Engineering Principles for Building Trusted Human-AI Systems

In the process engineering reliable and trustworthy AI systems there is significant wisdom to be gained from traditional engineering domains. Extending on earlier work our attention is on topics that stress the principles of building human-AI systems. We plea for a reinforced attention for engineering methods and processes in order to urge the essence for improved scientific progress and industrial AI applications where one can stand on the shoulders of giants. On the one hand, we see their complexity increase on an individual level, as well as on their connected dependency levels, whilst on the other hand, we see a growing lack of experience on the level of their design and engineering. The complexity of current AI models often limits our understanding. The methods and processes to ensure safety, reliability, and transparency are insufficient. This poses serious risks at the level of trustworthiness, particularly when it comes to critical applications with significant social, economic or even physical impact. Future AI systems must adhere to stringent requirements, as mandated, for instance, by the European AI Act, ensuring meticulous design, validation, and certification based on clearly defined criteria.