Any essay, problem set, lab report, or case study that AI can pass — QLM converts it into a simulation that captures how students think, not just what they produce.
Why this matters now: A 2026 study in Science by researchers at Berkeley, UTS, and Cornell found broad GenAI use among students and an estimated 9% rate of AI-assisted cheating. The authors call for discipline-specific assessment reform rather than blanket bans or universal detection. QLM is that reform — converting AI-vulnerable assignments into process-based simulations that make cheating irrelevant.
Chirikov, I., Smirnov, I., & Kizilcec, R. F. (2026). Generative AI use and misuse call for assessment reform in higher education. Science, 392(6800), 818–820. DOI: 10.1126/science.aec5115
AI detection tools produce false positives on ESL students and false negatives on sophisticated users. The Berkeley study found detection is unreliable at scale.
Blanket bans are unenforceable and counterproductive. Students use AI anyway; bans just make the use invisible.
Honor codes alone didn't prevent the 9% assisted-cheating rate the researchers found.
Discipline-specific reform — redesign assessments so AI cannot substitute for the reasoning they measure.
Process-based evidence — evaluate how students think, not just what they submit.
Responsible AI integration — measure whether students used AI well, not just whether they used it.
"Rather than treating GenAI use as a binary (cheating or not), institutions should redesign assessments to capture the reasoning process and make AI substitution irrelevant."
— Adapted from Chirikov, Smirnov, & Kizilcec (2026), Science 392(6800)
QLM's Assessment Conversion Studio is the direct implementation of this recommendation. We don't detect AI. We don't ban AI. We redesign the assessment so that the simulation is the assessment and the reasoning process is the evidence.
Six dimensions of process evidence — not just whether the answer is right.
Every score is backed by behavioral evidence from the simulation — predictions made, decisions taken, explanations written, revisions after feedback.
Every step depends on the previous one. Students can't skip ahead or delegate to AI — because step 3 requires the actual output of step 2.
Empirically validated: interconnected assessments correlate r=0.954 with true competency (arXiv:2512.10758)
See every prediction, decision, revision, and strategy shift as it happened. The timeline IS the evidence — not the final submission.
Based on the Learning Visibility Framework (Davalos & Zhang, 2026)
AI-generated follow-up questions probe each student's understanding. Three independent evaluators score responses. Higher concordance than human graders.
Validated by Barba & Stegner (2026) and Ipeirotis & Rizakos (2026)
Students don't know they're being tested. The simulation IS the assessment. Behavioral patterns predict competency without explicit test questions.
Built on stealth assessment research (Shute, 2011–2025; JRTE 2025)
Every assessment offers multiple paths to demonstrate the same competency. Visual, written, interactive, or oral — students choose their strongest modality. Same rubric. Same rigor.
Implements UDL framework for simulation-based assessment (Advances in Simulation, 2025)
Upload or describe any existing assessment — essay prompt, problem set, lab procedure, case study, coding spec.
The system identifies learning objectives, maps to Bloom's taxonomy, and shows exactly which parts AI can substitute.
Your assignment becomes a scenario with decision points, branching consequences, and process-evidence capture — matched to the best simulation template.
Preview the simulation yourself. Edit decision points, adjust rubric weights, set AI-use policy. Approve when ready.
Every prediction, decision, explanation, and revision is captured. You see misconception clusters, mastery confidence, and oral defense probes — not just grades.
QLM identifies the skills and competencies your assignment was designed to measure — tagged by Bloom's taxonomy level.
See exactly which parts of your assignment AI can pass without understanding. Severity-rated by component.
A runnable scenario with decision points, branching consequences, and variables that differ between students.
Your old rubric mapped to 6 process dimensions: prediction, decision, explanation, revision, AI-use, and transfer.
Follow-up questions that verify authenticity and probe reasoning depth. Targeted to each student's evidence gaps.
See process evidence, misconception clusters, mastery confidence, and AI transparency for every student.
Paste one assignment. See the simulation before your students do. 2 minutes.
Convert Free →Share this tool with your department. It's free to try.
"I found a tool that converts AI-vulnerable assignments into simulation-based assessments. Free to try: quantumlearningmachines.com/conversion-studio"