Facebook’s image outage showed us that AI is tagging our photos

At around 10:30 pm yesterday, we noticed that users were having issues loading and downloading photos and videos on Facebook, Instagram, and Whatsapp. After the initial (and expected) barrage of complaints toward local telco companies from Malaysians, a quick search online through downdetector revealed that the issue was a worldwide outage. Facebook responded relatively quickly this morning, and things are back to normal.

But users noticed something peculiar during the outage—instead of their pictures, text tags used by Facebook’s machine learning systems were displayed instead. So instead of seeing pictures on Facebook, users were shown tags such as “Image may contain: dog”, or “Image may contain: 1 person, 1 beard”.

It seems innocuous, but this is actually how Facebook’s artificial intelligence (AI) processes and recognises your images. So don’t be offended if Facebook tags you as more than one person… it’s a work in progress. The tags also appeared to Instagram users as well.

This isn’t actually a new development. Facebook has been utilising AI to help with image recognition since 2016, and the whole effort is connected to the company’s movement for more accessibility to everyone. Practical uses include describing images and text to the visually-impaired.

The question, after Facebook’s data privacy issues, is whether the data here is used for ad-targeting. Deep machine learning has certainly come a long way in recent times, but it’s unclear if the data is being used by Facebook for more untoward ad-targeting. After all, we’ve all been subject to targeted ads on Facebook at some time or other.

Basically, these developments in AI and machine deep learning have come a long way over recent times—the outage last night showed us how far, of course. The worry here is that once Facebook’s machine learning algorithm is capable of reading your visual content as text, that this data can be utilised to infringe users’ privacy.

Facebook has yet to comment on the matter.

[ VIA ]