Explainable AI as Collaborative Task Solving

Arjun Akula, Changsong Liu, Sinisa Todorovic, Joyce Chai, Song-Chun Zhu; The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 91-94


We present a new framework for explainable AI systems (XAI) aimed at increasing human trust in the system's performance through explanations. Based on the Theory of Mind, our framework X-ToM explicitly models machine's mind, human's mind as inferred by the machine, as well as machine's mind as inferred by the human. These mental representations are incorporated to (1) learn an optimal explanation policy that takes into account human's perception and beliefs; and (2) quantitatively evaluate human's trust of machine behaviors. We have applied X-ToM in the context of visual recognition. Compared to the most popularly used attribution based explanations (saliency maps), our X-ToM significantly improves human trust in the underlying vision system.

Related Material

author = {Akula, Arjun and Liu, Changsong and Todorovic, Sinisa and Chai, Joyce and Zhu, Song-Chun},
title = {Explainable AI as Collaborative Task Solving},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}