The problem
AI can produce correct answers without real learning.
Generative AI can produce technically plausible outputs without building the reasoning process behind them. That creates cognitive debt: apparent competence, fragile judgment.
There are three paths: ignore it, ban it, or redesign the learning experience. This protocol works on the third path.
Pilot sequence
Five sessions. A sequence that cannot be skipped.
Think alone
The student produces a visible trace of reasoning without AI.
Contrast
A Socratic chatbot pressures through questions, not answers.
Interpret data
New technical evidence appears to test hypotheses.
Decide
The student defends a recommendation under uncertainty.
Transfer
A new industrial case, without scaffolding and without chatbot.
What it measures
It does not only assess the final answer. It observes the trajectory.
Causal complexity
Whether the student identifies mechanisms or only symptoms.
Technical specificity
Whether the student uses measurable variables, values, units and conditions.
Epistemic awareness
Whether the student knows what they know, what they do not know, and what evidence is needed.
Decision under uncertainty
Whether the student decides with explicit risk and defends technical judgment.
Designed instruments
The value is in the complete system, not in an isolated prompt.
Classroom
Session scripts, technical cases, timing, teacher intervention rules, and pre/post chatbot activities.
AI
Four chatbot modes: basic Socratic, deep Socratic, technical verification, and adversarial evaluation.
Evidence
Initial trace, hypotheses, contrast log, defensible decision, final transfer, and teacher observation.
Research
Longitudinal rubric, methodological frame, evidence matrix, and criteria for analyzing cognitive displacement.
Why it matters
"If you assess only the final product, AI solves it. If you assess the trajectory, AI becomes a tool."
This work connects three worlds that are usually treated separately: engineering, artificial intelligence, and pedagogy. The question is not whether students will use AI. The question is whether the design of the class will force them to think better with it.
This public page summarizes the protocol. The complete operational instruments remain curated while the pilot and academic work continue.
Status
Protocol complete. Pilot in progress. Paper in preparation.
Methodological design
Five-session sequence and evidence matrix defined.
Teaching material
Scripts, industrial cases and rubrics built for classroom execution.
AI system
A restricted chatbot designed to ask, contrast and pressure without replacing technical judgment.
Next step
Evidence analysis, protocol iteration and academic preparation.
Conversation
AI does not replace judgment. A good class makes judgment visible.
I work with institutions, teams and communities that want to use artificial intelligence without losing human responsibility.
Let's talk