KIT Department of Informatics

Making AI manageable for engineers

Making AI manageable for engineers

Reliable and secure AI systems for engineers are missing - Karlsruhe Competence Center offers assistance for companies

Artificial intelligence (AI) is embedded in smartphones, search engines, or navigation devices and thus facilitates everyday life. There is also great potential for their use in engineering, as in smart factories or autonomous vehicles. However, there is a lack of procedures that make it possible to plan the behaviour of the systems and make their decisions comprehensible. The "Competence Center for AI Engineering" (CC-KING) under the leadership of the Fraunhofer Institute for Optronics, Systems Engineering and Image Exploitation IOSB with the participation of the Karlsruhe Institute of Technology (KIT) and the FZI Research Center for Information Technology is intended to remedy this situation.

Classical engineering is characterized by plannability: Already in the design phase, developers know how the individual components and thus the overall system will later behave. Systems with components of artificial intelligence (AI) or machine learning (ML) are not so predictable; they develop further during their runtime and only unfold their final functionality during operation. This is a great challenge for the reliable control of exceptional situations - and the economic benefit is hardly quantifiable in advance. Without the calculability of classical engineering, the use of intelligent systems is therefore difficult for companies.

The "Competence Center for AI Engineering" (CC-KING) combines the concentrated information technology and engineering competence of the Karlsruhe location in order to decisively facilitate the use of AI in practice: The Fraunhofer Institute for Optronics, Systems Technology, and Image Exploitation IOSB, the FZI Research Center for Informatics, and the KIT conduct research in close contact with companies on fundamental questions, practical methods, and concrete application problems.

Basic methodological questions

The (in)predictability of the behaviour of learning systems is a central theme of AI engineering. "AI engineering aims to make AI and ML usable in engineering terms, comparable to classical engineering. It is a very young discipline that bridges the gap between basic AI research and engineering sciences," says Professor Jürgen Beyerer, scientific director of the Competence Center, head of the Fraunhofer IOSB institute, and professor of computer science at KIT. "Besides predictability, the researchers focus on the security of AI-based systems, the explainability of decisions, or the integration of preliminary and expert knowledge in data-driven approaches. The aim is to develop a standard procedure model for AI engineering that makes AI technologies suitable for large and heterogeneous teams.

"As a technology region with a long tradition in both engineering sciences and computer science, the Karlsruhe location offers optimal conditions for the competence center," emphasizes Beyerer. With the Baden-Württemberg Autonomous Driving Test Field and the Karlsruhe Research Factory, which is currently under construction, there are also suitable real laboratories for the application fields of mobility and industrial production. "Under these conditions, AI engineering could become the unique selling point of German AI."

Consulting and Learning Laboratory for SMEs

CC-KING is intended to enable small and medium-sized enterprises (SMEs) in particular to make controllable use of AI components. "Even highly innovative SMEs often lack AI competence. This gap is difficult to close because AI experts are rare and, moreover, are usually not familiar with the typical application domains," says Beyerer. CC-KING therefore offers companies concrete support. Companies can, for example, make use of QuickChecks or TransferChecks in a very unbureaucratic way. A consulting centre and an AI engineering learning laboratory for the training of company employees are currently being set up.

Interested parties are already invited to contact the CC-KING coordination office located at the Fraunhofer IOSB (under kompetenzzentrum∂ki-engineering eu or by phone at the project assistance under 0721/6091-290). In August 2020, the Baden-Württemberg Ministry of Economics, Labour and Housing approved three million euros in funding for CC-KING.

The contributions of the participating research institutions

As the leading consortium partner, Fraunhofer IOSB contributes its broad information technology competence in industrial automation and control technology as well as in the fields of AI and ML to the competence center. "In particular, we have already developed a tool-supported process model for AI engineering in industrial production over the past three years as part of Fraunhofer's internal lead project 'ML4P - Machine Learning for Production'," explains Dr. Julius Pfrommer, research group leader at the institute and technical-scientific director of CC-KING, "It allows us to apply AI processes in a plannable and repeatable manner. The AI algorithms are of central importance, but often only make up a fraction of the overall solution". One focus is on the deep integration of existing tools from the engineering disciplines with the AI procedures. "Only in this way is it possible for AI to do a good job even in those areas where it has little or no data and experience from the past".

As an ideal testing environment for the use of AI in industrial production, the Fraunhofer-Gesellschaft and KIT are currently building the Karlsruhe Research Factory, which will start operating in 2021.

"There are challenges with AI- or ML-based systems, for example in terms of plausibility and flexibility," says Michael Beigl, Professor for Pervasive Computing at the KIT Faculty of Informatics. The traceability of decisions of AI systems must also be improved, says the smart data expert who coordinates KIT activities within CC-KING. Another research topic is the integration of AI procedures and AI systems such as the Smart Data Innovation Lab (SDIL) with existing models, simulators, and expert knowledge from the engineering disciplines. "For this purpose, we at KIT develop methodological foundations and problem solutions," says Beigl. In addition, the KIT institutes contribute to the field of tools and components.

This includes, for example, the procedure model in AI engineering, assistance functions for knowledge acquisition and optimisation of AI components or the application of AI and ML procedures with limited resources. The FZI Forschungszentrum Informatik is in charge of this work package and the application domain "mobility" of the competence centre. In addition, the FZI, as an institution for practical knowledge and technology transfer, contributes its expertise in the field of mobility research and artificial intelligence, especially on embedded AI as well as AI methods. For research and demonstration of the AI methods to be implemented in the competence centre, both the infrastructures of the Autonomous Driving Test Field Baden-Württemberg and the infrastructures of the FZI House of Living Labs can be used.

Details on the KIT Center Information - Systems - Technologies: http://www.kcist.kit.edu