Microsoft has published a code of conduct for its own AI models, which the company links to the concept of "humanistic superintelligence." The document is currently considered a working draft and is open to suggestions from experts and the community. Therefore, its provisions may change, and some of them are more about guiding the direction of development than establishing technical requirements.
The code is based on ten principles. Microsoft prioritizes human interests over those of AI and does not believe it is appropriate to grant models rights or legal personhood. The company also insists on maintaining the ability to interrupt operations, adjust behavior, and completely shut down the system. The model must adhere to the code even when it hinders the completion of assigned tasks. There is a specific emphasis on the need to maintain human control over AI system interactions and to avoid creating excessive dependence on assistants among users.
Another block of rules addresses the structure and behavior of models. Microsoft opposes the development of systems that simulate consciousness, as well as interactions of AI in forms that are inaccessible for human understanding and control. At the same time, the company supports diversity of cultures, views, and values, but ties it to fundamental principles of safety, autonomy, dignity, and human rights. The document itself acknowledges the complexity of such an approach: in practice, different values may conflict, and the definition of acceptable behavior directly influences the operation of models.
Related articles
