Facebook has been taking on some of the top names in AI and making management changes to help head off criticism about addiction to the site and its â??fake news' problem as artificial intelligence becomes ever more important
Facebook's hiring of French artificial-intelligence trailblazer Yann LeCun in 2013 to start its AI research lab signalled that the social media giant was serious about competing in the kinds of technologies revolutionising the web. Its highly visible brand of AI helped turn the flood of pictures, pokes and personal data into one of the world's most popular websites.
But criticism over election-meddling ads, "fake news" and the social network's impact on mental health - problems that Facebook is looking to artificial intelligence to help solve - has sparked questions over whether Facebook is keeping up with its rivals in the aggressively competitive world of AI development and research.
Facebook recently shook up its AI management, shifting LeCun to a more limited role in AI strategy, direction and external "evangelism." To manage the technology's growth, Facebook hired JÃ©rÃ´me Pesenti, who previously led IBM's AI platform Watson, as vice-president of AI.
The change in leadership highlights the rapid advancement of AI at Facebook - and how far it still needs to go.
"They are a significant player in AI today, where they totally weren't five years ago," says Pedro Domingos, a University of Washington professor, AI researcher and author of The Master Algorithm.
He says Facebook's team of roughly 100 AI researchers is a small fraction of the team at Google or Microsoft and far more limited in its scope.
"This is the Red Queen hypothesis," he says, referring to a concept in evolution stating that organisms must constantly adapt to survive. "It's not how fast you're running but how fast you're running compared to everyone else."
Facebook has said it would double the size of its AI lab in Paris. In all, the company currently employs more than 100 AI researchers in the United States, Montreal, Tel Aviv and Paris.
"The reality is that AI is more important than ever to Facebook," says company spokesman Ari Entin. "Our teams are growing. We're continuing to publish and open-source more than ever before, and deploying AI across Facebook at a really high level."
AI has long been the bedrock for the key features Facebook needs to gain new users and keep people engaged, such as facial-recognition systems in photo tagging and the algorithms that decide where posts land on users' News Feed.
Facebook chief executive Mark Zuckerberg personally recruited LeCun, a New York University professor known for his breakthroughs in deep learning.
Facebook's early AI lab was limited, but Zuckerberg voiced grand ambitions. "One of our goals for the next five to 10 years," he said in 2015, "is to basically get better than human level at all of the primary human senses: vision, hearing, language, general cognition."
AI seized a growing portion of Facebook's core technologies, including in recognising faces, language translation, targeting advertisements, captioning videos, pinpointing inappropriate content and recommending "people you may know."
Yann "has built a world-class team at Facebook, having it publish great papers and ship products", says Andrew Ng, a co-founder of Google Brain and former chief scientist of the Chinese tech giant Baidu, including what he says are impressive feats in natural-language processing and image recognition.
The technology is also central to some of Facebook's most controversial, high-visibility issues, including how to deal with misinformation or hate speech, discriminatory advertising, and questions of attention manipulation and social media addiction.
But some experts say Facebook's AI teams have, like at other tech giants, struggled with the natural tension between the teams in research, with their long timelines and academic pursuits, and application, whose mandate involves building features that actually work.
Hilary Mason, a vice-president of research at Cloudera and founder of the machine-learning research firm Fast Forward Labs, adds that few companies of Facebook's scale have achieved the perfect balance between research and results.
The company, she adds, remains incredibly alluring to AI talent: "Data science types are attracted to where the data is, and Facebook has some of the most interesting data."
Facebook competes for talent with tech giants such as Amazon and Google's DeepMind, which offer researchers a chance to work outside a social network and explore the possibilities in voice assistants, health, gaming and drone delivery.
At Facebook, Pesenti will oversee the division LeCun runs, Facebook AI Research, and Facebook's Applied Machine Learning division, which builds and deploys the AI used by more than two billion people around the world every month.
Domingos says AI is critical to keeping the site successful, engaging and alive.
"Facebook has this amazing business where they don't even have to troll the web for content. People just upload their stuff and then they serve it back out with ads attached, and they print money. It's great to be Facebook," Domingos says. But its "machine learning has to respond. And if it doesn't respond, the whole site will be in much worse shape."