The many terms for AI that people can understand and control

Scrutable AIED

Abstract

This chapter describes scrutability in AIED. Scrutable systems are designed so that learners can answer key questions about the way those systems work. This is becoming increasingly important as AIED applications become more pervasive, and their impact more extensive. We show how scrutability fits in the landscape of many similar terms; it reflects a human-centred perspective and describes the effort a learner needs to invest to understand the way an advanced educational system works, even for selected aspects they care about. We explain how scrutability should be integrated into the whole learning engineering enterprise. This includes the well studied area of scrutability of learner models and the link with open learner models (OLMs). But it goes beyond that. It involves a scrutability design mindset for all elements in the architecture of AIED systems. We illustrate these ideas in two case studies. A key con- tribution is a set of benchmark questions − a scrutable AIED system should be designed so that a learner can answer at least some of these. These questions can guide the design and test-driven development of AIED systems. We discuss why scrutability is particularly important in education and how AIED makes an ideal context for scrutability research. We conclude with an agenda for future research on scrutability in AIED.

Publication
In Handbook of Artificial Intellidence in Education
Date