AI in biology education assessment: How automation can drive educational transformation

Abstract

In this chapter I review a decade of research on the artificial intelligence (AI)-based EvoGrader assessment system and its associated ACORNS instrument to demonstrate the complex consequences of developing technological tools designed to replicate human tasks (in this case, computer scoring of biological explanations). Many of the examples highlight the challenge of cleanly disentangling automation efforts from the educational systems in which they operate. AI scoring performance is impacted by instructor discourse practices, disciplinary language, writing fatigue, implementation conditions, English learner attributes, and many other variables. Consequently, efforts to improve automated scoring may (intentionally or unintentionally) reveal crucial insights into the educational contexts in which AI-based assessments operate. These findings suggest that future AI-assessment development work should be grounded in a "systems" perspective that explicitly acknowledges and integrates features of the broader educational ecosystem and considers benefits of the development process beyond the assessments themselves.