NU710 Unit 4 AI Documentation Ethics Analysis Example

The Microphone at the Bedside: A Unit 4 Ethical Analysis of Ambient AI That Drafts Insulin Teaching Notes

[Student Name]

Purdue University Global

NU710: Advanced Health Information and Technologies

Unit 4 AI Ethics Analysis

[Instructor Name]

[Date]

The hospital, the vendor pilot and the patients are composite and illustrative; the studies and statements cited are real.

The Proposal

The tool runs on a tablet in the patient's room. When the nurse starts a teaching session, it records the conversation, transcribes it and uses a large language model to draft three things: a structured teaching note for the chart, a plain-language plan for the patient and a short list of "comprehension concerns," moments when the patient's answers suggested confusion. The nurse reviews and signs each draft. The vendor stores recordings in its cloud for 30 days, then deletes them, and says the data will not train its models without a separate agreement.

What It Could Give

The case for the tool rests on documentation burden. Michalowski et al. (2026), in a position paper on multimodal language models, argue that tools combining audio, video and text could update records in real time, reduce manual entry and return time to direct care, while warning that accuracy, privacy and bias must be addressed before such systems are trusted. Early large-scale experience with ambient scribes for physicians in one integrated health system reported time savings and generally good note quality, alongside occasional errors that clinicians had to catch (Tierney et al., 2024).

On the pilot units, an insulin teaching session runs about 25 minutes, and the teaching note takes another 8 to 10. A nurse who teaches three patients a shift spends about half an hour writing about teaching. A draft that needs only review could return most of that time. The plan written for the patient could also be better than today's, since it would reflect what was actually said rather than a generic template.

The comprehension flags are the most interesting feature. A nurse may miss the moment a patient says "so I take the long one at breakfast" when the plan says bedtime. A tool that catches such moments could prevent the dosing errors the bundle is designed to stop.

The same microphone that could catch a patient's confusion also records everything else the patient says.

Consent and Privacy

Teaching conversations are rarely only about insulin. Patients mention money, family conflict, drinking, a fear of needles left over from a childhood hospital stay. A recording captures all of it. Consent must therefore be specific and freely given: patients should be told what is recorded, where it goes, how long it is kept and that declining changes nothing about the teaching or the discharge. Family members in the room are recorded too and must also be asked.

The vendor's 30-day cloud storage means protected health information leaves the hospital. The hospital would need a business associate agreement, a security review and a clear answer on whether recordings could be subpoenaed or breached. The vendor's promise not to train models without consent should be written into the contract, not left to a marketing page.

Accuracy and Accountability

Language models can produce fluent text that is wrong. In a teaching note, a wrong dose, a wrong time or a claim that the patient demonstrated correctly when they did not becomes part of the legal record. The nurse who signs the note is responsible for it. The risk is that, after weeks of accurate drafts, nurses stop reading carefully, an automation bias that turns a time-saving tool into a source of error. The ANA's 2022 position statement on AI keeps responsibility for practice with the nurse, whatever tool drafted the words, and warns against letting technology stand in for clinical judgment (American Nurses Association [ANA], 2022).

Fairness

Speech recognition and language models perform unevenly across accents, dialects and languages. The pilot units serve patients who speak Spanish and Marshallese, and many older patients with hearing loss who speak softly or haltingly. If the tool transcribes them poorly, their notes will be worse, their plans less accurate and their comprehension flags less reliable, so the patients most at risk of misunderstanding would get the least benefit. The vendor has not reported performance by language or for interpreted sessions.

Nursing Judgment and the Relationship

There is a quieter risk. Teaching is a relationship as well as a transfer of information. A patient who knows the session is being recorded may say less, especially about the fears and barriers that matter most. A nurse who knows a tool will flag confusion may listen less closely. Neither effect is certain, but both should be watched.

What Nurses on the Pilot Units Said

The proposal was discussed at a staff meeting on 5 West. Nurses were split. Several welcomed anything that would cut charting time, and one said the comprehension flags would help newer nurses most. Others worried about patients who would not speak freely and about being recorded themselves, asking whether recordings could be used in performance reviews. That question deserves a written answer: the hospital should state that recordings will not be used to evaluate staff, and the contract should prevent the vendor from offering such features.

Weighing the Considerations

The benefits are real but uncertain; the risks are specific and in most cases manageable. Refusing the tool outright would forgo time savings and a possible safety gain. Accepting it on the vendor's terms would expose patients to privacy risks and the hospital to accuracy and fairness problems it cannot yet measure. The analysis therefore recommends a limited pilot under conditions.

Recommended Safeguards

1. Specific opt-in consent from the patient and anyone else present, with a scripted explanation and no effect on care for those who decline.

2. A signed business associate agreement with no secondary use of recordings or transcripts and deletion verified by audit.

3. Mandatory line-by-line review of the dose, timing and demonstration fields, with the draft and the signed version compared in a monthly audit of 20 notes.

4. Performance by group: the vendor reports transcription accuracy for English, Spanish, interpreted sessions and older speakers before the pilot, and the audit tracks errors by group.

5. No use for interpreted sessions until accuracy for them is shown.

6. An evaluation after 90 days on two units, comparing documentation time, note accuracy, patient-reported comfort and dosing errors at the 72-hour contact with the units' prior data.

Conclusion

Ambient AI could give nurses back time and catch the misunderstandings that cause dosing errors at home. It also records patients' most private conversations, can be confidently wrong and may serve worst the patients who need it most. A limited pilot with specific consent, contractual privacy protections, enforced review, performance reporting by group and a defined evaluation would let the hospital learn whether the benefits are real without asking patients to carry the risks unprotected.

References

American Nurses Association. (2022). The ethical use of artificial intelligence in nursing practice [Position statement]. American Nurses Association.

Michalowski, M., Topaz, M., & Peltonen, L. M. (2026). An AI-enabled nursing future with no documentation burden: A vision for a new reality. Journal of Advanced Nursing, 82(1), 907-912. https://doi.org/10.1111/jan.16911

Tierney, A. A., Gayre, G., Hoberman, B., Mattern, B., Ballesca, M., Kipnis, P., Liu, V., & Lee, K. (2024). Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catalyst, 5(3). https://doi.org/10.1056/CAT.23.0404