Designing Human-Centered Assessment in an AI World: A Framework for Making Student Thinking Visible

Synthesis: This conceptual paper argues that the generative-AI assessment debate has been framed around the wrong question. Detection and Academic Integrity enforcement, it contends, do not answer what evidence is needed to make valid judgments about student learning. Building on authentic assessment and evidence-centered assessment, it proposes Human-Centered Assessment — an approach organized around authentic voice, reflection, judgment, context, process transparency and ethical AI integration — and supplies three complementary frameworks that move faculty from principle to assignment redesign. The shift it recommends is from detecting AI to designing assessments that produce richer evidence of thinking, in a world where the artifact alone no longer certifies the learning.

Key Findings

1. The paper argues that generative AI has not changed the purpose of Assessment but has exposed how much current practice relies on the artifact as a proxy for the thinking behind it.
2. It introduces Human-Centered Assessment as a conceptual approach organized around six emphases: authentic voice, reflection, judgment, context, process transparency and ethical AI integration.
3. It presents three complementary frameworks, sequenced from understanding the principles, to evaluating existing assignments, to redesigning assessments for AI-mediated learning environments.
4. Detection-based responses are treated as insufficient because they address authorship rather than validity, leaving the underlying question of what counts as evidence unanswered.
5. The paper builds explicitly on authentic assessment and evidence-centered assessment rather than proposing a new assessment theory.
6. It concludes with implications for assessment practice, faculty development and future research, and is published in the AALHE conference proceedings rather than as an empirical study.

The argument

The paper starts from a question that detection debates tend to skip: what evidence do educators need to make valid judgments about student learning? Its premise is that the purposes of higher education have not changed — students are still expected to think critically, solve problems, communicate and exercise professional judgment — but that generative AI has made the old evidentiary shortcut, where a submitted artifact stood in for the reasoning behind it, unreliable.

On this account the integrity conversation and the validity conversation are different, and the field has spent its energy on the first. That does not make detection pointless, but it makes it incomplete: catching AI-generated work says nothing about whether the assessment ever captured the learning it claimed to measure.

The framework

Human-Centered Assessment is offered as a conceptual approach rather than a measurement model, and its six emphases are the operational core. Authentic voice and reflection push toward artifacts a student can genuinely call their own and can explain; judgment and context tie assessment to the professional situations graduates will face; process transparency makes the route to a product visible; and ethical AI integration accepts AI as part of the environment rather than an intruder to be excluded.

The three frameworks are sequenced for faculty use: one to understand the principles, one to audit an existing assignment against them, and one to redesign for AI-mediated learning. The sequencing is the practical contribution — a department can start from an assignment it already runs rather than from a blank page.

From detection to design

The paper's through-line is a redirection of effort. If the artifact can no longer be trusted as evidence, the response is to design assessments that generate richer evidence of thinking — drafts, reflections, oral defenses, justifications, process documentation — rather than to police the artifact more aggressively.

That is a defensible position, and it is also the harder one to implement, because process evidence costs faculty time and does not scale like a machine-graded exam. The paper acknowledges the workload and faculty-development implications and closes by naming them as the next research questions, which is where the framework's practical limits sit.

What this means for practice

• Instructors. Audit one assignment for what it actually evidences. If the submitted artifact is the only evidence of thinking, the redesign target is process transparency — drafts, rationale and reflection that make the reasoning visible.
• Assessment designers and professionals. Separate the validity question from the integrity question. Detection answers whether AI wrote something; only redesign answers whether the assessment measures the learning it claims to.
• Faculty developers. Use the three-framework sequence as a workshop structure. Starting from an existing assignment and auditing it against the six emphases is more tractable than introducing a new assessment theory.
• Administrators. Budget for the workload the approach implies. Process-rich assessment needs faculty time and support, and a framework that increases marking load without resourcing it will not survive a semester.

Limitations

• The paper is conceptual and published in conference proceedings; it presents a framework and three design tools but no empirical evaluation of whether the approach improves assessment validity.
• It draws on authentic and evidence-centered assessment rather than proposing a new theory, so its contribution is synthesis and operational sequencing rather than a novel construct.
• The six emphases and three frameworks are not accompanied by reliability or validity evidence, and no instrument for applying them is offered.
• The paper is written for a higher-education assessment audience and does not address the resource constraints of large-enrollment or under-resourced settings.

Citation

De La Cruz, L. (2026). Designing Human-Centered Assessment in an AI World: A Framework for Making Student Thinking Visible. Intersection: A Journal at the Intersection of Assessment and Learning, 7(4), 42-53.