AI regulation: maintaining interoperability through value-sensitive standardisation

What the paper says

Artificial intelligence (AI) regulatory frameworks are being used to emphasise specific values in previously purely technical areas. One clear example is the European Union’s (EU) AI Act regulation, which sets rules on AI while relying on technical standards to guide compliance and strengthen global cooperation. However, it also states that regulatory efforts have to be in accordance with Union values. Hence, we analyse the EU's approach as an example of the importance that values have gained in AI regulatory frameworks and standards. Therein, we argue that diverse conceptions of values could prevent interoperability between AI systems across regulatory contexts. To illustrate our argumentation, we identify values stated in AI regulatory frameworks from the EU, the United States and China and compare them according to the Theory of Basic Values. The comparison highlights that maintaining interoperability is crucial for regulatory frameworks and standards aiming to advance AI-based systems’ successful deployment and function in and across different regulatory contexts. Consequently, we propose a value-sensitive AI standardisation approach based on our analysis of AI regulatory values, a previous ISO/IEC/IEEE standard’s approach and value sensitive design. Our approach aims to relate AI regulatory values to requirements in AI regulations and provide suggestions for value-based configurability in AI standards. While we find there is no harmonised approach for the embedment of value-based consideration in AI standardisation, this could be a tool used to maintain the interoperability of AI systems that have global reach. Our proposed approach aims to address this shortcoming within the interplay of global AI markets and regulatory strategies.