Introduction
The T2 – Standardisation activity engagement indicator provides an overview of the AI standardisation landscape across multiple international and European standards development organisations (SDOs). The indicator is composed of the total count of AI standardisation deliverables identified over the course of an extensive study, together with the subset of those that have been classified by AI Watch experts as significant in the context of the European AI regulation. The criteria for classification as significant favoured first-level standards dealing with AI-related risks in a horizontal manner and covering implementation aspects, as opposed to foundational and basic standards.
The indicator is complemented by a measure of participation by EU27 Member State national standardisation bodies in a key ISO/IEC technical subcommittee expected to be an important source of future harmonised standards supporting the European AI market.
World
An initial, broad analysis of the AI standardisation landscape resulted in approximately 140 standards and standardisation deliverables identified (left hand side of graph) (Nativi and De Nigris, 2021). The analysis drew on multiple specialist sources, including publicly available surveys on AI standardisation and scientific publications, as well as direct interaction with experts from international and European Standardisation Development Organisations (SDOs), e.g., in the context of events and activities oriented towards the creation of standardisation roadmaps. The overall landscape includes AI standardisation deliverables from four of the major international and European SDOs, namely ISO/IEC, ETSI, IEEE and ITU-T.
In a second step, a subset of significant standards was selected for further analysis from the perspective of the requirements laid down by the proposed European regulation on AI (right hand side). More than half (57%) of the initially identified standards and related deliverables (such as technical specifications or reports) were deemed significant in that sense, already hinting at a substantial degree of a priori alignment between the activities of SDOs and European regulatory needs in the field of AI.
A final analytical step assessed the level of maturity of the sub-population of significant standards in terms of their degree of coverage of the different requirements of AI regulation, including data governance, technical documentation, record keeping, transparency, human oversight, accuracy, robustness, cybersecurity, risk management and quality management. Multiple attributes were analysed and quantified for each standard, notably the degree to which they operationalise the high-level requirements of the legal text, but also including a range of suitability aspects, such as typology, domain generality, maturity and compliance management,.
A detailed operationalisation and suitability analysis is not reflected in the current indicator, due to its ongoing nature, as standards from only two of the aforementioned SDOs have been assessed to date. Despite this, a significant number of standards from ISO/IEC have already been identified as offering high operationalisation and/or suitability values. This implies that future standardisation activities aiming to produce European and harmonised standards for the upcoming AI regulation should be able to leverage and build on existing work at the international stage.
A final observation supporting the above conclusion is the high level of European participation in international AI standardisation activities. With regard to ISO/IEC, the largest standard development body composed of representatives from national standards organisations, EU Member States participation on Subcommittee 42 (Artificial Intelligence) of the Joint Technical Committee 1 on Information Technology is substantial. Indeed, there are 18 EU national standardisation bodies among its participating or observing members, being 13 of those actively participating, of the current 33 participating members. The significance of European participation in international standardisation activity on AI is further reinforced by the technical cooperation between European SDOs (CEN/CENELEC and ETSI) and ISO/IEC through the Vienna and Frankfurt agreements.