At the recent I / O 2018 developer conference, we were shown Google Duplex technology. In a nutshell, this is a new artificial intelligence system used in the Google Assistant app and Home speakers.
Duplex was created to save the user's time to call different establishments and services instead. More specifically, you can ask AI to order food delivery, book a table in a restaurant, sign you up for a beauty salon, find out the availability of a specialist you need, the availability of a particular vehicle in a car rental, and so on. Of course, this cannot but cause another wave of controversy about the danger of artificial intelligence, the uprising of machines, Skynet and all this. However, Google Duplex today interests us more from a technical point of view than from a sociopathic one.
The scenario of use is obscenely simple: you tell the virtual assistant, for example, to find out the availability of plane tickets for the required date. He calls, finds out the availability of tickets, their price, class and other details that you asked to clarify in your request. The voice of artificial intelligence sounds amazingly natural, so the person on the other end of the wire will not suspect that he is communicating with a machine, and not with a living client.
The only thing not completely clear is how Duplex will react to clarifying questions, the answers to which it cannot know. Let's say Google knows your home or work address, but what if you need to name the symptoms when making an appointment with the doctor? Or if you asked Assistant to order a pizza, but didn't say which filling? In general, there are still many different nuances here. Perhaps complex questions that require additional information will come to the smartphone in the form of push notifications or in some other form of interaction with a mobile device. But, most likely, the developers will quickly solve this problem, literally lying on the surface. Either they have already foreseen it, otherwise they would not have spent so much time and effort on creating such an advanced AI for dialogues: in addition to super-realistic speech synthesis, the system was even taught to use various parasite words and expressions such as 'uh', 'mmm' or 'uuu', which people use to fill in pauses and link sentences.
In the end, it should be noted that Google Duplex is an independent project that is not part of Assistant, but only integrates into it. If the latter can conduct more or less lively dialogues with the user on various topics, then Duplex is intended exclusively for ordering certain services and goods. Nevertheless, Google is able not only to close promising projects (take at least smart glasses Google Glass), but also to develop them to incredible heights (Android, let me remind you, was originally supposed to be an OS for cameras).
What do we have?
On the one hand, for the Russian-speaking IT community, Duplex is now nothing more than an interesting news for discussion about the success of a Western company.
On the other hand, a couple of weeks ago Yandex, using the example of 'Alice', who already knows how to order pizza and call a taxi, proved that it is quite possible to create artificial intelligence that works with a relatively complex Russian language. Therefore, as soon as Duplex is launched on the American market, we should expect something similar in our country.