Recognizing a robot when it rings your phone is undeniably easy. The reason is that a robot’s voice is unnaturally efficient, melodic and it rarely stumbles. In fact, the voice betrays the source even before it delivers the message, for instance, telling you that your mortgage is overdue, or requesting your input in a customer survey. For most people, the knowledge that they are talking to a robot makes it easy for them to hang up the call. Based on the brains behind Google’s Duplex, this situation may soon change.
Duplex was unveiled recently during Google I/O, the company’s yearly developer conference. According to Google, the project encompasses a new technology that allows Google’s machine intelligence-driven virtual assistant to carry out a natural conversation with humans through the phone, imitating the chit-chattiness that characterizes human speech while completing simple real-life activities.
Although the on-stage demonstration was prerecorded, hearing and seeing the concept at work amazed the audience present at the conference. In the initial demo, a woman called a hair salon, whereby another woman a picked up the phone. Both women engaged in a conversation for about one minute before agreeing on the ideal time for a hair appointment. The second demonstration involved a man calling a restaurant to make a reservation. Since the woman on the receiving end failed to offer the preferred information, the caller opted for a new request.
What surprised the audience was the fact that none of the callers was a human being. The voices were initiated by bots that had been authorized by Google Assistant and activated via a back-end system. Despite that fact, they sounded human. They even said things like “Ohh” and “um” as well as completed questions with the raised pitch that accompanies a question mark.
For Google, Duplex serves as the next significant venture into natural-sounding, fully-independent robot conversations. However, the project appears to be operating on a thin line between being eerily deceptive and largely convenient. Although Google is yet to launch the feature, it is expected to run in Assistant on various gadgets including compatible smart speakers and phones.
According to Google, Duplex was launched as an experiment several years back by Yossi Matias and Yaniv Leviathan, the VP of engineering and the company’s principal engineer consecutively. The project combines deep learning, text-to-speech, and natural language processing technology into a single service.
This technology raises the question regarding etiquette. There have also been questions regarding the possibility of human beings outsourcing calls that are uncomfortable for them to Google Assistant. Nevertheless, Google said it is currently limiting Duplex to only specified domains, even though John Havens, the executive director of IEEE Global Initiative on Ethics of Autonomous and Intelligent Systems, sees the possibility of expanding those limits in the near future.
According to Havens, it will soon be easy for somebody to key in words in a bid to have a virtual assistant break up with their partner or boyfriend. There’s also the risk of abuse whereby people may try to program Google Assistant to spam-call an enterprise even though Google said it would curb that by setting an unspecified number of calls by a single user.