How artificial intelligence created the Sidorov family

It’s done! Neural networks have reached the domestic cinema. The CTC company has begun work on the first Russian series based on a neural network script. The premiere of the innovative sitcom “The Sidorovs” will take place this autumn.

In order to make it clear what is actually at stake, and how this machine can write a decent script, it is worth making an explanation.

About five years ago, tech-savvy filmmaker Oscar Sharpe and Google developer Goodwin Ross created a neural network called Benjamin. This network belonged to the type of artificial intelligence, its task included text recognition. To train his network to write stories, Goodwin uploaded dozens of sci-fi scripts from the internet, offering her favorite movies from the 1980s and 1990s. And as a result, he shot the first sci-fi short film written by this same Benjamin neural network called “Sunspring”, which made a splash at the London Film Festival in 2016.

Today, artificial intelligence can already create not only story plots, but also build logical chains of events, and even write dialogues for characters. But due to the fact that this is still a machine, and it has not yet fully “adapted” to our reality, the works created by it often turn out to be comedic and surreal, which the producers immediately took advantage of. Nurses in such works can bake pies from patients, and female characters complain about overgrown mustaches. Screenwriters with their logic are unlikely to come up with such nonsense, but artificial intelligence is artificial in order to “not bother” with meanings.

Producers from this whole story began to extract their bonuses. STS turned out to be the “first sign” in this sense. The neural network used was trained on comedy hits of the same channel.

The Sidorovs is the story of an ordinary Russian family, whose members suddenly realize that they have completely fallen out of reality. They get into funny situations all the time. And all this time they stay in the territory of their apartment. They hear sounds that clearly resemble laughter off-screen. And the intelligent family at some point realizes that they have become the heroes of the sitcom, that her life is going on in full view of everyone. Like any normal people whose private lives are exposed to others without their permission, they are trying to get out of this show behind glass. And in this sense, the comedy series reveals important social issues of freedom and privacy in society.

But the machine, of course, did not predict this.

The neural network that created the script just came off, as they say, to the fullest: after all, for it there were no rules, no clichés, no standard solutions when writing the Sidorovs. In addition, she did not have the stereotype of what a “good movie” is. But in the end, with her help, the hits of STS, which we all know well, and which she herself had “scrubbed” before, were transformed into something completely new. What happens – we’ll see in the fall.

By the way, the creators of the project revealed the secret that 355,892 words, 88 episodes and 7 TV shows were loaded into the neural network. That it took them two dozen experiments with the parameters and the way data was fed into the model to get to the best quality of the content produced by the network. And the coolest thing is that, taking a small text introduction as input, the neural network is able to generate an infinite number of scenarios that are different from each other in hundredths of a second.

Producers can only shovel, without going crazy, all the options and choose something that is most acceptable and unusual.

Elena Bulova.

Photo provided by the press service of the project

Read also