On June 23, 2022, the partners of the KI Absicherung consortium gathered for the official project closing event in Berlin and presented the results of their three years of research and development work.
Project coordinator Stephan Scholz (Volkswagen) opened the event for the approx.120 visitors on site as well as the approx. 200 participants connected via livestream and led through the day-long program. After another brief welcome by Ernst Stöckl-Pukall from the Federal Ministry of Economic Affairs and Climate Action, the project's funding body, a keynote address was given by Wolfgang Müller-Pietralla from Volkswagen on the topic of "Guiding principles for the responsible use of AI". Afterwards, the two coordinators Stephan Scholz (Volkswagen) and Michael Mock (Fraunhofer IAIS) as well as the subproject leaders presented their most important project results in short speeches.
The goal of the project was to build an industry consensus regarding a general assurance strategy for AI functions. For this purpose, specific methods, measures, and metrics have been developed in a systematic way to determine and reduce the shortcomings of the different AI functions. These mechanisms serve as the basis for reliable system assurance. Depending on the AI function module, either architecture-based, deep neural networks (DNN)-based, test-based, or data-based metrics provide the assessment-critical data. For demonstrable security reasoning, a reliable process for holistic assessment of the various AI functions has been developed.
This chain of argumentation was built up and illustrated in KI Absicherung using the example of pedestrian detection. Existing algorithms for pedestrian detection have been tested and corresponding critical scenarios (corner cases) were identified. Targeted training of the AI with modern simulation technologies improved pedestrian detection even under difficult conditions. For example, the AI correctly and reliably identified objects even when lighting conditions changed or when the view of the pedestrian was temporarily obscured.
In the afternoon, visitors then had the opportunity to gain in-depth insights into the contents of the project through an interactive poster session with more than 40 posters as well as three parallel lectures. In addition, the topic of standardization was discussed in the afternoon - on the one hand from an ISO perspective, and on the other hand from a DIN/BSI perspective. At the end of the event, a panel discussion by all four coordinators of the KI family projects elaborated on the positioning of the KI Absicherung project in the overall context of the project family. This also set the connecting point for the follow-up event on the next day: On June 24, 2022, the consortium of the sister project KI Data Tooling presented its interim results.
Further information, such as the poster booklet of the closing event for download, can be found here: https://www.ki-absicherung-projekt.de/final-event-results