Meta’s artificial intelligence (AI) branch held a Facebook livestream overnight Asia time to showcase how AI is being used to build the metaverse and announce a new AI-enabled voice assistant program that will give Apple’s Siri a run for its money.
The social media giant rebranded to “Meta” last year to signal a pivot in company focus to the metaverse, sparking a race for brands to rush in to capitalize on the renewed interest in the subject.
“The way that we use computers now adapts much more to what you’re doing, and as devices have gotten better at understanding and anticipating what we want, they’ve also gotten more useful,” said Meta CEO and founder Mark Zuckerberg early in the livestream. “Now, I expect that these trends will only increase in the future.”
While voice assistants like Siri or Amazon’s Alexa have been limited to people’s phones or smart home terminals, Meta’s new assistant “CAIRaoke” (pronounced karaoke) will operate as an augmented reality program through a pair of smart glasses. Seeing what the user sees, it can provide visual cues, recommendations, and analysis through a heads-up display in the lenses.
Meta has made voice recognition a major part of its AI development and has started training the programs on actual speech as opposed to text-based learning.
Many AI systems have been criticized for carrying with them the subconscious biases of their developers — such as facial recognition software often displaying higher accuracy in identifying white, male faces, as that is the data set they were trained on. The same is true of translation services, which often use English as the default language and the reference point for all translating. Meta is looking to avoid these biases by training its programs to translate directly between languages.
It is hoped soon programs like CAIRaoke and others in the metaverse will be able to provide speech-to-speech translation in real-time. Zuckerberg said five years ago their programs could translate a dozen languages and hopes it will be capable of over 100 by the end of the year, even for languages with little online presence.
Speaking with Forkast recently, CEO of SingularityDAO Marcello Mari said a future where AI is making more decisions for human society is inevitable, and blockchain can help it be developed fairly and transparently.
“The question is: ‘Do we want this artificial intelligence to be developed behind closed doors by Google, Facebook, Alibaba or Amazon?’” Mari asked rhetorically. “Or do we want this artificial intelligence to be developed by the people on a transparent and permissionless ledger where everybody can see what’s happening?”
Meta is using a new form of machine learning called “self-supervised” learning (SSL) to dramatically improve the learning time and capacity of these programs. SSL means the programs are fed huge troves of incomplete data and learn to form connections and patterns on their own, as opposed to “supervised” learning in which curated data samples are shown to them by programmers and the pathways explained.
Supervised learning is by necessity much slower due to the level of human input required, as opposed to simply uploading huge amounts of data and letting the machines figure it out themselves.
“This actually seems closer to how the brain learns,” Zuckerberg said. “For example, you don’t need to show a kid thousands of pictures of a cat for them to understand what a cat is.”
Meta’s brand has long centered around connecting people, and it seems determined to carry that branding mission across to its next stage as a metaverse company. It has held to this branding even as it has been rocked by a series of scandals in recent years regarding the type of content it chooses to promote to its users and how it handles users’ data.
As the firm is under more scrutiny than ever before and recently reported its first-ever quarterly decline in user numbers on its flagship platform Facebook, many will be watching to see if it can live up to these ideals.