Operator: Your next shopping experience starts with a text
Operator wants to “unlock the 90% of commerce that’s not on the Internet”, CEO Robin Chan tells me. After two years in stealth, Chan was finally willing to give TechCrunch a peek at his startup, which he sees as the convergence of the biggest themes in tech: mobile, messaging, and the on-demand economy.
Operator calls itself a “Request Network”. It’s an app that uses a network of human ‘Operators‘ to fulfill customer requests. It can handle a broad range of commercial requests. For now it’s focused on “high-consideration” purchases that may require expertise or have lots of options to choose from.
Mr Camp co-founded Uber along with Travis Kalanick. Operator does not have any formal agreements in place with the ride-hailing app, but is closely watching the development of UberEverything, Uber’s logistics and delivery service, as a potential partner.
The upcoming holiday season is poised to be the first big test of digital concierge services as consumers turn to Christmas shopping or make reservations. A challenge for traditional mobile commerce has been getting customers to complete the purchase — users often find it too time-consuming or inconvenient to input their credit card number into a webpage on their smartphone, for example — and digital concierges are trying to change this
Facebook’s new virtual assistant for Messenger, M, is pretty darn impressive.
At this point, M can do pretty much everything an actual human assistant might be able to do, short of picking up your dry cleaning. (Although it could arrange to have it delivered!) That’s great news for Facebook. The company is rolling out M as a way to keep people using Messenger and, eventually, get them shopping inside of it. An assistant to make that easier will certainly grease the skids on those efforts.
But there’s actually a simple reason for why M is so advanced. For the most part, M is much more human than it is software. Or rather, it’s powered by actual humans much more than it is by software.
The artificial intelligence technology used to power M is still in a very early stage, which means that while the system is learning some of the basic responses for popular requests, human moderators handle the bulk of the interactions with actual users, according to Facebook’s chief technology officer Mike Schroepfer.
“It’s primarily powered by people,” Schroepfer explained. “But those people are effectively backed up by AIs. The idea here is, you can ask it any question, not just the set of questions that it’s capable of. The thing that’s cool about this is it gives us a much wider training set, like what are the things people actually want it to help them [with].”
In other words, making it human-powered versus machine-powered allows Facebook to get a more authentic glimpse at how people want to use the product.
Right now, Facebook is training M with supervised learning, a process where the computer learns by example from what human trainers teach it. If a user asks A, you respond B. Eventually, the idea is that M will know enough to operate without a human handler. Facebook has a team of people building neural networks — applications that help machines think and act like humans — and many of those applications are already live inside of M, Schroepfer says.
That doesn’t mean that M will fly solo any time soon. The feature is only available to a small group of beta testers in Silicon Valley, and the technology needs to become much less human-dependent before Facebook passes it out more broadly, Schroepfer said.
“The reason this is exciting is it’s scalable,” he added. “We cannot afford to hire operators for the entire world to be their personal assistant.”
Schroepfer also showed off a new tool Facebook is building that can actually describe what’s in a photo, and vocalize it through a verbal Q&A process with a user. So, if you asked Facebook what was in a picture, it could — without ever having seen the picture before — respond correctly, based on other photos it has seen. This tech hasn’t rolled out to users yet, but Schroepfer hopes that someday it will.
These efforts are part of a much broader push from Facebook to dive into artificial intelligence and deep learning as a way to personalize its service. It has one of the world’s top deep learning experts, Yann LeCun, running its AI division; the eight-person team from machine-learning startup Wit.ai, which Facebook acquired in January, is running M. The company won’t say how many operators it’s using for M, but BuzzFeed found that Facebook is using outside services like TaskRabbit to complete some of the requests.