The AI Tutor in a Mandatory Ethics Course
A hospital system requires all 1,400 clinical staff to complete a 2-hour annual ethics and patient-privacy course. To cut costs, the education team proposes replacing the facilitated case discussions with an AI chatbot that answers staff questions about policy and generates practice scenarios. The vendor claims the chatbot is "trained on our policy library." A pilot with 60 staff showed high satisfaction, but during the pilot the chatbot gave a confident but incorrect answer about when a privacy breach must be reported to a regulator, citing an outdated policy version. Two staff acted on the answer in a tabletop exercise before a facilitator caught it. The compliance officer is enthusiastic about the cost savings; the privacy officer is alarmed. The course is due to launch in six weeks.
1. Evaluate the proposal against ethical, accessibility, and professional standards. What are the non-negotiable conditions for using the chatbot, if any?
Read the worked answer
The pilot surfaced the central risk: a system that gives confident wrong answers about regulatory obligations in a domain where wrong answers have legal and patient-safety consequences. High satisfaction is not evidence of accuracy, and in an ethics and privacy course, satisfaction can even correlate with the chatbot telling people what they want to hear. The fact that two staff acted on an incorrect answer before a human caught it shows the failure mode is not theoretical.
Non-negotiable conditions, if the chatbot is used at all: it must be grounded in a version-controlled, current policy library with the source and effective date shown for every answer; it must decline or escalate when it lacks a confident, sourced answer rather than generating one; it must not be the sole source for any regulatory obligation; and there must be a human review and escalation path that staff can reach easily. The privacy officer must sign off on the content and the update process, and the outdated-policy failure must be root-caused before launch.
I would also insist on a pre-launch accuracy test against a set of known-answer questions, including the breach-reporting question that failed, with a pass threshold agreed in advance. Accessibility matters too: the chatbot must work with screen readers and offer a non-chat alternative, since not all staff can or will use it.
Professionally, the chatbot can support practice and routine questions, but the facilitated case discussion carries judgment, nuance, and accountability that should not be removed to save cost in a mandatory compliance course.
2. The compliance officer wants the full replacement to capture the savings. How do you negotiate the decision, and what would you recommend as the final design?
Read the worked answer
I would reframe the decision as a risk trade-off rather than a cost-saving debate, and put the pilot failure in front of both officers in writing: the specific question, the incorrect answer, the outdated citation, and the fact that staff acted on it. I would ask the compliance officer to state, on the record, who owns the consequence if a staff member relies on a wrong chatbot answer about breach reporting. That question usually changes the tone of the conversation.
My recommended design is a blended model: keep a shortened facilitated discussion for the highest-stakes content, especially breach reporting and consent, and use the chatbot for practice scenarios and routine policy questions with source citations and escalation to a human. This captures much of the savings while keeping a human accountable for the judgment-heavy content.
I would propose a staged launch: run the blended model with one cohort, track accuracy-test results, escalation rates, and facilitator-caught errors, and set a review gate before expanding. I would also require a named owner for the policy library and a change process so outdated content cannot silently persist. If the compliance officer still insists on full replacement, I would document the privacy officer's objection and my recommendation, and escalate to the executive sponsor rather than proceed silently.