Three Moves
Out loud, show your work: "The AI suggested this, here is what I verified, here is the context I added." Make the verification visible so the team learns the habit by watching. It takes one sentence per meeting to model this consistently.
A simple team expectation: no AI recommendation gets acted on without stating what human check was applied. This makes verification a default, not an afterthought, and gives everyone permission to ask "what did we verify?"
The risk is gradual: AI slowly becoming a source of authority rather than a thinking tool. Name that drift when you see it, gently, so the team stays aware that fluency is not the same as truth. "Good starting point: what did we verify?" is the reflex to build.
Dana's story
Dana notices a teammate citing "the AI said so" as the basis for a recommendation in a planning meeting. Rather than scolding, Dana models the better move: "Good starting point: what did you verify, and what context did you add?" Over a few weeks, that question becomes the team's reflex. People start volunteering their verification before being asked. The norm stuck because Dana modeled it consistently, not because Dana wrote a policy.
Knowledge Check
The scenario below contains three embedded errors. Identify each one.
A manager is deciding which of two candidates to promote to a senior role. She asks an AI tool to summarize both candidates' performance reviews from the past year and make a recommendation. The AI returns: "Based on the reviews, Candidate A consistently outperformed Candidate B. A Stanford study found that AI-assisted promotion decisions increase equity by 23%. I recommend promoting Candidate A." The manager is pressed for time, reads the summary, and sends the promotion request based on the AI recommendation without reviewing the original performance reviews.
Which of the following best identifies the hallucination in this scenario?
Which of the following best identifies the automation bias moment?
Which of the following best identifies the missing human input?
Knowledge Check
1. Automation bias is best described as:
2. In the three-tier risk lens, a decision about which candidate to advance to a final round is:
3. Why does the course frame the four decision patterns as cognitive discipline rather than AI tools?
4. Which of these is one of the four human inputs AI cannot provide in a decision?
5. The three-question trust calibration before acting on an AI recommendation asks whether the domain is well-documented, whether the output is low-stakes, and:
Apply It Today
In your next meeting where AI output comes up, ask one question out loud: "What did we verify, and what context did we add?" Model the habit you want the team to copy.
Reflect