Machine Learning and Word Embeddings technology was used to reveal the difference between propaganda and news. Two AIs were “brought up” by a state and an independent TV-channels for 6 months. Then they were taught to answer questions about Russia, but only using the data they “had learned” while watching news programs of those channels.

Russia’s media channels are subordinated to the government. They always present the news in the way that is beneficial to the government. They are completely “pro-Putin” and know how to brainwash people’s heads. There is a channel that is opposed to them; it is not financed by the government and doesn’t take instructions from anyone, so it can tell about things that are hushed up by the federal channels.

Obviously, the content, vocabulary and manner of news representation is very different. The worldview of a person who watches a different channel is different too.

The goal of the campaign was to increase awareness of the independent channel. It was also to reveal a deep mental problem of information environment as the crucial factor influencing people’s minds.

With Google Speech-to-Text service and Word Embeddings technology the trained neural networks could identify the language of TV programs, recognize words and break texts into paragraphs. Then they were taught to choose relevant questions. The AIs are childlike, immature minds who don’t have all answers – only those they can compose from the content they “consumed” for 6 months.

As a result, two pure and pristine AIs were created; they are like twins, but with no experience and no outside factors influencing their outlook. The two systems were given a task to “watch” daily news. Then the AIs were taught to answer any questions about political, economic and social life of Russia. Asking a question, a person becomes a participant of the experiment, because the result is unpredictable and it depends on how a question was asked. It provoked interaction between users and the AIs. People were also eager to share the results of communication on social media, it led to many reactions and posts.

The chosen strategy was targeted at politically active people interested in obtaining true information. They actively use social networks and live in metropolitan cities. There’s no use in trying to impose an opinion on them, as they are intolerant towards direct advertising. They are used to making choices, comparing and experimenting on their own. That’s why they were offered to participate in an experiment that would show how a worldview is formed depending on the information environment. Asking a question, a person becomes a participant of the experiment, because the result is unpredictable and it depends on how a question was asked.

A mix of innovative technologies (Google Speech-to-Text, Trained Neural Network, Question Answering Model (Q&A), ODQA, Word Embeddings) helped AIs absorb terabits of information. After 6 months, they could answer any questions. But only using the knowledge and vocabulary of the programs they had watched. Using the website interface, everybody can ask a question to the AIs and get answers from them or read related answers. Also, one can vote for the system he/she likes more.

The story rocketed around the Internet and media.

In one week

14 048 080 media impressions

193 427 visitors

93% voters favored TV Rain's AI