In our studies and white papers, we describe the latest scientific findings on topics such as digitalisation, artificial intelligence, machine learning and quantum machine learning. True to our approach of application-orientated research, we place particular emphasis on possible applications, examples and guidelines for companies. Our publications are all available free of charge, although some require a short registration process.
A systematic approach to testing AI applications using foundation models
Foundation models have great potential for the economy and society – but how can AI applications be designed to not only be innovative, but also secure and trustworthy? This question is currently on the minds of many experts in politics, research and companies. The new white paper "Developing trustworthy AI applications with foundation models" provides answers to this question and explains how specific risks of foundation models can affect the AI application and presents a risk-based system that can be used to evaluate and ensure the trustworthiness of such AI applications.
The trustworthiness of AI applications has been the subject of recent research and is also addressed in the EU's recently adopted AI Regulation. The currently emerging foundation models in the field of text, speech and image processing offer completely new possibilities for developing AI applications. This white paper shows how the trustworthiness of an AI application developed with foundation models can be evaluated and ensured. For this purpose, the application-specific, risk-based approach for testing and ensuring the trustworthiness of AI applications, as developed in the "AI Assessment Catalog – Guideline for Trustworthy Artificial Intelligence" by Fraunhofer IAIS, is transferred to the context of foundation models. Special consideration is given to the fact that specific risks of foundation models can have an impact on the AI application and must also be taken into account when assessing trustworthiness.
The publication was released as part of the KI.NRW flagship project Zertifizierte KI.
Methods for the certification of Artificial Intelligence
There are many possible applications for Artificial Intelligence. AI systems are increasingly taking over complex work processes and making autonomous decisions. Consequently, high quality and safety requirements as well as strict ethical guidelines are necessary to guarantee the reliability of AI technologies. Therefore, the KI.NRW flagship project “Zertifizierte KI”, led by Ph.D. Maximilian Poretschkin, Head of AI Assurance and Certification at Fraunhofer IAIS, is developing an approach for assessing the trustworthiness of artificial intelligence, taking into account AI test criteria and quality requirements.
The Springer publication “AI and Ethics” was published as part of the AAAI Spring Symposiums 2023. In the chapter “AITA: AI trustworthiness assessment”, Poretschkin et al. describe options for certifying AI systems, including methods for operationalizing ethical values or evaluating AI processes. In the future, the interaction between humans and artificial intelligence should be optimized and controlled in this way.
Guideline for Designing Trustworthy Artificial Intelligence
Artificial intelligence promises to provide data-driven solutions to optimize problems that are difficult to solve using other methods – including in the field of industrial production and logistics. However, engineers are faced with the challenge that predictable and permanently reliable performance is expected when designing industrial systems and processes. These framework conditions must be considered and implemented in the development and operation of AI-based systems.
The new discipline of AI engineering provides a solution to this problem. Its subject is an engineering and systematic approach to the use of AI processes as part of a holistic systems engineering process. The new white paper of the same name from the Fraunhofer Institutes for Intelligent Analysis and Information Systems IAIS and for Optronics, System Technologies and Image Exploitation IOSB describes how AI engineering can be used beneficially in production. “In our projects with the industry, it becomes clear time and again that AI as a technology alone does not solve the problems,” emphasize the two editors Ph.D. Engineer Thomas Usländer from Fraunhofer IOSB and Daniel Schulz from Fraunhofer IAIS. The systematics and methodology of AI engineering are needed to ensure the sustainable business benefits of this innovative technology."
A comparative study
Data protection, risk and compliance management: new AI technologies present companies and developers with new challenges. Various institutions are working on guidelines and standards for the management of artificial intelligence (AI) so that they can tackle these in a systematic and structured way. In the study “Management System Support for Trustworthy Artificial Intelligence”, Fraunhofer IAIS compared the draft standard for AI management systems from the International Organization for Standardization (ISO) and the guidelines to date. The study, commissioned by Microsoft, shows the extent to which AI management systems can support companies in the trustworthy use of AI systems and at the same time strengthen trust in AI applications.
White paper
The team from the Universities of Bonn and Cologne together with Fraunhofer IAIS presents its interdisciplinary approach in a white paper for the certification of AI applications and explains the fields of action from a philosophical, ethical, legal and technological perspective. The publication forms the basis for the further development of AI certification.
Fraunhofer Institute for Intelligent Analysis and Information Systems IAIS