Authors

Simone Kopeinik • Dominik Kowald • Tony Ross-Hellauer

Language

English

Keywords

trustworthy AI, reproducibility, machine learning, barriers, drivers

License

CC BY-SA 4.0 International

Target audience

AI/ML researcher, funders, publishers

Prerequisites

Basic computer science knowledge

Abstract

This learning module explains the critical relationship between reproducibility and the trustworthiness of artificial intelligence (AI). It emphasizes the understanding of how reproducibility impacts credibility and explains the different levels of AI reproducibility that researchers may strive for. The module also elaborates on barriers, such as inconsistent data collection and lack of transparency, as well as drivers, like standardized practices and tools that can support reproducibility. Finally, it illustrates how the presented barriers and drivers interact and aims to foster an understanding of this interaction in order to enhance the reproducibility of AI systems, thereby leading to a more reliable and valuable research practice in the field of AI.

Learning outcomes

1. Understand the relationship between reproducibility and the trustworthiness of AI
2. Understand the relationship between reproducibility and the trustworthiness of AI
3. Be aware of Barriers and Drivers of reproducibility
4. Understand the relation of barriers, drivers, and degree of AI reproducibility