Explainable AI Design & Human-AI Interaction

Course Description

This is a design course. Increasing the trustworthiness of machine learning solutions has emerged as an important research area. One approach to trustworthy machine learning is explainable and/or interpretable machine learning, which attempts to reveal the working mechanisms of a machine learning system. However, other than on-task performance, explainability is not a property of machine learning algorithms, but an affordance: a relationship between explanation model and the target users in their context. Successful development of machine learning solutions that afford explainability thus requires understanding of techniques beyond pure machine learning. In this course, we will first review the basics of machine learning and human-centered design. Then, we will introduce several techniques to explain machine learning models and/or make them interpretable, and through hands-on sessions and case studies, will investigate how these techniques affect human-AI interaction.In addition to individual homework assignments, students will work in groups to design, justify, implement, and test an explainable machine learning algorithm for a problem of their choosing.Additional Recommended (601.454/654, 601.290, 601.490/690 or 601.491/691) and 601.477/677. Prerequisite(s): Students may receive credit for EN.601.484 or EN.601.684, but not both.

Code

EN.601.684

Credits

3

Subject

Computer Science