Autoren: Westhoven, Vock, Adolph
Kategorie: Technikfolgen und –abschätzung
Advances in the field of Artificial Intelligence (AI) generate pressure to apply the technology in more and more areas. As such, numerous political and regulatory as-pects are in need of revision, highlighted by efforts such as the European Commis-sion’s White Paper on Artificial Intelligence or the national AI-Strategy of the German Federal Government.
One important aspect in the professional area is the appropriate consideration of occupational safety and health (OSH) when deploying of AI. With our work, we provide an overview over current research activities in Germany and future research topics.
We analyzed 600 AI projects compiled by the “Plattform Lernende Systeme” accord-ing to their relevance for safety and health. Most projects focus on relatively direct sensor data processing, with some also providing more far-reaching predictions or inclusion of more abstract models. Few projects address AI safety itself, with verifica-tion or automated safety configurations. The remaining projects study various other topics, such as a documentation assistance.
The addressed characteristics are of limited scope when aiming for a comprehensive view of safe AI. We thus compiled further theoretical considerations on AI safety from literature, loosely based on work on autonomous agents by Franklin and Graesser (1996) and integrating more current work, e.g. regarding interactive machine learning systems (Dellermann et al., 2019) and regarding a clearer distinction between transparency and interpretability (Ventocilla et al., 2018). We distinguish technology-inherent aspects of AI from those only becoming effective in certain contexts. Inherent factors mostly influence the error probability and include transparency, robustness, and verifiability, whereas context factors mostly affect the damage potential and include human involvement and the environment.
Our project analysis highlights current work in Germany towards safe AI. We complement these with considerations drawn from literature regarding the safety-relevance of AI systems to provide a comprehensive overview over current research and future research opportunities at the intersection of AI and OSH.
Zeitfenster der Veranstaltung (1)
Westhoven, Vock, Adolph