Read Time: 6 minutes │ Subscribe to the FREE Newsletter │Join the Premium Patch

Hi Ammis & friends!

Welcome to the 🍃 Premium Patch, where we share fresh perspectives from fellow Ammis & friends on AI and their areas of expertise.

Earlier this year, I had the immense privilege of discussing AI + Human Brains + Healthcare with Dr. Beatriz Thames, my childhood neighbor, schoolmate, college roommate, and certified lifelong bestie.

Today, we'll explore Beatriz's evolving relationship with AI as a senior neurology resident, from initial indifference to regularly using tools like ChatGPT, and her insights on incorporating AI into her medical workflow to enhance practice without replacing the human elements essential to quality healthcare.

🍃 You’ll discover:

• Beatriz's journey to Neurology and embracing AI

• How she approached her personal AI literacy journey

• Her view on how AI can enhance healthcare

• Her essential technologies for home and work

👩‍🌾Let’s dig in!

Beatriz Thames on Neurology + AI

Get to know Beatriz

R: How would you fill in this blank to describe your relationship with AI? __________.ai

B: Evolving.ai. Honestly, in the first few years of my residency, we did use AI tools, but they were kind of in the background, and they didn't seem so robust.

We used AI software called Rapid AI to help triage stroke patients. The software had AI technology to look at the patient’s scan to identify whether they might have a large vessel blockage. Frankly, I didn’t look at the software results; I would look at the pictures myself.

We also used Dragon dictation software to help us write our notes faster. But similarly, I found it clunky and was better off writing the notes myself.

But what really got me excited about AI in this last year, my senior year, was when my friend was casually playing with ChatGPT. He was making poems about neurology. And the results were so clever! Sometimes, I get a little isolated in my medical world. I was like, "Wait, what is this? This is super cool."

I had no idea ChatGPT had these capabilities. I thought, maybe aside from playing around with it, what can it actually do for us to make our work faster? So, within the last six months, I’ve tried to self-educate on how foundation models and large language models work. I would love to use this technology to help me with what I call pain points in the neurology workflow.

The biggest challenge is just the sheer amount of knowledge that we have to learn and how much we have to constantly learn. If this tool is so good at taking information and summarizing it, it’s a good tool to address that pain point. And so that's why I think my relationship with AI is evolving.

Six months ago, I didn't understand anything about it. Now, thanks to my excitement about ChatGPT, I'm feeling a little bit more comfortable. It doesn't seem like magic anymore. I kind of understand the underlying processes at work that are leading it to produce answers. And I'm also better understanding its limitations.

R: Evolving.ai - I love it. Before we go any further, could you share a little bit about yourself and your path to the point that you are in life?

B: Yes! I am currently a senior neurology resident at Methodist Hospital in the Texas Medical Center. As early as high school, I knew I wanted to go into medicine and deal with the brain, because I feel like it's the seat of our existence – the organ responsible for who we are as people, our emotions, memories, etc. Anytime something goes wrong with that very important organ, it's high risk, but also high reward if you can help it.

I thought I wanted to do surgery until I got into medical school and did my surgery rotation, which stressed me out. So I chose neurology instead, which still deals with the brain but in a more non-invasive way. I wanted to stay close to my family in Houston, so it worked out well that I matched into this residency program.

Along the way, I found out that I really like the neurology subspecialty of movement disorders, which is very focused on diseases that either cause excess movement or too little movement. For example, too little movement is Parkinson's disease, and hyperkinetic disorders include things like essential tremor, Huntington's, Wilson's, and ataxias.

I'm really interested in seeing how I could use AI in this field, maybe steering a little away from larger language models. I have so much to learn about how AI is being used in video analysis. In movement disorders, we take videos of patients with abnormal movements, and then make human judgments to diagnose. I would like AI to help us analyze better.

Beatriz’s AI Literacy Journey

R: How did you approach learning about AI, especially large language models, as someone knowledgeable about the human brain? What was easier or particularly interesting, considering the parallels?

