Facebook’s AI research lab in Montreal to expand as it marks 1-year anniversary
Social media giant heavily relying on technology to hold empire together
Facebook doesn’t make magnetic resonance imaging machines and has no plans to start.
But that hasn’t stopped researchers at the social networking giant’s Montreal artificial intelligence lab from working with NYU School of Medicine to make MRI scans faster and more accurate.
Facebook celebrated the oneyear anniversary of the Montreal lab Thursday by highlighting the MRI project and some other research work. The company also announced that the AI lab will move into a new space that can accommodate up to 80 people, which gives the team room to expand from the 20 or so researchers currently working there.
Joelle Pineau, an associate professor at McGill University and the lab’s head, said that medical imaging was a good fit for the kind of “fundamental research” on artificial intelligence that Facebook hired her to do.
“I had one earlier this year; they had to image my shoulder. It was like 45 minutes sitting in the machine,” she said.
Part of the problem is that MRIs involve a series of pictures that act like “slices” of the human body, that are assembled to create a 3D model, a time-consuming process.
“What we’re looking at with our technology is, what if you don’t need to look at all the slices to build up a full 3D image?” she said. “We take a few slices and based on the information in those slices, we decide what slices to get next. So now we have an adaptive acquisition procedure.”
The idea is that if an algorithm can make guesses at how to fill in the missing parts of a 3D image, and then test those guesses to improve accuracy, there could be all sorts of applications beyond medical imaging.
This kind of research matters to Facebook in a big way, because the company is relying heavily on artificial intelligence to hold their social networking empire together.
When Facebook CEO Mark Zuckerberg testified to the U.S. Congress in the wake of the Cambridge Analytica privacy scandal, he repeatedly talked about artificial intelligence tools as a key tool for the social media platform.
Already, Facebook uses AI to detect terrorist propaganda and block bad actors from setting up fake accounts. Zuckerberg said other problems are more difficult for AI tools, such as the persistent problem with hate speech.
“Artificial intelligence” can be a nebulous term, but for the most part it involves using machine learning algorithms, and in particular deep learning neural networks.
Deep learning is extremely good at certain types of problems, especially when there’s a lot of clearly labelled data to “train” the neural network, and in situations where the system can make predictions that have clear “right” and “wrong ” answers.
This sort of thing is already very widely used; if you’re reading this, you’ve almost certainly interacted with a machine-learning algorithm already today.
“Our digital life is essentially completely mediated by machine learning: What you see, what order you see it, when, accompanied by what advertisement,” Pineau said.
“Not just on Facebook; across platforms, your whole life is mediated by machine learning right now.”
In another experiment, Pineau said researchers in her lab are trying to teach a neural network so that they can feed a photo of food into the system and the artificial intelligence will try to generate a recipe for how to make the food.
She said the results are “actually quite impressive” but it’s difficult to evaluate if the machine’s recipe is “correct” or “incorrect.” She suggested that they might have to do a cook-off to figure it out.
Pineau said that she’s noticed a big difference in how people talk about artificial intelligence in the past year since the Montreal Facebook AI lab was founded. It’s not just Zuckerberg talking about the opportunities and challenges with artificial intelligence; everybody is talking about it now.
“It’s a conversation that now is happening out in the open, not just in the lab,” she said.
“I think that’s healthy; I welcome that opportunity that we have to talk about it more broadly and to listen more broadly to the concerns.”
Our digital life is essentially completely mediated by machine learning: What you see, what order you see it, when, accompanied by what advertisement.