Information aus der Blackbox: Zum Verhältnis von Transparenz und Geheimnisschutz am Beispiel Künstlicher Neuronaler Netze

The rapid development and proliferation of Artificial Intelligence (AI) have led to increasing transparency obligations. These obligations encounter two challenges: the protection of trade secrets and the black box nature of AI
Analyzing the General Data Protection Regulation and the draft of the European Union's AI Act, this publication examines how existing and future transparency obligations can be met for trained artificial neural networks (ANNs), while still safeguarding trade secrets.
To achieve this, the author elaborates on the semantic information of ANNs and develops a hierarchical framework that outlines different representation possibilities, starting from machine code and progressing to techniques of Explainable Artificial Intelligence.
The resulting system of graduated transparency resolves the tension between transparency obligations and the protection of trade secrets in the field of ANN.