Nurse informaticist urges AI literacy and governance as clinical tools spread

Dr. Steph Heltzer, professor and MSN Informatics Program Director at Texas Tech University Health Sciences Center, told a Health Bites webinar audience that nurses must develop foundational AI literacy and be part of organizational decisions as clinical artificial intelligence becomes more common.

"Understand the origin and the data sources of any AI tool and the first step of assessing its trustworthiness," Heltzer said, arguing that human review against evidence‑based sources is the "most critical" step before acting on AI outputs. She framed the issue as a responsibility for clinicians and organizations rather than a technical curiosity.

Heltzer outlined a practical framework she helped publish, the Nurses AI Guide to AI literacy, which she said maps basic competencies into domains she summarized with the acronym NURSES. She emphasized the "N" (navigating basics) — knowing who built a tool and what data trained it — and the human review step that should follow any AI output. "A tool never really expresses doubt in itself. It treats output like it's right," she said, describing how probabilistic token models can be confidently wrong.

The presentation reviewed common failure modes such as hallucinations (fabricated facts or references) and confabulations (partial truths), and Heltzer urged clinicians and educators to verify citations and outputs rather than accept them at face value. She also warned that bias in training data or algorithmic design can perpetuate unfair decisions, and raised the possibility of "data poisoning," where attackers skew training data.

Heltzer listed clinical areas where AI has shown value — early warning systems, documentation aides, radiology/pathology tools, drug‑interaction checks and predictive models for falls or readmissions — while cautioning against inappropriate uses such as replacing hands‑on assessments or bypassing informed‑consent processes.

On workforce impacts, she warned that over‑reliance on AI risks "deskilling" clinicians and stressed the need to teach downtime procedures and preserve critical judgment. Heltzer urged nurse leaders to secure seats at procurement and governance tables: nurses control much of day‑to‑day workflow, and their input is essential to safe deployments.

In a question‑and‑answer session, Heltzer said organizations often adopt tools rapidly for competitive or financial reasons but should pair deployments with governance, transparency and training. Citing early pilots and published studies, she said ambient scribing and transcription tools have shown time savings for nurses — "nurses are reporting getting a couple hours back" in some studies — but she stressed review before signing and attention to licensing/equity of access.

She described librarians as pivotal partners in research workflows and scoping reviews, helping speed literature screening while maintaining a human‑in‑the‑loop for verification. She recommended microlearning and university certificates (Coursera, Udemy, university programs) and named available certifications (AIMP, ABAIM) while declining to endorse specific vendors.

Responding about environmental impacts, Heltzer noted that training and hosting large models require substantial power and water for data centers and urged communities and planners to consider infrastructure and environmental impacts when data centers are proposed.

The webinar was produced by the National Library of Medicine network. The session closed with a call to action for nurses to move from literacy to fluency, build competencies, participate in governance and advocate for ethical, transparent AI use in healthcare.