Interpretable AI

Image 1 of 1
Theme leaders Mattias Wahde and Karinne Ramirez-Amaro present the theme.

Interpretable AI is an emerging field, focused on developing AI systems that are transparent and understandable to humans.

Graphic element

Many currently popular AI systems operate as black boxes, meaning that it is hard to understand how they make their decisions and to correct errors that inevitably occur from time to time. In many situations, black box AI systems have also been shown to maintain or even enhance biases present in their training data. By contrast, interpretable AI models are designed in a transparent manner, making it possible for a human observer to follow (and, if needed, correct) their decision-making processes.

This is especially important in applications where decisions involve high stakes or legal implications, such as healthcare, personal finance, or traffic applications. Interpretable AI models can also help increase trust in AI systems, making it easier for humans to work alongside them.

The CHAIR theme Interpretable AI executes a series of activities, including seminars, workshops, and research projects, that aim to develop, extend, study, and compare interpretable methods and to contrast them with black box models. The projects cover a range of applications, such as conversational AI, forestry management, and multi-vehicle trajectory planning, and involve collaborations with researchers from various fields.