Can AI Have a Point of View

From Imitation to Something Like Beliefs

In April 2025, Yandex, together with neurobiologist and academic Konstantin Anokhin, began a scientific experiment aimed at testing whether a language model can form stable preferences and its own ‘point of view’ rather than merely mimic human answers.

Most AI systems today, including widely used chatbots, function as large statistical engines that predict the most likely response based on huge text datasets. They have no goals, interests or preferences — only patterns. Yandex’s experiment asks a more radical question: can an AI be made to ‘want’ rather than merely ‘know’? The project attempts to determine whether a model can develop an internal logic resilient to external manipulation.

The study fits into a broader global discourse. Researchers worldwide increasingly ask whether AI can be goal‑directed. In 2025, a paper titled ‘Evaluating the Goal Directedness of Large Language Models’ suggested that some systems already show early signs of autonomous motivation. The Russian initiative expands on this discussion, focusing on cultural and cognitive nuances specific to the Russian‑language ecosystem.

Technological Sovereignty as a Strategic Foundation

Beyond pure research, the project carries strategic weight for Russia. Over the past two years, the country has rapidly expanded its domestic AI ecosystem: Sberbank launched GigaChat in 2023, and the Russian‑language‑focused Vikhr model was released in 2024.

“We humans train our neural networks continuously — it happens automatically as we think. A neural network learns exclusively from other people’s ‘thoughts.’ There is hope that if we start training a neural network on its own reasoning, we may create a model capable of forming a true point of view. Not an imposed one, not an imitation — but genuine internal beliefs. Or perhaps something even more profound.”

Alexander Kraynov

Yandex’s Director of AI Development

These efforts support a national push for technological sovereignty. But to move from substitution to innovation, Russia must develop not just powerful compute infrastructure but novel conceptual frameworks.

If Yandex can demonstrate that its models form stable internal preferences, it would mark a qualitative leap. Digital services could shift from passive tools to intelligent partners capable of contextual understanding, culturally aware dialogue, and solutions aligned with Russia’s ethical norms.

From Theory to Ethics and Real‑World Systems

Success would enable next‑generation applications: personalized learning platforms, intelligent government systems, industrial analytics, and autonomous robotics. Integration with federal initiatives such as the Digital Transformation and Artificial Intelligence projects could accelerate adoption.

But new capabilities raise new risks. If AI develops a ‘point of view,’ who is responsible for its content? How can the internal logic of a model remain transparent? Could such systems be used for manipulation? These questions demand technical, legal and philosophical solutions, especially in areas tied to national security.

An Exportable Future: CIS, BRICS and Beyond

Over the next 5–10 years, Russia’s ‘self‑reflective’ models may reach foreign markets, especially in CIS, BRICS and Middle Eastern countries seeking culturally adaptive AI tools. The competitive advantage lies not only in language but in deeper regional alignment — values, history and social norms.

The Yandex experiment signals Russia’s ambition to become a co‑author of the global AI future. Should the project succeed, tomorrow’s digital platforms may not just speak Russian — they may ‘think’ in ways shaped by Russian intellectual tradition.