Visually impaired individuals will now be able to “see” photos on Facebook, with the introduction of an accessibility feature that describes the contents of uploaded pictures using a voice extension.
The new feature, which currently runs on iOS, is only accessible in five countries at the moment, but will soon be functional in other regions and on additional platforms. And while it will only describe pictures for now, the company is working on extending the functionality to videos.
This is just but one of the many upgrades that the tech giant has unveiled recently. Disabled people already have a wide range of accessibility features that have enhanced their interactions on the social network over time. Commenting on the new developments on his Facebook page, Mark Zuckerberg said that the use of Artificial Intelligence (AI) “is an important step towards making sure everyone has equal access to information and is included in the conversation.”
Facebook’s Accessibility Improvements
On most operating systems, Facebook already provides users with an option to activate sound accessibility, through which menu contents are read out to them. The number of blind users on Facebook has increased over time because of this.
As Matt King, a member of Facebook’s Accessibility Team, notes however, more needs to be done in order to bring the user experience of disabled social media users up to par with those of their seeing counterparts. Matt, who is visually impaired, is among the people working to improve the platform’s usability for the disabled.
Facebook’s New Feature Uses AI
Facebook is not the first platform to offer a descriptive feature for photos. However, its own is considered more advanced because it uses artificial intelligence. Unlike Twitter, which requests that people uploading pictures write a description for them so that it can be read out to blind users, Facebook’s AI identifies the image by itself.
Using the AI for picture identification tasks is a breakthrough as it does not depend on users writing descriptions, which some people find boring, going by the relatively low number of clicks on the option reported by Twitter. Another issue with descriptions inputted by users is that similar images end up having different descriptions and sometimes different photos may also bear the same descriptions, leading to confusion.
The Facebook feature is sophisticated, as it uses deep learning, and this helps avoid providing misleading information to blind users.
Evaluations of descriptions inputted by users on other social media platforms have shown that pictures are prone to mislabeling, either as a result of mistakes or because some users give incorrect information with intent to deceive others. In theory, the use of AI should avoid such issues.
This technology minimizes data redundancy as there will only be one description for the same picture and it will not change regardless of the number of times it is uploaded.
Possible abuses and errors
It is not clear which precautions Facebook is taking to protect its AI from abuses and errors, some of which have the potential to seriously damage its reputation.
Companies like Google and Microsoft have had some problems with their AI applications which the Facebook team hopefully took note of. In July 2015, for instance, Google’s search AI, which identifies pictures, erroneously labeled a black couple in a photo as “gorillas”. While these mistakes are unintentional, they must be avoided at all costs as they give companies a very bad name.
To correct the problem, Google removed most of the sensitive terms from its AI and now it only uses common, neutral names for labels. Still, the problem still doesn’t seem to be fully solved, as some users report that the AI labeled pictures of their children as “pets.” Other users have reported that they have seen molehills being tagged as “mountains” by the Google AI.
With of all of its glitches, the opportunities that AI will provide for the visually impaired sure paints an optimistic picture for the future.