As a newly minted MD, I know every recent med school graduate has something to say about how their med school experience could have been different. My previous article commented on the ways medical school could be more valuable in this age of technology and information overload, and I have continued to think long and hard about what new waves of physicians should learn in medical school.
The topic that immediately comes to mind is AI. The field of medicine has faced the conundrum of how to implement and balance the use of this cutting-edge technology without compromising academic integrity, patient data privacy, or clinical expertise. In fact, many institutions lack the knowledge, experience, and expertise to teach AI at all.
Perhaps the biggest problem I saw in med school was unsupervised learning and use of AI, as students and trainees are not constantly monitored by their mentors. I have seen students use large language models (LLMs) like ChatGPT in classroom settings, typing in the exact questions their professors ask so they can respond within seconds. They no longer open textbooks because they can feed classroom material into Google’s Notebook LLM and have it generate summaries, flashcards, and questions for test preparation. All of this can be done in seconds to minutes.
On the hospital floors, I once saw a senior resident physician use ChatGPT to generate a note. After patient encounters, he put the mic on, told ChatGPT about the patients' histories and his proposed assessment and plan, and then asked it to organize the content in the format he preferred. He then asked it to check his documents had holes and mistakes. In a matter of minutes, he completed a preliminary note he could copy and paste to the EHR.
The convenience we have at our fingertips with AI is far greater than many physicians imagine, and I have encountered attending physicians who have not experimented with AI at all. They know neither what it can possibly do to help us nor the pitfalls of improper AI use.
A few of the biggest issues with AI use in healthcare are misinformation and data privacy. The concept of hallucination should not be news to many of us by now. AI providing false information and incorrect references has led to numerous retracted academic articles, and, in the worst cases, patients have acted on incorrect medical advice from AI. There is also no guarantee that physicians, as users, will not make the same mistakes.
Another consideration is that general consumer AI products like ChatGPT and Gemini are not HIPAA-compliant. Any queries or data that should be treated as PHI can be used as training material for LLMs. Many clinicians, including the resident I described earlier, may have unintentionally provided free resources for the LLM to train on, benefiting companies rather than our patients, not to mention compromising PHI. Data privacy and security are no small concerns; thus, all LLM users in healthcare, including students and trainees, must be taught proper ways to avoid misusing the technology readily accessible to them.
While it is imperative to understand how AI can endanger patients and clinicians, we must also explore how AI can help us. Due to the aforementioned concerns, I know many attending physicians are reluctant to implement AI in their clinical workflow and hesitant to explore AI further. My concern is that these reluctant physicians will fall behind. As Dr. Eric Topol, author of "Deep Medicine," said, “Physicians will split into two groups in the near future — those who can use AI well and those who don’t.”
Aside from AI scribe, the biggest benefit I see in using AI is the ability to retrieve accurate, up-to-date medical information through tools such as OpenEvidence and Doximity Ask. As these products offer cross-referenced information from reputable academic journals and official guidelines from medical societies, we, as physicians, can more reliably seek information and trust the process by which it is gathered. In practice, patients can be better helped to understand anything confusing about discussions they had with clinicians, and physicians can easily look up the latest medical guidelines, research, and data as needed.
Since the advent of ChatGPT, I believe we are at a point of a paradigm shift in which knowledge delivery and learning methods are being revolutionized by AI. Long gone are the days when students would physically go to libraries and search through journals and textbooks to find the information they needed. Even Google searches and looking for appropriate content on YouTube are no longer the fastest way to get the information. Students can and will turn to AI as a comprehensive first-line resource for learning. Now is not the time to play the Luddite card, but to face the reality that AI could disrupt the current workflow and bring about changes at a pace we have never witnessed before.
What is extremely difficult is that academic institutions lack consensus on how to use AI. Creating campus-wide guidelines is challenging because different fields have varying uses and interests regarding AI. To keep things simple, the responsibilities have been given to each department and its instructors. In other words, physicians who are actively teaching and educating must learn and set the boundaries of AI, so students and trainees can be aware of its benefits and pitfalls.
As I have just graduated from medical school, I wish we had had more lectures on how we, as students and trainees, could experiment with and learn from AI. Without proper guidance, students may be using AI in ways that violate institutional guidelines, despite their efforts to be careful.
As previous generations have had to embrace computers, gradually replacing paper charts, physicians must welcome the new wave of AI technologies as part of our practice. You may disagree with how AI should be implemented and tested, but that alone is not a compelling reason not to experiment with AI.
New physicians will inadvertently introduce new technologies, learning methods, and ethical concerns into the profession, and we need guidance from veteran physicians on how to incorporate them appropriately without compromising our professional integrity and patient safety. I hope institutions’ administrations take AI issues quite seriously, publish a formal guideline as a living document, and establish workshops to educate all physicians.
It would be appropriate to consider inviting younger physicians and students to discuss how they have been using AI, so learning could take place from the bottom up. There is no point in saying “Do not use AI.” Instead, we need careful hands and eyes so we do not run into legal, security, and medical issues never dreamed possible.
My hope is that physicians learn to use AI as a carefully-wielded tool, not as something that will eventually replace us. For those who have less interest in using AI, I urge you to at least spend some time asking how AI could be helpful in your practice and, if it is disruptive, to what extent. These questions will be important for guiding all AI users in healthcare around you, and this step to address a communal quandary must happen without delay.
What do you think the parameters of AI teaching in medical school should be? Share in the comments.
