Experimentation, testing, and auditing are indispensable for developing evidence-based policies, ensuring their compliance, and verifying their efficacy in real-world scenarios. These processes help identify potential issues, mitigate risks, and ensure the ongoing integrity and safety of AI systems. With the rapid evolution of AI systems, they serve as critical elements in creating a secure, trustworthy, and innovative digital ecosystem in the European Union.

Since 2020, TFS has been developing recommendations that promote these techniques, beginning with our response to the European Commission’s call for comments on its White Paper on AI, which outlines its proposed AI policies for the European Union. In our responding policy memo, Experimentation, testing & audit as a cornerstone for trust and excellence, we proposed that the EU should establish strong capabilities for experimentation, testing and auditing of AI systems, through instruments like risk assessment frameworks, benchmarking protocols, and engineering labs. We furthermore recommended designing these capabilities to promote compliance and innovation both within the EU and abroad, making them accessible and evolving them over time.

In October 2021, in providing further input on various aspects of the EU AI Act, we published a second policy memo on this topic: Trust in Excellence & Excellence in Trust. In this memo, we advocated for numerous policies that aim to make trustworthiness the EU’s competitive advantage globally, including well-funded and resourced testing and experimentation facilities, an ambitious, EU-wide and SME-friendly regulatory sandbox program, and the establishment of direct complaint procedure to empower citizens to defend their fundamental rights, health and safety.

We are reassured that many of these considerations, such as regulatory sandboxes, measurement and benchmarking capabilities, and a direct complaint procedure have been reflected in the European Parliament’s negotiating position on the EU AI Act.