May 30, 2018
Google Ups the Chatbot Ante with Duplex
Earlier this month CEO Sundar Pichai revealed Google’s latest virtual assistant, Duplex at the company’s I/O keynote. The company’s vision for the virtual assistant is, unsurprisingly, to help people get things done.
Pichai emphasizes some of the most cumbersome items on our to-do list can be calling businesses and service providers to schedule appointments. In the US alone, 60% of small businesses do not have an online booking platform setup. Google thinks AI can help us with this time- wasting problem.
What if your virtual assistant could call your salon and schedule a haircut for you?
Pichai continues the demo by showing Duplex doing just that. The voice of the virtual assistant sounds shockingly human. It speaks to the receptionist at the salon with natural intonations such as “um” and “m-hmm”. When the receptionist challenges Duplex when the first-choice appointment wasn’t available, Duplex reciprocates with ease.
After the appointment was successfully confirmed, the audience roars with applause.
The difference between Duplex and traditional chatbots is its ability to handle nuances. The technology leverages deep learning to carry on seemingly natural conversations.
The demo continues with a tougher test. Duplex, with a male voice this time, calls a small restaurant to make a reservation. The hostess on the other end speaks with a thick foreign accent that doesn’t faze Duplex. When Duplex requests a reservation for four people, the hostess says they only take reservations for five or more. The chatbot responds by asking how long the wait is on weeknights. The hostess responds and the conversation naturally comes to a close.
The demo was nothing short of amazing. But skeptics wonder how many calls did it take to produce something so flawless? Did Google fudge anything? The company will not disclose the names of the businesses called in the Duplex tests. The bot also raises some ethical questions about computers posing as humans. Duplex represents the culmination of years of work and pioneering R&D in machine learning at Google. Project leads and executives probably didn’t expect the demo to garner some of the controversial reactions that it did.
While many questions remain, the quality of the interactions exemplified by the Duplex virtual assistant reveals that the future of natural language chatbots may be closer than we think.