ChatGPT in the Classroom
Examples for how AI could be used in teaching
Students and teachers alike are anxiously wondering how Large Language Model (LLM) AI tools like ChatGPT and others will disrupt education. Will it lead to a rash of plagiarism and corner-cutting, with students no longer learning anything? Will legitimate uses get flagged by overzealous teachers as “cheating”?? Can we even tell the difference between those possibilities???
These concerns are amplified further when you consider the high stakes world of graduate and medical education. The New England Journal of Medicine recently published an excellent editorial on this topic:
“Medical schools face a dual challenge: they need to both teach students how to utilize AI in their practice and adapt to the emerging academic use of AI by students and faculty. Medical students are already starting to apply AI in their studying and learning, generating disease schema from chatbots and anticipating teaching points. Faculty are contemplating how AI can help them design courses and evaluations…
At the graduate medical education level, residents and fellows need to be prepared for a future in which AI tools are integral components of their independent practice. Trainees will have to become comfortable working with AI and will have to understand its capabilities and limitations, both to support their own clinical skills and because their patients are already using it.”
This is the optimistic side of an AI revolution—it can act as a personalized tutor to students, come up with new research questions for faculty, and reduce busywork headaches among instructors. The last sentence also provides a reality check: patients have graduated from playing “Dr. Google” and looking up their symptoms online to using AI chatbots, and they will need medical providers who are well-versed in both the benefits and limitations/dangers of those tools. In the spirit of embracing generative AI, I asked ChatGPT1 to brainstorm some possible use cases for AI in a medical school context, and this is what it came up with:
Virtual Patient Simulations: Generative AI can create realistic virtual patients with a variety of medical conditions, allowing students and residents to practice diagnosis, treatment planning, and patient interaction in a safe and controlled environment.
Case Generation and Analysis: AI can generate diverse and complex medical cases for study, enabling learners to develop critical thinking skills by diagnosing and managing different scenarios. AI can also analyze learner responses and provide personalized feedback.
Personalized Study Guides: AI can analyze individual learning patterns, strengths, and weaknesses to generate personalized study guides and learning pathways, optimizing educational content delivery.
Medical Research Assistance: Generative AI can aid in literature review and research by summarizing articles, generating hypotheses, and suggesting relevant studies, speeding up the research process.
Clinical Decision Support: AI-powered tools can simulate clinical decision-making, providing students and residents with practice in making evidence-based decisions for patient care.
Patient Empathy Training: AI-generated patient stories and scenarios can be used to train medical professionals in empathy, communication, and patient-centered care.
This is a big one, as research suggests ChatGPT can produce results more empathetic than human doctors!
Mnemonic and Memory Aids: AI can generate mnemonic devices, acronyms, and memory aids to help students memorize complex medical information more effectively.
Here are a few detailed examples I came up with that illustrate how ChatGPT could be used in medical education…
Generating Case Studies and Quiz Questions
Writing case scenarios for student lectures, homework, and exams takes a lot of time, especially if instructors want to keep the course fresh and change these each semester. Generative AI tools like ChatGPT can help reduce this burden so they can focus on higher-level teaching instead of administrative work. Here is one example I created (none of the output below is edited):
Prompt:
Create a clinical training scenario for fourth year veterinary students. Generate real-world clinical chemistry data for a dog with severe hepatitis and liver failure, and ask three multiple choice questions to test their knowledge
Output:
Clinical Training Scenario for Fourth Year Veterinary Students
Case: A 6-year-old male Labrador Retriever named Max is presented to your veterinary clinic with lethargy, anorexia, vomiting, and icteric mucous membranes. The owner reports that Max has been acting unusually weak and disinterested in his usual activities over the past week. On physical examination, you observe jaundice and palpate an enlarged liver. Blood is drawn for clinical chemistry analysis.
Clinical Chemistry Data:
Total Bilirubin: 8.5 mg/dL (Reference Range: 0.1 - 0.5 mg/dL)
ALT (Alanine Aminotransferase): 750 U/L (Reference Range: 10 - 100 U/L)
ALP (Alkaline Phosphatase): 1200 U/L (Reference Range: 10 - 150 U/L)
Total Protein: 5.0 g/dL (Reference Range: 5.2 - 7.8 g/dL)
Albumin: 2.2 g/dL (Reference Range: 2.5 - 4.0 g/dL)
BUN (Blood Urea Nitrogen): 35 mg/dL (Reference Range: 6 - 24 mg/dL)
Creatinine: 1.8 mg/dL (Reference Range: 0.5 - 1.6 mg/dL)
Multiple Choice Questions (correct answers in bold):
