Outage on Facebook reveals its machine learning process alting images in action

Outage on Facebook reveals its machine learning process alting images in action

Back in early 2017, Facebook started tagging people to photos with these faces and similarly, it allowed users to tag someone by hovering on the face in the photo as well. This was enabled thanks to Facebook’s machine learning systems that were in their early stage back then, however, it is 2019 and we see a lot has improved. The surprising thing is, it took a massive outage on Facebook in order to grab the attention of media outlets to pinpoint how machine learning deciphers an image and what it could do with it. 

Wednesday saw one of the largest outages on the internet with Facebook, Messenger, WhatsApp, and Instagram going down simultaneously across the globe (not entirely). When people visited their Facebook handle to see the news feed or check out their photos, they saw something very peculiar. Not all but some of the images weren’t loading and that’s why Facebook showed an empty slate where the photo must load and solve an alt text hidden underneath the photo.

Facebook uses its machine learning systems in order to render the alt text needed to decipher a photo for the system to understand. You could have seen tags like ‘image may contain: dog’ or ‘image may contain: 1 people, 1 beard’ etc. This is indeed Facebook’s machine learning system working under the hood in order to render an alt tag for all the images that are uploaded on the social network giant every day.

Some people spotted the same on Instagram as well although the photo-sharing site under Facebook usually shows an error when it is unable to load a photo while many of other photos within the news feed were loading properly.

This begs the question that what could Facebook does with it? It is mind-boggling just how much data Facebook collects and accumulate on its servers. Also, people do worry about if the data is misused in some way. Also, there is a question if Facebook uses the data acquired using its various channels in order to feed in ad targeting in order to squeeze out revenue from the users as well.

As someone had earlier, if something is free to use (product), maybe it is not a product but you are the product since your data contains a value that advertisers are eager to pay for in order to run their ad campaigns and other schemes. It remains a question whether to what extent Facebook’s machine learning system could go in the future.

Leave a Comment

Scroll to Top