AI Use Within Clinical Settings and Pragmatic Best Practices

• Anonymously summarize a challenging actual clinical encounter and use the same question/prompt in Open Evidence, UpToDate AI, Chat GPT and other LLMs and see which was produces the most appropriate and clinically grounded answer.

• Ask your patients if they use Chat GPT, Google Gemini AI or other LLMs before and/or after their visits and ask how they use it.

• Read Thinking Fast and Slow by Daniel Kahneman.

Legal and Ethical Implications of AI in Clinical Settings

• Create a Google alert to keep yourself updated on AI and Healthcare.  There is rapid adoption and you want to be aware of the trends.

• Invest in one of the AI platforms and become proficient at prompt engineering.  It will help you personally, but also grow a better understanding of how AI works.

• Play with AI. Use it outside of the clinical space so you can better understand how it works in the clinical space.

AI Case Studies

Presentation 1:

• Treat AI like a very fast intern who never admits uncertainty.

• Don’t ask “Is this right?” → Ask “How could this be wrong?”

• You are accountable for outcomes.

Presentation 2:

• Human data is tricky. Test models on the data from populations they will be used on.

• Robust learning is hard. Models choose “general” performance on data they have SEEN, and this hides failures. Test models on corner cases that they (but not humans) may fail at.

• Humans are fallible. Test models+humans in high risk/uncertainty situations beforehand.

"So What? How Will AI Redefine the Healthcare Workforce?"

• Physicians need to actively participate in AI in healthcare, whether it's participating in a pilot, having AI-focused journal clubs, or providing structured feedback on workflow and usability.

• AI is most useful to a clinician as an assistant working under your supervision, not as an authority working unsupervised. Think of a task you've been hoping your own Medical Assistant could help you with, but hasn't had the time, or something you've been meaning to research and read about, but similarly haven't had the time. Ask an AI tool to support you.

• Consider asking patients about their LLM use (Claude/ChatGPT/Bard)