- Oops!Something went wrong.Please try again later.
Facebook has built what it says is the world’s most powerful supercomputer devoted to artificial intelligence, which the social media company believes will drive previously-impossible advances in the technology.
Meta, Facebook’s parent company, said the Research SuperCluster (RSC) machine was up to 20 times as fast as its existing machines, and would bring AI programs a step closer to human levels of capability in some areas.
It said the system would build on recent improvements in AI such as language translation and image recognition to pave the way for the “metaverse”, the immersive virtual worlds that the company has said will be its future.
For example, the computer will mean groups of people all speaking different languages will be able to speak with one another in real time by instantly translating their speech to other people’s mother tongue.
Meta said the RSC system would help the company better analyse videos that feature harmful material or hate speech.
The company has sought to deal with widespread criticism over its content moderation by relying more on AI. However, the monitoriing of videos or speech in real time, such as groups of people conversing on a voice chat, requires massive computing resources.
The supercomputer will be the world’s most powerful dedicated to artificial intelligence when it is completed this summer, Meta said. At that point it will be able to learn from trillions of individual photos or video clips. It said the system needed to handle quintillions - millions of millions of millions - tasks each second.
“AI can currently perform tasks like translating text between languages and helping identify potentially harmful content, but developing the next generation of AI will require powerful supercomputers capable of quintillions of operations per second,” Meta said.
The company is likely to contend with growing pressure over keeping users safe as it invests in the metaverse, which experts have said could pose a new set of harassment problems.
Its chief technology officer Andrew Bosworth has said he wants the metaverse to have “almost Disney levels of safety” although policing it fully will be “practically impossible”.
Meta said: “With RSC, we can more quickly train models that use multimodal signals to determine whether an action, sound or image is harmful or benign. This research will not only help keep people safe on our services today, but also in the future, as we build for the metaverse.”