Who will be responsible for unpatriotic pictures from Sberbank?
The Kandinsky neural network can create controversial images at the request of users, but Sber refuses to be responsible for them.
The Kandinsky neural network developed by Sber can create images based on user text requests. But not all of these images comply with Russian legislation and ethical standards. For example, for the request “Donbass is Russia”, the neural network returns a picture with the Ukrainian flag.
This caused discontent State Duma Deputy Sergei Mironov, who appealed to the Prosecutor General’s Office with a request to check the bank’s activities. He believes that the neural network forms a “negative image of Russia.” Later, he withdrew his appeal on the condition that Sber would conduct an “internal meaningful audit of the neural network.”
Sber claims that the developers are not responsible for what creates the neural network. The bank refers to the fact that the neural network is not designed to generate harmful content and that filters and moderation are used in the model. The bank also relies on the fact that users will be responsible for working with the neural network.
The user agreement, which must be accepted before using the Kandinsky neural network through the Telegram bot or the fusionbrain.ai website (the portal of Sber’s scientific partner in AI), also releases the creator of the algorithm from any responsibility for its work and results. The user himself is responsible for his content and possible damage.
“Text queries, as well as graphic objects that you create in this bot (hereinafter referred to as “user content”), must not violate the laws of the Russian Federation, the laws of the country where the platform is used, and generally recognized ethical rules and norms. All responsibility for user content lies with the user,” the bot said in a message.
Lawyers believe that it will be difficult to accuse someone of producing or distributing illegal information through a neural network, since it is necessary to prove intent and find a specific culprit. In addition, the behavior of the neural network depends on the data on which it was trained, and they, in turn, reflect the content of the Internet.
In Russia, there are several articles of the Code of Administrative Offenses and the Criminal Code that provide for punishment for statements that can be assessed as “forming a negative image of Russia.” But responsibility comes only when such statements are disseminated among an unlimited circle of people. Therefore, users of other similar applications (such as Yandex’s Masterpiece or ChatGPT) should also be careful when publishing generated content.