Towards AI Transparency and Accountability: A Global Framework for Exchanging Information on AI Systems

TL;DR

The paper tackles the fragmentation of global AI regulation by proposing a federated, open standard for exchanging information on AI systems. It outlines a Global AI Information Framework centered on a Data Model, federated Identifiers, AI Offices and Registers, a Catalog of Standardized Measures, AI Cards, and Automated Assessment Programs, all overseen by a Global AI Technical Foundation (GAITF). Key contributions include a decoupled technical-regulatory architecture, standardized measures with public AI Cards, and automated/supervised assessment pathways that align with frameworks like the EU AI Act while accommodating diverse jurisdictions. The framework aims to reduce regulatory burden, enable meaningful public comparisons, and promote responsible innovation through transparent, interoperable information sharing among regulators, industry, and the public.

Abstract

We propose that future AI transparency and accountability regulations are based on an open global standard for exchanging information about AI systems, which allows co-existence of potentially conflicting local regulations. Then, we discuss key components of a lightweight and effective AI transparency and/or accountability regulation. To prevent overregulation, the proposed approach encourages collaboration between regulators and industry to create a scalable and cost-efficient mutually beneficial solution. This includes using automated assessments and benchmarks with results transparently communicated through AI cards in an open AI register to facilitate meaningful public comparisons of competing AI systems. Such AI cards should report standardized measures tailored to the specific high-risk applications of AI systems and could be used for conformity assessments under AI transparency and accountability policies such as the European Union's AI Act.

Figures (2)

• Figure 1: Example entries in three hypothetical AI registers for AI systems whose intended use is hiring automation. The listed standardized measures are exemplary -- respective AI Offices shall determine appropriate Measures (possibly specific to intended use of AI systems). The same system may be in multiple registers, sharing System IDs but complying with local requirements, e.g., regarding the disclosed Measures. Here, comparisons of input influence Measures in Register 1 suggest that System A764 is directly discriminatory with respect to race, while System B766 is indirectly discriminatory via a proxy of race (Zip code).
• Figure 2: Global Framework for Exchanging Information on AI Systems. A Global AI Technical Foundation (1) creates an extensible AI data model (2). Governmental policy-makers in an AI Office create jurisdiction specific policies (3), Measure Catalog (4) and official Assessment Programs (5). They enhance the data model with jurisdictional taxonomies (6). AI Operators (7) comply with government regulations to register key reference data in AI Register (8). AI systems (9) are evaluated via an AI Office's Assessment Programs (5) and the resulting computed Measures (10) are stored in an AI Register. Industry Backed Assessment Programs (11) can mirror AI Offices' Measure Catalog and AI Offices' Assessment Programs as a baseline to produce new measures and assessments. The public (13) can make informed comparisons and decisions on their direct and indirect use of AI.