Teaching AI Well · Facilitator Library
How do you teach the hardest things about AI?
The hardest things to
teach about AI.
Ten lessons for educators delivering AI instruction. Not what AI is; that's the research. This is how to teach it: the misconceptions learners arrive with, the demonstrations that break them, and the pitfalls of teaching each one.
Every lesson is rooted in CPAI's six domains of AI proficiency, the same framework behind the Six Risks, and cross-links to the risk it teaches against.
Conceptual distortions: what AI is, who's responsible, when to engage.
The Dangers of Anthropomorphizing AI
“It understands.” “It knows.” “It wants.” The words install the overtrust.
Why person-language drives overtrust, and how to teach mechanism in its place.
Bias That Compounds
“Bias in, bias out” understates it. The real story has an exponent.
How AI amplifies bias and re-ingests its own output — feedback loops, model collapse, bias laundering.
The Metadata Economy
The risk isn’t only what they type. It’s the exhaust around it.
What AI companies and data brokers aggregate beyond the prompt — and why it can de-anonymize.
The Disclosure Problem
There’s no settled norm for disclosing AI use. You have to model one anyway.
Honest practice, attribution, and integrity when the rules don’t exist yet.
Cognitive Offloading to AI
Calculators offloaded math. AI offloads reasoning — and that’s different.
Which mental muscles atrophy under AI use, and how to design use that preserves them.
Knowing When Not to Use AI
The master skill isn’t fluency. It’s judgment about task-fit.
Teaching abstention — the deference decision, and the capability that doesn’t expire.
Practice deformations: how repeated AI use reshapes thinking and relating.
Prompting Is Dead
Prompt tricks were scaffolding for dumber models. Communication is the skill.
Why intent, context, and clear thinking transfer when prompt formulas don’t.
Attachment & AI Companionship
Parasocial bonding with AI is real — and heaviest for the most vulnerable.
Recognizing emotional dependence on AI, especially among at-risk learners.
Verification Is the New Literacy
The skill isn’t using AI. It’s checking it — with no source to check against.
Reframing hallucination from “AI makes mistakes” to verification as a teachable discipline.
Regression to the Mean
AI pulls every output toward the average. At scale, that erodes voice.
Teaching learners to feel the pull toward generic output — and keep their own signal.
Deliberately deferred. Orchestration (managing AI across tasks and tools) is the most workflow-oriented domain, and the craft guidance for teaching it lives in the Applied AI frameworks: DANCE, and Use Cases → Habits → Workflows. A trainer teaching Brain Fry in week 3 goes there. A dedicated facilitator lesson here is a candidate for a later addition.
The AI Proficiency model
Proficiency is the journey. The Applied AI Certification is designed for applied AI fluency — and advances you toward it.
One framework runs through all of CPAI's AI work: six questions everyone using AI should be asking, across two tiers, Literacy and Fluency. The pieces below are different doors into the same model.
The six questions and their risks, free and public — where AI use goes wrong.
Learn to answer all six questions — applied AI fluency.
The interactive curriculum, for enrolled learners.
Facilitator lessons — how to teach each domain.
For the people who teach
Where these lessons come from.
These ten lessons are part of the Certified Applied AI Trainer Program.