Facebook has been pushing hard into the realm of video, both live and prerecorded, in their latest efforts. It should come as a surprise to nobody that video was featured heavily at their F8 developer conference. It should also be no big shock that Facebook is attempting to use new tech, in this case artificial intelligence, to push the video envelope to make content more interactive and engaging. The latest innovations in video that they’re playing with include using A.I. to identify people in videos, as well as to automatically generate captions for the videos based on what people are saying.
Facebook’s head of applied machine learning, Joaquin Quiñonero Candela, called what they’re working on “image search on steroids”, comparing it to Google Photos and saying that Facebook’s A.I.-based system would be able to further refine the formula, check out more details and make searching more natural. As for video, he said that the new tech that Facebook is working on, incorporating A.I., would allow users to search the video in detail by content. Users may be able to type in a word or phrase and find videos bearing it, or find that content within a specific video. They may also be able to search a person and find what videos they’re in and when they appear, even down to warping to the exact second the person steps into the frame in the videos they feature in. He also said that scalability will be key with applying the technology to video search.
Facebook’s plans to intertwine A.I. and video run quite deep, including the creation of large-scale neural networks, hardware made specifically to run A.I. algorithms and open-source software. They have shared the plans for the specialty hardware, essentially allowing any entity with deep enough pockets to build an A.I. using open-source code on the same scale as Facebook’s A.I. efforts. Plans were also announced earlier in the year to send some of their specialty A.I. servers to academic establishments all over the world, allowing widespread analysis of their network hardware and software in the hopes of improving the field of A.I. overall.