For many, it was the discovery that the artificial intelligence of social networks “sees” every post and does it quite accurately.
On July 3, Facebook, Instagram and WhatsApp had a major failure – users couldn’t go into applications and access news feeds for several hours. Also, all three services (WhatsApp and Instagram belong to Facebook) did not open images and avatars.
Due to the failure, users could observe an unusual picture on Facebook and the web version of Instagram — instead of photos, rather accurate text descriptions appeared. For example, a snapshot of a girl with a dog was designated by a set of tags: “In this image there can be a girl and a dog.”
What are these tags instead of photos?
The Verge publication explains : in short, the set of tags is how each photo of each user sees the artificial intelligence embedded in Instagram and Facebook. Machine vision systems analyze each image, breaking it into objects. In some cases, Facebook algorithms can even determine exactly which user is shown in the photo.
Facebook has been using the system of “reading” images in its services since at least 2016. In 2018, Instagram launched the function of retelling the content of photos for the visually impaired – in fact, read a set of tags.
For those who are immersed in the field of technology, such text descriptions in social networks are not something new. Programmers even made extensions for Chrome , revealing sets of tags. But the major failure of Facebook and Instagram showed the usually “invisible” part of the work of neural networks to a wide range of people.
Against the background of Facebook scandals and privacy, the media and users wondered why they could collect this type of data on social networks. The Verge suggested that information can be passed on to advertisers. Journalists turned to Facebook for comments, but received no response.
At the same time, the editors said: “Regardless of how the data is used, this is a fascinating look behind the scenes of the world’s largest data collection operation.”
Some examples for comparison
TechCrunch Editor joked: “One man, beard – a fairly unmistakable description of me”
Avatar editor TechCrunch, which described the algorithm
“Facebook’s image recognition missed one person. It’s either me, for Judas “