With A.I. advances, Facebook tests M, your newest assistant
Today the company reported that it's making headway in this area to create and test an artificial intelligence assistant dubbed "M."
Unlike other intelligent systems that might tell the user what the weather will be or to pull up a map, M is designed to complete multi-step tasks.
M, according to Facebook, is set up to purchase a gift, for example, and have it delivered to your mother. It also will be able to make travel arrangements and appointments and book restaurant reservations.
The new assistant program is in what Facebook calls a "small test" that is showing promise. However, to handle complicated requests and tasks, the system needs more advances in machine language, vision, prediction and planning.
"The amount of data we need to consider when we serve your News Feed has been growing by about 50% year over year -- and from what I can tell, our waking hours aren't keeping up with that growth rate," wrote Mike Schroepfer, chief technical officer at Facebook, in a blog post. "The best way I can think of to keep pace with this growth is to build intelligent systems that will help us sort through the deluge of content."
He added that with text, photos, video and soon virtual reality, Facebook is dealing with an every growing amount of information that users are posting and searching for.
Schroepfer noted that the Facebook A.I. Research team, also known as FAIR, has focused some of its research on image recognition and natural language understanding and is making gains that could one day benefit users searching for information on the social network.
Next month, Facebook's A.I. team will present its research on object detection, a subset of computer vision, at NIPS, an annual conference on Neural Information Processing Systems, being held in Montreal.
Object detection is a difficult challenge. While people can see stripes and shapes, the computer only recognizes pixels, making it more difficult to distinguish what is in the images.
"Our researchers have been working to train systems to recognize patterns in the pixels so they can be as good as or better than humans at distinguishing objects in a photo from one another -- known in the field as 'segmentation' -- and then identifying each object," Schroepfer wrote.
Facebook reported that its new A.I. system can detect images 30% faster, using 10 times less training data than previous industry benchmarks.
The company also said that it has combined its Memory Networks (MemNets) system, which is designed to read and answer questions about short texts, with new image recognition technology. By connecting the two, users should be able to ask the machine what is in the photos.
"We've scaled this system from being able to read and answer questions on tens of lines of text to being able to perform the same task on data sets exceeding 100,000 questions, an order of magnitude larger than previous benchmarks," Schroepfer wrote. "These advancements in computer vision and natural language understanding are exciting on their own, but where it gets really exciting is when you begin to combine them."
Though this research is still early on, Schroepfer said it could be a major development one day for people with visual impairments. With the A.I. technology, when their friends share photos on Facebook, the visually impaired will be able to find out what is in them.
Ezra Gottheil, an analyst with Technology Business Research, said Facebook is doing work that is important in advancing A.I. but also important to the social network.
"Facebook is in the content delivery business," he said. "It wants to give you more of what you want and less of what you don't want. That makes the user experience better. If you can find pictures, or particular pictures, with the objects you are looking for, with fewer mistakes, you are going to be happier. If you can speak your requests instead of typing them on those horrible smartphone keyboards, you will be happier."
Schroepfer added that Facebook's A.I. team also is working on teaching computers how to plan, as well as predictive learning, in which machines can learn through observation instead of direct instruction.
"Some of you may look at this and say, 'So what A human could do all of those things,' " wrote Schroepfer. "And you're right, of course, but most of us don't have dedicated personal assistants. And that's the "superpower" offered by a service like M: We could give every one of the billions of people in the world their own digital assistants so they can focus less on day-to-day tasks and more on the things that really matter to them."