B: There are no AI courses in medical school, so I had to seek out mentors, read papers, watch videos, and self-educate. I found it really cool that neural network architectures inspiring deep learning are structured similarly to the brain at a molecular level with nodes connected in networks, though simplified.

It's amazing that we've managed to create a computational formula mimicking the brain's molecular structure that can produce generative AI capabilities previously thought impossible to program, like creativity. I'm not sure if we can program other very human qualities like wisdom, ethics, and values into AI yet.

R: You recently presented on this topic - ‘Demystifying and Leveraging Large Language Models in Neurology Practice’ - to your Neurology department members at Houston Methodist. I was lucky to attend virtually. I really loved the chart in your slides with brain functions matched to the AI functions that mimic them. When you first saw that chart, was it an instant "aha" moment of seeing how these things mapped, and did that accelerate your understanding coming from a neuroscience background?

B: What I loved about those diagrams is that I knew I would be presenting to neurologists. I think we like to see familiar things and make connections. As a neurologist, neuroanatomy is our bread and butter. So I thought it was really cool how those diagrams, the data intelligence continuum and the anatomical brain view, showed how the brain's functions can be matched to machine equivalents. That all came from an excellent book, though regrettably, I haven't finished its 600 pages yet.

Even around 200 pages in, I struggled with parts regarding the different algorithms and statistics involved in computing. I question how much technical detail I truly need to understand versus just grasping the limitations. I'm not trying to become a programmer—I want to stay a neurologist but learn to leverage this technology properly.

Using AI in Life and in the Neurology Practice

R: How do you use AI today, personally or professionally? What are some of your favorite ways to use it?

B: It's still an evolving relationship where I'm learning to use it effectively. But as you know, I've become a fan of ChatGPT. Based on colleagues' recommendations, I got the ChatGPT Plus subscription, which let’s me access the GPT-4 AI model that tends to have less errors.

I've learned you can create custom, domain-specific GPTs focused on sources you provide. My colleagues have uploaded textbooks so they can ask GPT clinical questions based on guidelines.

I'm working on a "NeuroBuddy" custom GPT by uploading residency handbooks and reliable sources to assist with institution policies and clinical queries, initially focusing on stroke and emergency scenarios.

For example, I asked about evidence for venous stenting in idiopathic intracranial hypertension cases. The model provided some study references, likely extracting them from PubMed, which is exactly what I wanted as a starting point, though we always watch for potential hallucinations requiring verification.

R: One of my favorite analogies is that ChatGPT is like a smart intern. You definitely need to review its output carefully, but it gets you a good portion of the way there. I can’t think of anyone who could use an intern’s help more than overworked medical residents.

My last question is on your view of AI's role moving forward in medicine? Many feel apprehensive it will replace their profession.

B: I consider myself an optimist based on what I've read. I don't think AI will replace any specific profession because what we do, especially as neurologists, involves more than just memorizing guidelines.

There's pragmatism involved with complex patient cases lacking clear evidence one way or another. And the physical aspect of examining patients and interpreting findings is likely very challenging for AI currently. Even routine tests have false positive/negative rates, so you likely want humans interpreting those while AI handles more concrete tasks.

So I'm less concerned about AI taking our jobs for now. It could make them easier if leveraged properly with its limitations understood. But it could also provide no benefit, like triage software flagging excessive false positives we then have to double-check anyway. We're still working on integrating AI seamlessly into workflows.

What’s in Beatriz’s Tech Stack? 🪴 

Apple or Android?

Both

Mac or PC?

PC

What are your favorite or most essential technologies right now?

ChatGPT

What technology or app would you recommend to Ammis & friends to make their life easier?

Strava for tracking workouts

🌾 Cream of the Crop

Many thanks to Betariz for sharing her invaluable insights and helping us grow.

🍃 If you have ChatGPT Plus, check out this GPT she created, NeuroBuddy!

Thank you for spending a few of your precious, precious minutes with us.

We’d also love your help to grow the ammi.ai 💚 Garden!

See y’all soon!
Ruqaiya
Ammi by day, Ammi by night