Previous poll results: The integration of AI

Artificial intelligence (AI) represents one of the most significant changes in health care in decades, with the potential to change the way we practice medicine forever. In particular, the advent of large language models (LLM) has super-charged the AI revolution due to their advanced natural language processing (NLP) functions, generative capabilities and ability to think creatively. LLMs like OpenAI’s Chat GPT represent true general AI models, able to learn and utilize information across multiple domains. They are also the first AI that can rightfully claim to have passed the Turing Test, found to be indistinguishable from humans in a study of about 4000 virtual therapy patients on the app Koko using the GPT-3 model, and in a different study, to actually exceed humans in performance of emotional awareness understanding.

In the face of this metaphorical impending tidal wave, students and post-doctoral trainees might feel a range of emotions: from excitement, to trepidation, to confusion about how to get involved.

When I was going through med school, there was a huge movement to teach ultrasound. I remember one of my professors calling it “the stethoscope of the future.” With the amount that ultrasound was highlighted on my school’s advertising brochures, it was readily apparent that ultrasound training might not be a standardized part of medical education at that point and schools that were teaching it in an organized way were on the “cutting edge.” During residency, I rotated frequently through a large allopathic Tier 1 academic center, the flagship medical school and teaching location for the state. Even in that setting, less than 10% of the fourth-year medical students I worked with in the ICU and on the medicine wards were able to do even basic point-of-care ultrasound (POCUS) exams with proficiency. Now, I use POCUS every day in practice. In fact, I would say that not only would it not be possible to practice as effectively without it, but my patients would actually be getting substandard care from me if I declined to use ultrasound.

So how does POCUS compare to AI and LLMs? Like ultrasound, clinical AI and LLMs are relatively novel “tools” that have significant ability to improve both patient care outcomes and physician well-being/the enjoyment of medicine. In the near future, AI has the potential to become so powerful that failure to utilize it during the diagnostic process may actually breach the standard of care. Being aware of AI is a huge first step, and already puts you ahead of the curve. Many attendings and teachers in medical school don’t have a lot of experience or know very much about it, so the playing ground is fairly level. For students and trainees, this is an opportunity to push the envelope on your own and experiment where safe and reasonable with generative AI in patient care. LLMs have multiple immediate applications in medicine that are accessible to students and trainees, including as virtual medical scribes (ambient dictation), point-of-care clinical decision support and research review. Two LLM AI products you can use today are Open Evidence and Nabla Copilot.

Disclaimer: I have no financial incentive or connection to either of these products.

Open Evidence is a clinical decision and research support tool that searches only peer-reviewed papers and evidence-based guidelines to answer your question. Best of all, it is excellent for research because it produces a fully sourced bibliography for every query, complete with in-text citations. Open Evidence is not-for-profit and free as long as you can provide an NPI number.

Nabla Copilot is a HIPAA-compliant ambient dictation scribe that will listen to you talking to a patient and create a SOAP note based on your exam. While there are many apps that do this, where Nabla stands out is that it allows an unlimited Pro subscription for free for students and trainees. That’s incentive enough to give it a try, as you would have to pay at least $20 per month to get similar performance out of Chat GPT.

Remember, when using AI, trust but verify. Hallucinations, or moments when an LLM creates fictious information to satisfy a prompt, are rare but not impossible. Additionally, it’s on you to check with your institution and make sure you’re not violating their policy by using a certain AI tool.

Your other checklist items: read as much as possible coming out of publications like JAMA AI and NEJM AI and keep abreast of the big issues and developments in AI. Find ways to get involved in AI policy and add your comments from your perspective as students. Finally, your most important task: ask your teachers and administration officials to develop a curriculum to train you on these new health tools and other emerging technologies. There are several schools with “digital health” models or curriculum to use as models, but they are far from the norm.

We have an opportunity as DOs and future DOs to give our profession a leg up on the rest of health care. It’s up to us to make our voices heard and ensure we are being properly trained in and given the opportunity to access AI as we prepare for practice.