Research project: Transparency of AI decisions

How do AI models make their decisions – and how can the underlying mechanisms be explained in a way that specialist users can understand? This is the question addressed by the research project ‘eDIF – NoCode Platform Concepts & Features’, in which the Research Centre for AI and Robotics Law is participating.

As part of the Master’s programme in “Applied Artificial Intelligence & Digital Transformation” at Ansbach University of Applied Sciences, the project is investigating how the internal concepts of AI models can be presented in a way that is visible and comprehensible to non-technicians. The focus is on the approach of mechanistic interpretability: this is intended to help understand which learned features and concepts a model uses when making a decision.

The aim of the project is to develop a no-code prototype that makes complex AI interpretability processes accessible without in-depth prior technical knowledge. This will enable subject-matter experts not only to see the result of an AI system, but also to understand the concepts and decision-making structures on which that result is based.

The platform will be tested and evaluated as part of an empirical study. We are particularly looking for students and other interested individuals who are willing to test the prototype from a subject-matter perspective. No technical knowledge is required. Feedback from participants will help to identify relevant use cases and further develop the clarity of the information presented.

The project is being carried out in collaboration with eDIF and is intended to result in a functional prototype and a scientific publication. Further information on eDIF can be found at: https://www.edif.ai/