2026 ASEE Annual Conference & Exposition
Designing Assessments for the New AI-era
Presented at DSAI-Session 8: Generative AI in Assessment, Grading, and Accreditation
The rapid integration of generative artificial intelligence (AI) into higher education has disrupted traditional paradigms of teaching, learning, and particularly assessment. As AI systems such as ChatGPT, Copilot, and Gemini increasingly mediate the production of text, code, and data analysis, conventional models of student evaluation, centered on recall, replication, and closed-ended testing, have become increasingly untenable. This paper argues that higher education must transition from an “AI-policing” stance toward the deliberate design of AI-inclusive and AI-tolerant assessments that preserve academic integrity while advancing student learning, scalability, and agency. Drawing on design frameworks from Jisc [1], UNESCO [2], and the U.S. Department of Education [3], as well as case studies from STEM education, the paper proposes a new typology of hybrid assessment. These models integrate asynchronous AI-amenable components (such as written projects, data analyses, and programming assignments) with synchronous or live performance elements (such as oral defenses, debates, hackathons, and simulations). The hybrid format ensures that students can leverage AI as a creative partner while remaining accountable for understanding, reasoning, and improvisation- skills that cannot be outsourced to generative systems. The study further introduces the concept of process-based evaluation, emphasizing the analysis of a learner’s digital footprint - draft versions, reasoning logs, and iteration histories - captured through process-mining tools and learning analytics. The framework operationalizes academic integrity through a Minimum Viable Evidence (MVE) approach, requiring a bounded set of process artifacts, AI-use disclosure, documented revisions, and explicit validation checks that shift evaluation from polished outputs to demonstrable reasoning and professional judgment. This evidence of engagement and cognitive progression provides a richer, more authentic picture of competence than static end-products. Complementing this, the paper advocates for a shift from traditional rubric-based grading, which can be easily gamed by prompting AI tools, to cohort-relative assessment models that benchmark performance against dynamic group patterns rather than fixed descriptors. This approach not only mitigates volatility in grading but also restores fairness by contextualizing achievement within each learning community. Ethical, technical, and logistical implications are critically examined, including concerns about faculty workload, data privacy, and accreditation compliance. The analysis highlights that while hybrid assessments demand greater initial preparation and technological infrastructure, they ultimately foster integrity and efficiency by automating feedback and reducing grading bias. The paper concludes by outlining a roadmap for developing AI-resilient assessment ecosystems: learning environments that embrace generative technologies as collaborators rather than threats, maintain authenticity through live verification, and cultivate reflective, adaptive learners prepared for a digitally augmented world. Accordingly, this work advances a proof-of-work principle for assessment: students are evaluated not only on outputs but on the documented reasoning and iteration that produced them.
-
Andrew D’Amico Northwestern University [biography]
The full paper will be available to logged in and registered conference attendees once the conference starts on June 21, 2026, and to all visitors after the conference ends on June 24, 2026