AI in Medicine with Dr. Brian Patterson : WHSLA Web Conference 23 Summary

Friday afternoon of the WHSLA conference featured a fireside chat with Dr. Brian Patterson, Associate Professor and Medical Informatics Director for Predictive Analytics and Clinical Decision Support, UW Health.

Dr. Patterson shared his knowledge of artificial intelligence including large language models (LLMs) such as ChatGPT and how they have evoloved over time and will continue to evolve. 

What are people doing with AI in Medicine? People are using 3rd party apps on the web like Nabla Copilot, DocsGPT.  Many companies have outlawed these tools *, but people will be using them anyway on their personal devices because they see the value in it.

Dr. Patterson expects there will eventually be integration with existing IT products -- including the EPIC project that you mentioned--as businesses license closed shop versions that won't recycle and sell company secrets.  

He suggested that AI would be more of a trust-but-verify copilot, "like a book-smart intern who comes in hung-over sometimes" vs. "the crystal ball is broken if it doesn't always get the right answer."

• ChatGPT for discharge instructions
• Drs will spend less time on documentation and other drudge work, which may reduce burnout
• These tools make it very easy to draft content, but the end-user is still responsible for the final draft.

During the Q&A portion questions were posed about the reliability of ChatGPT especially in regards to citing accuracte evidence and not hallucinating (i.e. making up citations). This is something that ChatGPT might get better at in the future, but human authors using LLM tools will always need to verify, confirm, and attest to the content they create with the help of ChatGPT. 

Dr. Patterson shared that he has been surprised that the uptake and interest in adopting new LLM technologies has happened as quickly as it has. Right now UW-Health is working with Epic to use AI: UW Health using Microsoft AI in Epic Systems medical records (news article). It's likely that more and more health systems will allow uses of AI as the technology evolves and the benefits are seen. One area Dr. Patterson hopes AI can help with is by decreasing paperwork burdens such as charting on providers. An April 5, 2023 story on NPR talked about just that.

For more reading, check out this book mentioned by Dr. Patterson: AI Revolution in Medicine: GPT-4 and Beyond. It may tell the all-too-rosey side of the AI story. 

For another POV, check out this recent Medlib-L post on AI for cataloging and literature searches. 

Prompt by JPxG, model by Boris Dayma, upscaler by Xintao Wang, Liangbin Xie et al. (Apache License 2.0 or BSD), via Wikimedia Commons

Practical AI for Instructors and Students: Prompting AI

Practical AI for Instructors and Students Part 3: Prompting AI (about 12 minutes)

Wharton Interactive Crash Course: Practical AI for Instructors and Students

In this video, Wharton Interactive's Faculty Director Ethan Mollick and Director of Pedagogy Lilach Mollick discuss how to effectively prompt AI like Midjourney, ChatGPT, Microsoft’s Bing, as well as how to take the lead, weaving your own expertise into the interaction. 

This video is Part 3 of a five-part course in which Wharton Interactive provides an overview of AI large language models for educators and students. They take a practical approach and explore how the models work, and how to work effectively with each model, weaving in your own expertise. They also show how to use AI to make teaching easier and more effective, with example prompts and guidelines, as well as how students can use AI to improve their learning.

 

MM: This is the part of AI that is so interesting to me: How to get it to do what you want it to do?

1. Tell it who it is -- Give it some context.
2. Give clear instructions about what you want.
3. Give it examples and steps
4. Interact with it - Push back
5. Practice until you get good at it!