Paper Number
1968
Paper Type
Complete
Abstract
Risk-based artificial intelligence (AI) regulations define risk categories for AI-enabled systems. The operators of such systems must determine the risk category applicable to their AI systems. This requires detailed knowledge of the classification rules defined in the regulations. Only a few supporting tools have been developed to facilitate the task of risk classification. This paper presents a novel method that describes all the necessary steps to develop such a tool. To demonstrate and evaluate the method, it is instantiated for the European Union’s AI Act. The evaluation shows i) that the classification model achieves promising performance in predicting the risk categories for AI systems, ii) that users can effectively use the web application to carry out a risk classification, and iii) that users find SHAP text plots integrated into the web application helpful for understanding the reasons of a classification prediction.
Recommended Citation
Weinzierl, Sven; Zilker, Sandra; Zschech, Patrick; Kraus, Mathias; Leibelt, Tobias; and Matzner, Martin, "How Risky is my AI System? A Method for Transparent Classification of AI System Descriptions by Regulated AI Risk Categories" (2024). ICIS 2024 Proceedings. 4.
https://aisel.aisnet.org/icis2024/data_soc/data_soc/4
How Risky is my AI System? A Method for Transparent Classification of AI System Descriptions by Regulated AI Risk Categories
Risk-based artificial intelligence (AI) regulations define risk categories for AI-enabled systems. The operators of such systems must determine the risk category applicable to their AI systems. This requires detailed knowledge of the classification rules defined in the regulations. Only a few supporting tools have been developed to facilitate the task of risk classification. This paper presents a novel method that describes all the necessary steps to develop such a tool. To demonstrate and evaluate the method, it is instantiated for the European Union’s AI Act. The evaluation shows i) that the classification model achieves promising performance in predicting the risk categories for AI systems, ii) that users can effectively use the web application to carry out a risk classification, and iii) that users find SHAP text plots integrated into the web application helpful for understanding the reasons of a classification prediction.
When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.
Comments
13-DataAnalytics