Auditing of AI

Artificial Intelligence (AI) is becoming increasingly integrated into our society with applications ranging from facial recognition to self-driving cars. The potential benefits of these systems are vast, as everything we have achieved as civilization stems from intelligence. Recognizing the potential of AI, it is crucial to explore how we can harness its advantages while avoiding drawbacks (Future of Life Institute, 2015). Ensuring that these systems function safely and reliably is of the utmost importance. One way to achieve this is through the emerging field of AI auditing (Mökander et al., 2023, p. 6). As the consensus on AI shifts towards acknowledging its socio-technical character, it becomes essential to develop strategies that address issues like interpretability and various other social, legal, and technological challenges associated with this technology. A practical solution is embracing a Trustworthy Artificial Intelligence (TAI) approach (Thiebes et al., 2020, p. 447). In this thesis, TAI is approached from an AI auditing perspective by examining how AI experts conceptualize ethical, legal, and socio-technical elements and how these affect the realization of TAI. The thesis concludes that concepts that form the basis of the trustworthiness approach to AI systems are not easily operationalized because of their complex interpretability and subjectivity. To understand these multifaceted concepts clearly, we must view them from their social and cultural context.