Would you send a robot to a meeting in your place if you could? Look back over your calendar from the past month and count all the times you were unnecessarily stuck in a video conference call when you would rather have been somewhere else. And I don’t mean water skiing around a Caribbean island, but actually working on something that needed to get done.
Many of the practical advances in computer intelligence over recent years have handed mundane formalities to machines, often the kind that come after we make a decision. I’ve decided that I want to go to Rome for two weeks in September, but I’m happy to let the artificial intelligence of my smartphone assistant trawl the internet for the best price.
Google recently took that process a step further by unveiling Duplex, an AI assistant that can interact with humans–not just search engines–in pursuit of booking our restaurant tables and scheduling our haircuts.
Could Duplex, with its humanized speech patterns and ability to respond to external questions, one day become smart enough to attend a business meeting in our stead? Now that we’ve got robots communicating with humans on our behalf, could Google assistant video calls be next?
The Duplex Potential
Google revealed Duplex at the I/O conference in May, but the next-gen AI assistant only moved into public testing in late June. As it currently stands, a select group of testers and businesses are using the service to engage in the first human/machine conversations over the availability of a table for four at 8 p.m.
Duplex is an expansion of the existing Google Assistant technology that powers the tech giant’s “OK Google” range of products, from smartphone search engines to home entertainment hubs. Whereas the current version is basically a voice-activated search engine, Duplex can actually place audio calls and speak with humans to perform simple tasks like making a restaurant reservation.
The video below shows the AI in action booking a haircut:
If you know you’re listening to a robot then the overly formal sentence structure and mechanical pitch within some words give it away. That said, the fact that the salon receptionist acted as though they were talking to a human seems believable.
Aside from improving the way the robot sounds and introducing some human pauses and “ums,” what Google has done is expand the parameters its assistant can understand. The machine can respond to a human voice, just as other devices like Siri or Alexa can, but it is also able to negotiate a solution within a defined range–in this case, suitable hours for a haircut. As with all AI, it’s a two-fold solution: the machine sounds human enough for us to be relaxed in its presence, and it’s functional enough for the conversation to flex within the usual bounds of this kind of interchange.
So, what if that was your boss on the other end of the line, chasing an update on your project status?
AI As Human Surrogate
You can expect Duplex to be in the testing phase for a while yet, and to have more than a few bugs once it becomes publicly available for all. There’s just so much complexity in human conversation that refinement is going to take years. There are all our different accents and speech patterns to deal with, plus the emotion of a situation and the possibility of a tangential question. Skype’s live audio translation service, for example, has been with us for several years and can still descend into absurdity trying to keep up with even simple English-to-English captioning.
That waiting period aside, though, this is exciting technology. Duplex’s machine learning is built on the detections of patterns within thousands of hours of analyzed human conversations. As is the case with Google’s other recent AI release, voice isolation software that can pick out a speaker within a crowded room, the machine treats human conversation like any other logarithmic problem and searches for the underlying, repeatable aspects of the exchange. It works best in closed conversations like a booking request because there’s a clear goal involved–a desired time and place to be negotiated. It could even make perfunctory small talk along the lines of “How is your day?” but it would be easily stumped by the kind of wide-ranging conversation you’d have with a friend.
In a meeting scenario, however, where it was ordered to report on a predetermined status and take notes on new information and new requests, it could come in handy–perhaps reducing time spent around the conference room table.
Google Assistant Video Calls
AI is already creeping into our video calls. VC vendor BlueJeans recently launched a partnership with an AI company to include automated, smart notetaking bots within conference calls, while hardware makers including Logitech and Cisco are introducing “auto” everything, from auto-adjusting lighting and focus to auto-framing and cropping.
Those smart technologies make attending meetings easier, but Duplex could make attendance optional. Video conferencing has enjoyed a steady rise to prominence over audio-only communication in recent years, so Duplex would have to be adapted to work with a platform that includes video as well as audio. That could be as simple as having a placeholder photo sit in your chat window, or it could be an animated, automated avatar talking for you.
You could provide Duplex with a detailed update on your current project, with milestones recorded next to keywords such as “completion date,” “current status,” or “pending decisions.” Armed with this information, Duplex could sit in on a meeting and answer any questions related specifically to your project, while recording the entire discussion to video. Any unexpected questions from the boss or terminology it didn’t understand could be recorded as separate video messages for later access. If it really got confused, Duplex could always put in a call to your mobile and ask for instructions–though, as you can hear below, it’s pretty good at thinking on its digital feet.
As the technology stands, Duplex isn’t ready to forgive your absence at every video meeting with the team, but if you’re needed elsewhere or if you’re stuck in traffic, it could become a competent assistant to make sure you’re not left out of the video conference loop.