If you've ever wondered how Facebook sometimes asks you to tag people and suggests their names with great accuracy, it's their artificial intelligence system at work here. Now Facebook's AI will go beyond identifying people and suggest captions for images, Wired reports.
Facebook calls the system Automatic Alternative Text and it uses deep neural networks to identify particular objects and facial expressions, read the picture to determine the location, and using a text-to-speech engine, it describes the image to users who cannot see the picture.
In a demo, Facebook engineer Matt King fed a picture of a friend who biked across Europe into the system. Facebook's AI reportedly described the picture as being set outdoors, adding that it included trees, clouds and grass, and that it was taken near water. Normally, the text-to-speech engine would have simply stated that King's friend has put up a picture.
"My dream is that it would also tell me that it includes Christoph with his bike," King told Wired. "But from my perspective as a blind user, going from essentially zero percent satisfaction from a photo to somewhere in the neighbourhood of half ... is a huge jump."
According to King, the system is still not perfect as it can't string words to form sentences, but he's confident that it will happen in the near future.
Twitter had recently introduced a similar feature, although it still requires human input. The feature allows users to key in alternative text to images, which a text-to-speech programme can read out to visually impaired users.