Can we use AI to explain its reasoning so that humans can systematically learn from it? Imagine new physicians being taught to detect cancer on X-ray images by the AI system pointing out suspicious spots. Or AI providing second opinions even to advanced radiologists. This opportunity would enable us to use AI to learn even when existing knowledge is either not explicit or when teaching by rare experts does not scale.

This is the topic (Dr.) Philipp Spitzer has done a remarkable Ph.D. thesis on that he successfully defended this afternoon: “The Impact of Explainable AI on Knowledge in Human-AI Interactions” – with a number of experiments where novices learn to interpret medical X-rays, identify architectural building styles or distinguish birds.

Congratulations to Philipp for the great work that has received reception at renowned outlets. It should provide an important milestone for a successful academic career and promise novel applications for the real world.
 

A big thank you also the defense committee at Karlsruhe Institute of Technology (KIT) with Sanja Lazarova-Molnar, Oliver Grothe and Emilia Graß!