On the trail of human cognition
Pragmatic AI for less bias
2024/05/29
Researchers at the research group of Artificial Intelligence and Machine Learning (AIML) in the Department of Computer Science are working on the question of how human thinking and learning can be transferred to AI systems. As part of the TU project “Reasonable Artificial Intelligence (RAI)”, which has submitted a full application to the Excellence Strategy of the German federal and state governments for the ‘Clusters of Excellence’ funding line, the comprehensibility of AI decisions in particular is being investigated.
Why does an AI system make a certain decision and how can the decision-making process be explained so that a human understands it? These questions are a focus of research at AIML and are of central importance for the understanding and further development of AI systems. Felix Friedrich is doing his doctorate at AIML and has identified a need for improvements in terms of fairness and bias in AI systems. He was able to show that many AI models, such as image generators, tend to reproduce stereotypical ideas, such as a banker as a man in a suit or the nurse as a woman.
Removing prejudices from AI models completely? Probably an impossible endeavour. Ethical considerations are a central component of Friedrich's research. The researcher therefore appeals to human pragmatism: Risks are to be expected with humans, for example in road traffic. It is therefore wrong to expect machines to never make mistakes. AI systems are heavily reliant on human feedback – and the better we understand how the machine works and where the errors lie, the better we understand the potential of AI systems to improve them.
Part of the Cluster of Excellence "RAI"
The research work at AIML is embedded in the “Reasonable Artificial Intelligence (RAI)” project, which was called for full application in the “Clusters of Excellence” funding line as part of the Excellence Strategy of the German federal and state governments . RAI has set itself the goal of developing and promoting AI technologies that are not only effective and efficient, but also prioritise ethical principles such as safety, fairness and transparency.
Hessian.AI/eml/cst
