# How to Assess Trustworthy AI in Practice

**Authors:** Roberto V. Zicari, Julia Amann, Frédérick Bruneault, Megan Coffee, Boris Düdder, Eleanore Hickman, Alessio Gallucci, Thomas Krendl Gilbert, Thilo Hagendorff, Irmhild van Halem, Elisabeth Hildt, Georgios Kararigas, Pedro Kringen, Vince I. Madai, Emilie Wiinblad Mathez, Jesmin Jahan Tithi, Dennis Vetter, Magnus Westerlund, Renee Wurth  
**arXiv:** [2206.09887](https://arxiv.org/abs/2206.09887)  
**Date:** 2022-06-20  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2206.09887v1/

## Key points

- The report discusses the Z-Inspection® initiative, which evaluates the trustworthiness of AI-based technologies at different stages of their life-cycle.
- The focus is on identifying and discussing ethical issues and tensions using EU HLEG guidelines for trustworthy AI.
- The initiative involves a team of experts with diverse backgrounds who create socio-technical scenarios to identify potential issues related to trustworthiness.
- Ethical tensions that arise during the assessment process are classified into categories using a consensus process based on mapping.
- Recommendations are provided for monitoring AI over time and conducting fundamental rights assessments as part of an overall trustworthy AI assessment throughout the life-cycle of an AI system.
- Shortcomings in this approach may include challenges with data availability or limitations in expertise, but it offers practical suggestions for ensuring rigorous trustworthy AI assessments.

## Summary

This report provides a comprehensive methodological reflection on the Z-Inspection® initiative, which is a holistic process used to evaluate the trustworthiness of AI-based technologies at different stages of the AI lifecycle. The focus is on identifying and discussing ethical issues and tensions through the elaboration of socio-technical scenarios, using the general European Union's High-Level Expert Group's (EU HLEG) guidelines for trustworthy AI. The report illustrates how these guidelines can be applied in practice for both AI researchers and practitioners. The Z-Inspection® initiative involves a team of experts chosen based on their diverse backgrounds, skills, and experiences. They define the boundaries of the assessment and develop an evidence base by creating socio-technical scenarios that identify potential issues related to trustworthiness. The team manages different viewpoints and uses free text with an open vocabulary to describe these issues while building concepts in practice. The initiative also identifies value conflicts and trade-offs through ethical tensions that arise during the assessment process. A consensus process based on mapping is used to classify these tensions into categories, which are then consolidated using various mapping strategies. This consolidation can be challenging but can be aided by natural language processing techniques. Finally, the report provides recommendations for monitoring AI over time and conducting fundamental rights assessments as part of an overall trustworthy AI assessment throughout the life-cycle of an AI system. While there may be some shortcomings in this approach such as challenges with data availability or limitations in expertise, it offers practical suggestions for ensuring rigorous trustworthy AI assessments. Overall, this report serves as a valuable resource for those seeking to evaluate the trustworthiness of AI-based technologies using a holistic approach that considers ethical issues and tensions throughout their life-cycle.