“Rain” taught two neural networks: one on its programs, the second on “Russia 1”. They can ask them questions • TJournal • RIMA

“Rain” taught two neural networks: one on its programs, the second on “Russia 1”. They can ask them questions

To the question “There are many rich people in Russia” “no more” and “literally one and a half replicas” answers.

Training of two AI systems: on the left on the "rain", on the right at "Russia 1" photo "Rain"

In January 2019, Rain, together with developers from ISD GmbH and The App Solutions, launched an experiment to teach two neural networks. They were given to watch two types of television shows: news on "Rain" and news on "Russia 1" - in order to find out how much their answers will differ. In early April, the channel launched the first version of “communication” with both AI systems. This was told by TJ Executive Producer of the TV channel Svetlana Gulyaeva.

Updated at 19:00: in "Rain" they explained that DDOS attacks were made on the site. Developers restore his work and will strengthen the defense.

The experiment was carried out in January, and now neural networks by human standards-10-12 years. According to the developer Andrei Sergeyev, they created two identical systems in which the same principles and algorithms work.

To begin with, we load large volumes of transcribed TV shows into a neural network. Then she independently converts words into vectors. To understand how it is, here is a popular example. If we take the vector representation of Moscow and add the vector representation of “Russia”, we will get the concept of “capital”. According to the same principle: “Moscow” minus “Russia” plus “Great Britain” will be equal to the vector representation of a close to “London”.
With such examples, the system begins to analyze the relationship in our speech.
Victor Schkurba

The difference between neural networks is only in what content they are studying - on the news of the state channel or independent journalists. On the site you can ask questions to both systems: they do not always give accurate answers and are often confused in the wording. For example, both the rain system and the Russian system 1 could not answer the question of whether the Russian Internet should be prohibited. Otherwise, they are sharpened to answer questions about politics, society and life in the country.

Several questions of the rain neuralates from TJ - about Putin, Freedom and Ukraine

#artificial_intellect #neural networks #journalism #experiments