What should assessment measure when AI can produce the output? A conceptual framework for human-centered assessment in online education

Abstract

Background: Generative artificial intelligence (AI) systems can now produce written texts, solve complex problems, and create academic artifacts previously treated as reliable indicators of student learning. This capability undermines the construct validity of output-focused assessment, a challenge especially acute in online and distance learning (ODL) environments where AI agents can complete assessments on students' behalf. Purpose: This paper identifies the distinctly human capacities that educational assessment should prioritize and proposes conceptual principles for an assessment paradigm adequate to cultivating and evaluating those capacities across educational levels and delivery modes. Method: A systematic conceptual review was conducted across ERIC, PsycINFO, Scopus, and Web of Science (November 2022 to March 2026). From 874 initial records, 90 sources (64 journal articles, 17 monographs, 9 policy reports) were selected through a PRISMA-adapted screening process and synthesized using an integrative conceptual approach. Findings: The review identifies four distinctly human capacities resistant to AI replication (contextual judgment, lived-experience originality, dialogic meaning-making, and reflective action under uncertainty) and derives a four-principle assessment framework: processuality, dialogicity, situatedness, and reflexivity. These principles are organized along temporal/contextual and individual/relational axes. Implications: The framework offers assessment designers, particularly in online and technology-enhanced learning contexts, principled foundations for creating tasks that evidence genuine human learning rather than automatable output production. Implications for educational policy, teacher education, equity, and future empirical research are discussed.