Facebook wants to know what you're going to do before you do it, and has hired a big-brained boffin to give it the AI chops to do this.
Machine-learning luminary Yann Lecun announced on Monday that he had been tapped to run the social network's nascent Artificial Intelligence laboratory. The hire represents another way the data-fattened social network plans to wring more cash from your photos, status updates, and aimless likes.
"Facebook has created a new research laboratory with the ambitious, long-term goal of bringing about major advances in artificial intelligence," Lecun wrote. "I have accepted the position of director of this new lab. I will remain a professor at New York University on a part-time basis, and will maintain research and teaching activities at NYU."
Some of Lecun's inventions include tech for recognizing and processing handwriting, and automatically recognizing and organizing images. At Facebook, the New York-based Frenchman is tasked with developing new approaches to machine learning to create weak artificial intelligence systems that have predictive capabilities, he told El Reg on Tuesday.
"Full prediction... for Facebook is very important," he said, adding he wanted to model how the actions of individual residents of Zuck's data farm "evolve over time."
"Being able to predict what a user is going to do next is a key feature," he said.
Though some may scoff at our conjoining of machine learning with the happy-go-Skynet term "Artificial Intelligence", Lecun thinks it's an accurate characterization.
"In some ways you could say intelligence is all about prediction," he explained. "What you can identify in intelligence is it can predict what is going to happen in the world with more accuracy and more time horizon than others."
Utility companies and telecommunications firms have been using machine-learning models to predict failures in infrastructure and figure out when customers are going to switch to a rival "for a very long time," he said. Google, meanwhile, uses machine learning for features like its voice recognition and image categorization technologies.
"There are much more complex systems to model, like human behavior, or language, or social behavior," he said, but cautioned this means "the model has to be extremely complicated."
The payoffs of developing an effective predictive model are huge – just imagine if a company had technology that could effectively predict house prices in advance, or a person's likelihood to click on a specific advert.
It's no coincidence that Google chief Eric Schmidt wants the ad-slinger to one day build not a search engine, but a "serendipity engine."
For this reason, both Facebook and Google are striving to push the tech further, and will use their vast stores of data to train new models on tough, multi-variant problems.
"The access to data is what allows you to nail a problem," he said. "It diminishes some of the problems we're facing and increases some of the others. When you have tons of data need very fast systems to train your models which requires digging down to the hardware."
Lecun's hire follows several years during which Facebook has rapidly grown its internal technical capabilities to let it better reduce its costs and analyze more data more effectively than arch-rival Google.
The period of rapid invention created the "Open Compute Project" for low-cost standardized hardware, and a bevy of open-source data analysis technologies such as RocksDB and Presto. Zuck & Co have also exposed this flashy new tech directly to users via the Graph Search feature.
With the formation of the AI Lab, Facebook is now commencing a period of heavy investment into a strategic area that could yield new non-advert revenue generation tools, and a research group that could help it keep its best and brightest from the headhunters of other Silicon Valley firms.
It also lets Facebook further refine its vast stores of data into higher value systems, letting it make more money and at a higher margin out of the data that some one billion accounts have freely donated to it.
Facebook's new artificial intelligence group was announced in September, and will have locations in London, UK, Menlo Park, California, and New York City. ®