AI at the bedside and in nursing

About 25 minutes · plain language

At the bedside, AI suggestions meet a real patient. The nurse is often the last person who can catch an AI error before it reaches the patient. This lesson explains the two biggest traps, and the nurse’s role before, during and after an AI tool is introduced.

Your assessment comes first

The figure shows your place: between the AI output and the patient. The tool gives information; you make the assessment and the decision. Document your own findings, not “as per AI”.

Automation bias

People tend to trust automated advice too much. They accept wrong suggestions (errors of commission) and miss problems the tool did not flag (errors of omission). Systematic reviews find this bias in many health-care settings and show it is more likely when workload is high or when outputs are hard to verify. Simple habits help: look at the patient before the screen, and ask “does this fit what I see?”

Alert fatigue

When a tool fires too often, people start to dismiss alerts without reading them, including the important ones. Good tools are tuned with clinical staff so that alerts are fewer and more useful. If an alert is almost always wrong, report it rather than silently ignoring it.

The nurse’s role across the tool’s life

Before: help choose the tool and test it, for example in a silent trial. During: use it with your own judgement, override it when needed, and document your reasons. After: report outputs that look wrong and patterns you notice, such as a tool that often misses certain patients. Nursing bodies, including the American Nurses Association and the American Academy of Nursing, say nurses should help shape AI, not only use it.

Speak up

Your reports are how local monitoring works. A short note such as “score low, patient deteriorated, sepsis confirmed” helps the team retune or stop a tool before harm spreads.

When a patient brings an AI answer

More and more patients ask chatbots about their symptoms, results and medicines before or after they see you. Studies show that chatbot answers can sound clear and caring, sometimes more so than busy clinicians, and still contain errors. A patient who brings an AI answer is usually trying to understand and take part in their care. Treat it as a starting point for a conversation, not a challenge.

• Ask and listen. “What did it tell you, and what worries you about it?”
• Check together. Compare the answer with their actual results, medicines and plan, and with a trusted source (Lesson 5).
• Correct clearly and kindly. Say what is right, what is wrong or does not apply to them, and why.
• Protect privacy. Remind them not to upload full records or other people’s details to public tools.
• Escalate and record. If the AI advice could cause harm, such as stopping a medicine, involve the prescriber and note the discussion.