A Human-Centered Agenda for Intelligible Machine Learning

Jennifer Wortman Vaughan, Hanna WallachThe MIT Press eBooks 2021 · DOI · OpenAlex

Abstract

To build machine learning systems that are reliable, trustworthy, and fair, we must be able to provide relevant stakeholders with an understanding of how these systems work. Yet what makes a system “intelligible” is difficult to pin down. Intelligibility is a fundamentally human-centered concept that lacks a one-size-fits-all solution. Although many intelligibility techniques have been proposed in the machine learning literature, there are many more open questions about how best to provide stakeholders with the information they need to achieve their desired goals. In this chapter, we begin with an overview of the intelligible machine learning landscape and give several examples of the diverse ways in which needs for intelligibility can arise. We provide an overview of the techniques for achieving intelligibility that have been proposed in the machine learning literature. We discuss the importance of taking a human-centered strategy when designing intelligibility techniques or when verifying that these techniques achieve their intended goals. We also argue that the notion of intelligibility should be expanded beyond machine learning models to other components of machine learning systems, such as datasets and performance metrics. Finally, we emphasize the necessity of tight integration between the machine learning and human–computer interaction communities.
BibTeX
@inproceedings{vaughan2021human,
  author = {Jennifer Wortman Vaughan and Hanna Wallach},
  title = {{A Human-Centered Agenda for Intelligible Machine Learning}},
  booktitle = {The MIT Press eBooks},
  year = {2021},
  doi = {10.7551/mitpress/12186.003.0014},
}

Loading related content...

Figure