Learning sovereignty: AI’s next enterprise battle

Artificial intelligence has emerged as a new battleground for enterprise sovereignty. While companies have long understood the value of their data and the importance of data protection, AI introduces a new question: who owns the learning and knowledge generated by the technology? Microsoft CEO Satya Nadella recently referred to this issue as the Reverse Information Paradox, highlighting that companies purchasing AI can end up paying twice: first for access to intelligence and then by supplying the proprietary knowledge required to make that intelligence useful.

Consider a refinery, for example. While a foundation model may understand the refinery's operations, it does not possess the accumulated operating experience of the specific refinery. Similarly, an AI model may grasp the general principles of procurement, but it lacks the nuanced knowledge of an organization's unique supplier history, exceptions, and risk appetite.

An engineer making corrections to AI recommendations or an operator overriding an agent due to unsafe conditions on a plant also contributes valuable learning to the system.

As enterprises grapple with these emerging challenges, they must also consider the complexities of "learning sovereignty." While data sovereignty focuses on data storage locations and legal jurisdictions, learning sovereignty pertains to the control, retention, and transfer of the mechanisms through which AI systems improve their understanding of a business.

This includes evaluations, agent traces, model adaptations, and other records of how work gets done. Enterprises need to address questions such as where these learning artifacts reside, who holds contractual rights over them, and whether they can be moved to different platforms while maintaining control over the learning process.