Facebook working on tech that can 'read' photographs by users
Published 04/11/2015 | 02:30
Facebook is working on artificial intelligence that will be able to evaluate the world in a similar way to humans, such as being able to tell what is happening in users photos.
Speaking at the Web Summit in Dublin, Facebook chief technology officer Mike Schroepfer outlined the main technology developments the US giant is aiming to make over the next decade.
Earlier this year, the social media giant unveiled its "Memory Networks" technology, which lets machines perform high-level questioning and answering.
As an example, Mr Schroepfer showed a situation where a computer was fed a synopsis of 'The Lord of the Rings' and was then able to answer questions about the plot.
The company is now planning to pair this application with technology that can identify individual objects from photos and label them, such as a bat being held by a baseball player.
Facebook is using the two to develop a mobile-based system called Visual Q&A.
It will allow people who are visually impaired to ask questions about a photo, such as "Is there a man in the picture?" and receive answers from the computer about what is happening in the frame.
"So much of the world, and so much of the sharing on social networks is photos and videos ... one of the keys to making systems that work is teaching computers to understand the visual world," Mr Schroepfer said.
The artificial intelligence system will give the most relevant answers and will also show other guesses and suggested objects.
The system is not yet available to the public. The announcement marks Facebook's increased focus on artificial intelligence.
The company is trialling a virtual personal assistant called 'M', much like Apple's Siri, through its messenger app.
Facebook has also completed a prototype of its Aquila drone, which can beam a wifi signal to areas of the globe that are not served by a cell tower, enabling more people around the world to get online.
The drone will be solar-powered and able to stay in the air for up to three months without needing maintenance. Each drone will be able to cover a radius of about 80km.
Mr Schroepfer said that the firm "has a full-scale version fully constructed which will be undergoing flight tests very soon."
Across the room, Mike Krieger, co-founder of photo and video sharing app Instagram, stressed that balancing machine learning and human curation had been the key to Instagram's success.
Since setting up in 2010, Instagram has grown to become one of the world's most popular and fastest growing social networks with more than 400 million users.
Mr Krieger, appearing on the Web Summit's main stage, outlined the problems the company has faced as it has attempted to continuously grow and attract new users but also keep existing users happy.
He said one early development that proved successful was a community manager who engaged with users.
The community manager was then able to suggest different features to users.
However, he said as the social network grew such an approach became almost impossible and more algorithms and "machine learning" was introduced to try and identify what users wanted.
He said the 'Explore' feature allows Instagram to monitor a range of different areas including how long people look at images, who they're following and what they're interested in.
Another addition that enabled users to navigate their way around the site was to arrange photos and videos in particular categories.
He said this was done through a combination of "machine learning and human curation".
"We realised that with all this cool technology really needed that human touch to actually make the product sing. So we had a group of editors that take data and break it into categories," he said.
He said another recent feature which had proved successful was the new storytelling tool.