is often an indicator that the user found that post interesting and the prediction that the user will share another post (by the same author) is “a factor that our systems take into account”.

As you can imagine, “no prediction, if taken individually, is a perfect indicator of the value that a post can have for some users” and therefore, “to suggest the most relevant content for people, we use a forecast combinationsome based on user behavior, others on feedback received from them through surveys”. Clegg explained in detail how everything works behind the scenes, understanding however well that “not everyone will be able to identify this information just because we have shared it on our site” and therefore announcing that “in the coming weeks we will integrate the function Why am I seeing this post? in the Reel and Search sections of Instagram and in the Reel section of Facebook, after having previously made it available for some contents of the Feed and for announcements on Facebook and Instagram”. In practice, by clicking on a Reel it will be possible to view further information on how the previous activity somehow affected the machine learning models which determine and suggest the Reels that are shown.

How to turn off the algorithms on Facebook

Clegg also explained that “on Instagram we are testing the new feature I carewhich allows you to highlight a suggested Reel and consequently show a greater number of contents” of that kind.

Likewise, “to personalize the experience and the content you see”, on Facebook there is the function Show more, show less for all posts in the Feed, Videos and Reels via the 3-dot menu: the person clicks and can decide to see more (or less) content from that user or from that page, while Meta has let it be known that “we are studying a way to make this function even more evident”.

Of course, and as we had already explained on Italian Tech“if you don’t want an algorithm-structured feed or if you just want to see what your feed would look like without algorithm intervention, you can use the tab Feeds on Facebook or the tab Followed on Instagram” for view posts in chronological order. How it used to be, before AI came along.

@capoema