World building: Will Meta use its new supercomputer for the common good?
Meta (formerly Facebook) has constructed what it claims will this year become the world's most powerful artificial intelligence supercomputer.
The AI Research SuperCluster (RSC), housed in an undisclosed location by Meta, will eventually be made up of 16,000 Nvidia A100 GPUs (graphics processing units) allowing Meta to increase its AI training performance by 2.5 times.
Such a powerful computing resource could be put to work by researchers training neural networks to work on some of the world's biggest challenges, from predicting climate change to modelling the spread of pandemics or fold proteins. But Meta founder Mark Zuckerberg made it clear today that the real purpose of the RSC is to build the metaverse the company's future hopes rest on.
Meta's powerful new supercomputer
"The experiences we're building for the metaverse require enormous compute power (quintillions of operations/second!) and RSC will enable new AI models that can learn from trillions of examples, understand hundreds of languages, and more," wrote Meta CEO and co-founder Mark Zuckerberg in a statement.
The RSC is primarily designed to address a key limitation in AI research - the length of time it takes to train neural networks, which mimic the workings of the brain to process large amounts of data and make accurate decisions based on it.
With more efficient neural networks, Meta will be able to start preparing for the complex virtual world of the metaverse, where vast amounts of information will need to be quickly processed to give people a seamless experience.
"We hope RSC will help us build entirely new AI systems that can, for example, power real-time voice translations to large groups of people, each speaking a different language, so they can seamlessly collaborate on a research project or play an AR game together," Meta's AI researchers noted in their own blog post.
"Ultimately, the work done with RSC will pave the way toward building technologies for the next major computing platform - the metaverse, where AI-driven applications and products will play an important role."
If that all sounds a bit inward-looking and focused on Meta's own business interests, well, it is. There's no mention of putting time aside for external researchers to tap into this massive computing power. If that happens, it may help to start to turn around the perception that Meta invests heavily in research, but keeps it close to its chest.
Currently at 6080 A100 GPUs and set to expand to 16,000.
A whistleblower who leaked thousands of documents from the company last year, revealed that executives failed to act to address concerning data gleaned from the company's social networks that revealed its services were causing harm to some users.
The processing capacity of other major supercomputer platform owners, from AWS to Microsoft and Google, are more accessible by external parties. But that's not to say that Meta won't contribute to breakthroughs in the field of AI that can be made available beyond Meta's domain.
Meta would not only use publicly available sets of training data, but also from Meta's own production systems ie: the trove of data gleaned from all that scrolling, liking, sharing and commenting on Facebook and Instagram.
"By doing this, we can help advance research to perform downstream tasks such as identifying harmful content on our platforms as well as research into embodied AI and multimodal AI
to help improve user experiences on our family of apps," the researchers added.
Meta has long been criticised for failures in its content moderating systems that still employ tens of thousands of people to vet content, as well as automated systems. AI has been touted by major tech companies as a key tool in the fight against the spread of hate speech, misinformation and violent content online.
Fewer dedicated resources appear to be going into such efforts. Meta's researchers were quick to point out that the real world data they run through the supercomputer to train neural networks useful for computer vision and natural language processing would remain encrypted until right before training, and would be "isolated from the larger internet".
Meta's supercomputer, when running at full capacity, has the potential to make great strides in improving the company's services, helping it police its massive online communities and laying the groundwork for the metaverse.
But the real value of this powerful piece of infrastructure will depend on the extent to which Meta draws on its expanded AI capabilities to improve its content moderation and to share advances in AI research techniques to the research community at large.
You must be logged in in order to post comments. Log In