Google: our “telephone” AI is not good enough to be dangerous

Not so long ago, Google demonstrated the digital assistant Google Duplex, which is able to talk on the phone, reserving a place in a hotel, hairdresser or restaurant instead of a person. Something like a virtual secretary, and with a very human voice. So human, that news resources began to publish information about the possible danger of an assistant.

Recently, Google representatives tried to do everything to convince those who are afraid of the safety of their offspring. In no case does the company intend to abandon the project. Moreover, during this summer, the capabilities of the assistant will become available to a limited number of users of Google Assistant. The assistant will help you to reserve a table in a restaurant or choose the necessary time for a haircut in the nearest barber shop.

Demonstration of the capabilities of the system struck many. And not only by the fact that the assistant clearly answered the questions and generally supported the conversation, but also by the fact that the communication between the machine and the person can hardly be distinguished from the communication of the person with the person. All these “Ugu”, “Emmm”, “Nuuuu”, which the program diligently emulated (of course, in the English version of all these exclamations), are very similar to how a person behaves during a conversation.

So what's the problem? Well, the program says to itself and says, if all this looks like communication between people, so much the better, the progress of computer systems is obvious. In fact, information security experts and ordinary people are afraid that the voice assistant can be used for a variety of purposes by attackers or representatives of various kinds of special services. And it is almost impossible to distinguish a computer from a person.

Well, to steal personal information with the help of such a system is easier than the easy one - she knows how to convince. In any case, in demonstration videos everything looks exactly like this.


“When demonstrating technology, many questions arise. One of them is: “Did you guys create a higher-order AI that can be very similar to humans, including phone calls?”, Said Scott Huffman, Google’s engineering vice engineer. Huffman says that in fact the digital assistant is very, very far from being human. “What we actually created are three automatically“ trained ”systems, each of which has a clear task: to ask about the schedule, to reserve and schedule the cutting time. That's all the car can do, "says Huffman.

Among other things, Google added presentation ceremonial to the system's functionality. The car reports that it is a car, so the interlocutor should not have problems understanding the situation.

As for the fact that the technology can not be used for quite positive purposes, Google employees have already spoken. “Many technologies can be used both for the benefit and to the detriment of humanity,” said Nick Fox, vice president of Google. He also believes that it is very important to let users know who exactly is calling at the moment.

To create such a system, the Google team asked regular call center employees to reserve a table in a restaurant or a haircut at a hairdresser. In the course of "communication", a certain character dialogue necessarily arose, which Google employees recorded and used for teaching AI. At first glance, it turned out quite well.


Now Duplex perfectly copes with the issue of booking tables, and it is quite anonymous. However, Google has a team of operators who can operate in the field.

What is interesting is that Google considers the current ability of the system to communicate is not very high.

In general, Google is now trying to prove that there is “nothing personal, only business” in its project. In principle, most likely, this is the case, because it is unlikely the company is trying to grow AI, which after some time will prey on humans. No, the point here is in pure commerce, and soon, I suppose, the issue will really be resolved.