The Inter-and Intra-Institutional Politics of AI Standards-Making

Technical standards are a mechanism for translating high-level aims into implementable practices. However, far from being "neutral" technical documents, it has long been recognised that standards, and the processes to develop them, are political. In this article, we analyse the politics of standards-making for artificial intelligence (AI). Drawing on fifty semi-structured interviews with standards-makers and standards-experts, we assess the political dynamics between AI standards development organisations (SDOs) and within the two most important AI standards committees: the international ISO/IEC JTC1/SC42 and the European CEN-CENELEC JTC21. We find that there are a multitude of international institutions developing technical standards for AI and that competition between them is facilitating low adoption rates. Within this landscape, the robust procedures of SC42 and JTC21 leave them relatively insulated from interstate geopolitics, but these same procedures create uneven participation and influence. Notably, they simultaneously lead to too much and not enough "Big Tech" influence. From these findings, we introduce the concept of "insulated ineffectiveness" to describe how the robust processes of technocratic standards institutions can undermine their practical authority in environments with low institutional switching costs.