Paper on modeling cognitive load in explainable AI published in ACM CHI 2020

Explanations of artificial explanations can be simplified by controlling the number of variables, but the complexity in how they are visualized can still impede quick interpretation. We quantitatively modeled the cognitive load of machine learning model explanations as visual chunks and introduce Cognitive-GAM (COGAM) to balance between explanation accuracy and cognitive load.

Congratulations to team members Ashraf Abdul and collaborators Christopher von der Werth and Mohan Kankanhalli!

Sparse Linear Models (sLM) Cognitive-GAM (COGAM) Generalized Additive Models (GAM)
Lowest Cognitive Load, but
Lowest Accuracy
Balances Cognitive Load and Accuracy
by increasing accuracy with marginal increase in cognitive load
Highest Cognitive Load, but
Highest Accuracy

NUS Computing Feature:

CHI 2020 Video Preview:

Recorded presentation at 2nd NUS Research Week:

Abdul, A., von der Weth, C., Kankanhalli, M., and Lim, B. Y. 2019. COGAM: Measuring and Moderating Cognitive Load in Machine Learning Model Explanations. In Proceedings of the international Conference on Human Factors in Computing Systems. CHI ’20.