McCance Seminar Series: Roy Perlis, MD, MSc

Why It Matters

AI tools could streamline psychiatric workflows and raise prescribing standards, but unchecked implementation may compromise clinical judgment and patient safety.

Key Takeaways

• Ambient AI scribes reduce documentation time but may alter clinical decisions.
• Scribe-generated notes are longer, capture more symptoms, yet lower intervention rates.
• AI decision-support tools can match expert prescribing for depression, outperform community clinicians.
• Zero‑shot language models achieve ~85% alignment with expert medication choices.
• Regulatory gaps leave AI scribes largely unregulated, raising safety concerns.

Summary

The seminar highlighted how artificial intelligence is reshaping psychiatric practice, from ambient AI scribes that automatically document visits to language‑model decision‑support tools that suggest treatments. Dr. Roy Perlis described early experiments at Mass General Brigham, noting that AI scribes saved roughly 13 minutes per week of charting but produced longer notes with more symptom detail and a 27% drop in odds of psychiatric interventions.

Data from a network‑wide study showed that while scribes increase documentation breadth, they may inadvertently reduce clinician action, raising concerns about signal‑to‑noise ratio and unregulated AI use under the 21st Century Cures Act. Conversely, a separate project applied zero‑shot language models to antidepressant prescribing, achieving about 85% concordance with expert recommendations—far surpassing the ~50% alignment of community clinicians—and demonstrated better avoidance of harmful medication choices.

Perlis emphasized the paradox of AI: it can alleviate administrative burdens yet subtly shift clinical behavior. He illustrated this with a quip, “I like growing AI and I’m worried about AI,” underscoring the need for careful evaluation of both benefits and unintended consequences.

The findings suggest that while AI holds promise for standardizing care and reducing burnout, the lack of clear regulatory frameworks and the potential for altered decision‑making demand vigilant oversight and further research before widespread adoption.

Original Description

Dr. Roy Perlis presents, "AI beyond evil robots: It’s transformative, but is that a good thing?" as part of the McCance Center for Brain Health Seminar Series. https://www.massgeneral.org/neurology/mccance-center