Why Facebook Is Building Intelligent Machines

In last week’s newsletter about intelligent assistants, I didn’t mention Facebook’s fledgling initiative in this area. Unlike Siri, Google Now and Cortana, Facebook’s “M” project (currently in a private beta) won’t be baked into the operating system of your smartphone. Instead it will be a new feature inside Facebook Messenger, the company’s popular instant messenger app. I’ve been dismissive of Facebook Messenger in the recent past, because it’s far less sophisticated than similar apps in Asia – particularly China’s WeChat. But with M, Facebook is finally adding smarts to its messaging app. Not only that, M may turn out to be Facebook’s first ever “intelligent machine.”

Intelligent machine? Yes, that term is used by Facebook itself on the home page of its AI research lab: “In the long term, we seek to understand intelligence and make intelligent machines.” (emphasis mine) At first reading, this goal doesn’t seem consistent with Facebook’s mission to help people connect and share. What do intelligent machines have to do with connecting human beings? Quite a lot, if Facebook’s AI projects will have anything to do with it.

Facebook & AI

Facebook has two separate divisions dedicated to AI work. The Facebook Artificial Intelligence Research program (FAIR) is its research arm, while the Applied Machine Learning (AML) division is product-focused. The latter is already powering some of the Facebook user experience, according to AML director Joaquin Quiñonero Candela. At the company’s F8 developer conference in April, Candela noted that AI powers Facebook’s translation, photo image search and real-time video classification.

In a nutshell then, currently AI is being used to help filter and sort content on Facebook.

But there’s a big jump from doing that to conversing with intelligent machines, which is the aim of Facebook’s research division. Its director, Yann LeCun, told Fast Company that Facebook wants to teach machines “common sense.” To achieve that, it is building “natural language understanding for dialogue systems, which will be the basis of Facebook’s intelligent voice assistants.”

M: Facebook’s Prototype Intelligent Machine

As I mentioned above, M is an intelligent assistant program for Messenger. It’s being tested behind closed doors right now and Facebook hasn’t revealed when it will be widely released. But we get an idea of how it will operate from this Wired article:

…users will tap a small button at the bottom of the Messenger app to send a note to M, the same way they might message anyone on Facebook. M’s software will decode the natural language, ask followup questions in the message thread, and send updates as the task is completed.

That doesn’t sound much like an intelligent machine. But in a November 2015 blog post, Facebook CTO Mike Schroepfer explained that the company’s ambitions for M go well beyond mere conversation:

Unlike other machine-driven services, M takes things further: It can actually complete tasks on your behalf. It can purchase items; arrange for gifts to be delivered to your loved ones; and book restaurant reservations, travel arrangements, appointments, and more.

We can deduce from this that Facebook’s intelligent machines won’t simply talk to us, they’ll do things for us. So that tells us more about how M – and future “intelligent machines” that Facebook develops – will fit into Facebook’s mission to connect people. The machines will be conduits between people, helping us make transactions and carrying out manual tasks.

Let’s say you want to buy a birthday present for your Mom. In the future Facebook, you’ll be able to ask M to do that for you. Possibly M will analyze all the shopping websites your Mom has visited recently (remember that Facebook tracks many of us via its login across a myriad of sites). Or maybe M will simply remember the time your Mom commented on your cousin’s post, mentioning offhand that she likes Ronan Keating. Turns out Keating has just released a new album, so M buys it for your mom on your behalf. That’s just one potential way M could do tasks for you in the near future. I don’t know how exactly it’ll play out, but one thing is for sure: Facebook isn’t short of (your) data to feed the AI machines.

It’s important to note that M is currently only half powered by AI. According to the Wired article, M is also powered by “a band of Facebook employees, dubbed M trainers, who will make sure that every request is answered.” So for those questions that M cannot answer itself, an M trainer will answer. Which of course enables the AI to learn the answer, so that next time it won’t need the trainer. Eventually the trainers will be redundant, when the AI knowledge base is large enough to answer practically everything.


FAIR only launched in December 2013, so it’s not even three years old. Since intelligent machines is a long-term goal for Facebook, we may not see the results of its research for many years. But its M prototype at least gives us a glimpse into why Facebook wants to build intelligent machines. Put simply, Facebook wants machines to automate tasks for its users. Just like every other form of automation we’re seeing – from driverless cars to factory robots.

In 2016, chatbots are a huge trend. But longer term, Facebook is looking beyond chat functionality. Facebook wants “machines” (which may simply be apps) to walk the talk, to carry out tasks for you. And preferably before Google, Amazon, Viv and others achieve the same functionality.

Perhaps the big test will be: which of Facebook, Google, Apple or Microsoft will be able to buy the best birthday present for your Mom? We’ll find out in five to ten years time!