Instagram picture on phone screen with automatically generated description: Might be an image of 1 person, standing, and Machu Picchu

Source: Facebook

With the launch of a new AI tool to create audio photo descriptions, Facebook has made the site a much more enjoyable place to be for blind and visually impaired users.

While much of Facebook’s content is text — which is simple enough for screen readers to read out with a synthetic voice — images and video have long provided a roadblock for users with visual impairment.

Yes, if visual content is tagged properly with alt text, accessibility programs can interpret the post… but images uploaded without that information (an unfortunately high number) are simply unrecognizable.

To help combat this accessibility issue, Facebook launched a technology in 2016 called automatic alternative text (AAT), which uses object recognition to auto-generate descriptions of photos on demand. (It worked so well that the tool even won a Helen Keller Achievement Award.) Now, a new generation of AAT represents a variety of technological advances to improve on the AI’s photo-recognition skills.

In this update, Facebook has dramatically expanded the number of concepts the AI can identify. Descriptions are also now much more detailed, and may include key details like activities, landmarks, and species of animal.

The new and improved AAT also can now include information about the positions of elements within the photo, as well as each object or person’s relative size. For example, instead of simply saying, “Maybe two people,” the AAT could identify that one person is the central focus of the photo and another is lurking in the back left corner of the photo.

Now everyone can enjoy my cat birthday party photos. Thanks, Facebook!

Learn more about Facebook’s new AI photo descriptions for visually impaired people here.