Facebook uses your Instagram photos to train its algorithms to recognize the content of images
- Facebook uses photos posted by Instagram users to train its visual recognition algorithms.
- The company wants to improve this technology, in particular to help it combat offensive and hateful content.
- But this can cause problems with respecting the privacy of users.
Improving the capabilities of its visual recognition algorithms is an important issue for Facebook. And to train them, the company has a huge image base at its disposal: Instagram, which has been owned by it since 2012.
Instagram Facebook teams explain in a blog post published on Wednesday, June 2, that they use hashtags associated with photos published publicly on Instagram to teach their algorithm to recognize what the photo represents.
By training its artificial intelligence on a database of one billion images and 1500 hashtags, Facebook achieved a success rate of 85.4% according to the ImageNet measurement tool, a record, explains the post. The company believes that by expanding this base and the number of hashtags, the results could be even more satisfactory.
Facebook sees a lot of possibilities opening up thanks to this technology, such as “using AI to generate audio captions of photos for people with visual disabilities”.
This technology can also help Facebook combat offensive content that the company is trying to moderate after the various scandals related to Russia’s interference in the 2016 US election.
According to Facebook’s chief technology officer Mike Schroepfer, as cited by The Verge, the majority of the moderation work on the platform is now done by AI, which searches for violent, hateful, nudity, or terrorist propaganda, in order to help remove them.
“Until recently […]”we had to wait until something bad was reported by someone to do something about it,” he says.
But as The Verge points out, this offers an undeniable competitive advantage to Facebook and also raises questions of respect for user privacy. The latter are not necessarily aware that their publications are used to improve the company’s AI systems.
Facebook is not the only company to engage in these types of practices. For example, Google uses captchas – these tests that are used to prove that a user is not a robot – to train its visual recognition tools, recalls BFM Business.
Receive our latest news
Every day, the main Business Insider news