Facebook's virtual assistant M is super cool, but can it scale?

Facebook M virtual assistant artificial intelligence Siri Cortana
Credit: Facebook

Is there any way this innovative new human/algorithm hybrid virtual assistant could serve millions of users?


Last week, Facebook unveiled M, a fascinating virtual assistant feature built into the company's Messenger app. A new challenger to fully automated virtual assistants like Google Now, Apple's Siri, and Microsoft's Cortana, M turns the volume up to 11 by backing up the algorithms with actual humans.

Hybrid virtual assistant

According to the Facebook blog post by David Marcus, "M is a personal digital assistant inside of Messenger that completes tasks and finds information on your behalf. It's powered by artificial intelligence that's trained and supervised by people."

An undisclosed number of these "M trainers" assist the AI in answering tough questions and completing tasks that even the best software can't yet do on its own. Marcus says that M can "purchase items, get gifts delivered to your loved ones, book restaurants, travel arrangements, appointments and way more."

See also: Is Facebook still an innovative company? 

Facebook testers have reportedly already been using M to help organize dinner parties, track down unusual beverages, make DMV appointments, and call the cable company to cancel service. Apparently, M claims to be able to do just about anything that isn't illegal, though that isn't always a clear line. And reports say that while M typically responds quickly, it can take half an hour to respond to difficult questions (perhaps that's the humans kicking in?). 

090215 facebook m2 Facebook

Super cool? Yes! Scalable? Who knows?

All of this sounds super cool to me, especially the part about dealing with the DMV and the cable company. And M obviously goes way beyond the capabilities of software-only virtual assistants like Google Now, Siri, and Cortana.

But unlike those services, Facebook M is currently available to only a few hundred selected Bay Area residents. And there are already competing human-powered virtual assistants/concierge services such as Magic (or even outfits like TaskRabbit), that promise to take on a wide variety of tasks, including supplying information.

If Facebook hopes to push M beyond the relatively limited audience of those services, it will need to rely much more heavily on AI and less on the trainers. That's clearly the company's goal, but here's the thing: It's difficult to see how M could ever get to the point where it could serve hundreds of thousands of users, much less the billion people who signed on to Facebook one day last week. Marcus' blog notes, "This is early in the journey to build M into an at-scale service."

Humans still required

No duh. Even if the M trainers are successful in improving M's AI algorithms, there are still lots of things that M promises to do that will be very difficult for even the best-trained software to successfully complete alone. And it will be prohibitively complex and expensive—even for a company as rich as Facebook— to put together a big enough corps of humans to ensure swift, reliable service for truly large numbers of people. 

Sure, targeted ads and business arrangements with preferred partners to whom M drives customers could help defray some of those costs, but taking M from an experiment for a few hundred users in one metropolitan area to a widely used, geographically distributed offering seems like an impossible task with a puny payoff.

Maybe it would help to think of virtual assistants the way I think of wearable technology today: As a potentially transformative idea, with lots of worthy attempts in the works, but still very far from delivering mainstream, everyday solutions. Facebook M's human/software hybrid represents an important advancement in this quest, but lots more work—and some big breakthroughs—will be needed to create a solution that economically solves these problems at scale.

Must read: Hidden Cause of Slow Internet and how to fix it
View Comments
Join the discussion
Be the first to comment on this article. Our Commenting Policies