Authentic assessment
Case study · AI Tools & Strategy · Learning Design
The part of the method that answers the academic-integrity problem head on. Not by policing what students use, but by changing what they are asked to do, so the graded object is something a generative model cannot produce on a student’s behalf.
The problem
A written or selected-response exam exists to certify that a student can do the work of a discipline. A remote student with a text field and a generative model can now satisfy that instrument without demonstrating the underlying competence, so it no longer licenses the inference it was built to license. Intensified proctoring treats the symptom. The durable remedy is to change the construct being assessed, so that passing the task is demonstrating the competence, whether or not a model is in the room.
That is what authentic assessment does. It asks the student to perform a real task of the field and to account for their own reasoning, and it grades the process and the judgment, not a producible final answer.
The test
The question the skill and the checker ask of every graded task
Could a student pass this with a generative model, without having done the thinking? If the honest answer is yes, the task is redesigned until the answer is no.
Authentic assessment is one line on the build skill’s quality checklist, and it is the same line the Dial Your Course checker flags on an existing course. Both look for the same failure: a graded task whose answer a model can supply directly. The fix is never a warning in the syllabus; it is a change to the task, toward one of the forms below.
Approaches
Authentic assessment is a family of approaches, not a single technique. Simulations are one high-fidelity member of it, not the whole idea. What they share is that the graded object is the student’s own reasoning over their own work.
- Process and reasoning artifacts. The research log, the rationale for each decision, the iterations, and the response to critique are submitted and weighted, not just the finished file.A model can generate a polished result; it cannot reconstruct the specific path a particular student took to get there.
- Portfolios and defended case studies. Work assembled over a term into one coherent narrative, presented and explained.The evidence is the accumulation and the through-line, which only the student who built it can account for.
- Oral defense and live questioning. The student answers unscripted, decision-relevant questions about their own work in real time.Live reasoning cannot be outsourced; the student either understands their choices or does not.
- Personalized, private-artifact tasks. The task is anchored to material unique to the student, their own brand, dataset, context, or a per-student seeded problem.There is no answer key in the public corpus, so a general model has nothing to copy from.
- Simulations of real professional tasks. A replica of genuine work, complete with the conditions and incidents the real task contains, scored on judgment rather than a producible answer.The graded object is live decision-making inside a scenario built for this learner, which a model cannot fabricate on their behalf.
In the courses
Each course chooses the forms that fit its discipline; the specifics live with the course, not here.
- Advanced Data Science leans on simulations: eight professional tasks with per-student seeded data and a live stakeholder defense. See the course.
- UX Design for Interactive Media is built on a defended case study anchored to the student’s own brand, carried across the term. See the course.
- Light & Lasers assesses hands-on competence in required in-person lab demonstrations, authentic and AI-proof by construction. See the course.
Grounding
The approach rests on established work: authentic and performance-based assessment (Wiggins), constructive alignment between outcomes and evidence (Biggs), and the principle of organizing assessment around the individual learner drawn from personal learning environments, the focus of my own M.Ed. Personalized, performance-based assessment carries that principle into how competence is measured, which is also what makes it resistant to generative substitution.