Teaching Others to Think Critically

The skills in this course are worth having. They're worth more if the people you work with, live with, and learn alongside have them too.

Critical thinking doesn't spread through lectures. Telling someone they're being credulous doesn't make them more critical — it makes them defensive. The most effective transmission method is modeling: asking the questions you want others to internalize, visibly and consistently, until asking becomes a shared habit rather than an individual quirk.

The Question, Don't Correct Principle

When a colleague shares an AI-generated claim that might be wrong, the instinct is to correct it. "That's probably a hallucination." "I don't think that statistic is right." "You should have checked that."

These responses are sometimes necessary. They're rarely the most effective way to build someone's critical thinking habit.

A more effective approach: ask the question they didn't ask themselves. "That's interesting — do you know where that figure came from? Could we trace it?" "I haven't seen that finding before — how would we verify it?" "What's the source for that one? I want to make sure I can cite it if someone asks."

This achieves two things simultaneously. It catches the potential error without framing the person as uncritical. And it models the habit — you're demonstrating, in real time, what it looks like to pause and verify, so they can see that behavior as normal rather than paranoid.

Designing for Critical Thinking in Teams

Individual habits matter. Team environments matter more. If your organization's workflow normalizes passing AI output to the next stage without verification, individual critical thinking instincts erode under time pressure.

Three structural changes that make a material difference:

Make the verification step visible. In a review or approval process, add an explicit field: "Factual claims verified: yes / no / N/A." This normalizes the check without requiring individual initiative every time.

Build in a pre-mortem moment. Before significant decisions, create a five-minute space for the question: "What assumption in our analysis, if wrong, would change this decision?" It doesn't require a framework or a facilitator — just the habit of asking before committing.

Celebrate catches, not just outputs. When someone catches an AI error before it causes a problem, name it explicitly: "This is exactly what good practice looks like." The behavior you celebrate is the behavior you get more of.

When to Push Back and How

Sometimes you need to correct, not just question. Someone is about to submit an AI-generated report with fabricated citations to a client. Asking "where did that come from?" isn't sufficient — you need to stop the output.

The most effective pushback isn't about the AI. It's about the risk: "Before this goes out, I want to make sure we can stand behind everything in here — can we take ten minutes to check the three main figures?" This frames verification as professional diligence, not distrust of the person.

It also avoids the AI blame-deflection trap: "The AI gave me that" is not a defense once something has gone wrong. The accountability sits with the person who used the output, not the model that generated it.