1. What is the most likely cause of Max's clinical presentation and abnormal clinical chemistry values?
a) Pancreatitis
b) Diabetes mellitus
c) Hepatitis and liver failure
d) Acute kidney injury
2. Which of the following enzymes is primarily associated with hepatobiliary disease in dogs?
a) Creatine kinase (CK)
b) Amylase
c) Alanine aminotransferase (ALT)
d) Lipase
3. Which parameter is likely to be decreased in a dog with severe liver failure?
a) Total bilirubin
b) Alanine aminotransferase (ALT)
c) Blood urea nitrogen (BUN)
d) Albumin
When I saw this output, frankly I was blown away. This is a well-written and concise case vignette complete with accurate medical details and labwork, and the multiple choice questions it asks are a solid start. There are some things I’d take issue with—for example, in #3, BUN can also be decreased in liver failure—but this is an incredible result from software that was not even specifically trained on veterinary medical content!
Another way I could see using ChatGPT in this way would be to flip the script and ask the students to generate case scenarios and questions, then have them analyze and critique the accuracy of the output, followed by revisions to make it better! There is a good chance that this more active and creative educational activity will lead to better retention than simply staring at PowerPoint notes.
Deconstructing Medical Lingo for Novices
There is a steep learning curve for vet students and new grads to understand the baroque language of medical specialty reports, especially from pathologists and radiologists. ChatGPT can act as a tutor to help students translate these reports into a more understandable format:
Prompt:
For teaching purposes (not my own medical information), what would the pathologist's diagnostic interpretation be for the following lymph node morphologic description? "Most of the nucleated cells are lymphoid and consist of approximately 70-80% small "mature" lymphocytes, with mildly to occasionally moderately increased intermediate to large lymphocytes. Plasma cells are moderately expanded. There are a few scattered bland macrophages. In a few fields, non-degenerate neutrophils and eosinophils may be slightly increased relative to blood. No obvious infectious organisms or overtly neoplastic cells are observed."
Output:
The provided lymph node morphologic description suggests a reactive lymphadenopathy, which is a benign and usually reversible enlargement of lymph nodes in response to an immune response or inflammation. Here's the pathologist's diagnostic interpretation for the description:
Diagnostic Interpretation: The lymph node exhibits features consistent with a reactive lymphadenopathy. Most of the nucleated cells are lymphoid in origin and primarily composed of approximately 70-80% small "mature" lymphocytes. These small lymphocytes are characteristic of a normal or reactive lymph node population. Additionally, there is a mild to occasional moderate increase in intermediate to large lymphocytes, which can be indicative of a response to an antigenic stimulus.
The expansion of plasma cells is moderately noted, which further supports the reactive nature of the lymph node. Plasma cells are mature B lymphocytes that produce antibodies and can be stimulated in response to infections or other immune challenges.
The presence of a few scattered bland macrophages is also consistent with a reactive process. Macrophages are involved in the phagocytosis of cellular debris and antigens, and their presence suggests the clearance of immune response products.
Non-degenerate neutrophils and eosinophils are slightly increased relative to blood in a few fields. This finding could be a reflection of ongoing inflammation or immune response.
Overall, there is no evidence of obvious infectious organisms or overtly neoplastic (cancerous) cells. The morphologic features described suggest an appropriate immune response to an antigenic stimulus rather than a pathological condition. This reactive lymphadenopathy is likely a response to an infection, inflammation, or immune challenge.
It's important to note that this interpretation is based solely on the provided description and should be taken as a general guideline. Actual diagnoses should be made by a qualified pathologist or medical professional who has access to the patient's complete medical history, clinical context, and any additional laboratory data.
The ChatGPT interpretation of that lymph node morphologic description (taken from a recent real report of mine) is longer and wordier than the comments I typically write, but the content and meaning is very similar. It also goes further and explains some of the basic pathophysiology suitable for someone learning this for the first time in a course.
Helping Students and Junior Residents Draft Report Templates
Along with having difficulty interpreting those reports, writing them from scratch can be even harder. Using ChatGPT to create a basic outline could allow folks early in their career to get a huge jump-start on the learning process. For example…
Prompt:
Write a pathology report for a canine mast cell tumor.
Output:
Pathology Report
Patient Information:
Patient Name: [Canine Name]
Breed: [Breed]
Age: [Age]
Sex: [Male/Female]
Date of Biopsy: [Date]
Referring Veterinarian: [Veterinarian Name]
Clinical History: [Brief summary of clinical presentation and history]
Gross Description: Received for histopathological examination is a formalin-fixed, paraffin-embedded tissue biopsy from a canine mast cell tumor. The specimen consists of a single irregularly shaped, firm, pale-tan tissue fragment measuring [dimensions in cm]. The tissue is submitted in one cassette, labeled [Cassette Number].
Microscopic Description: Histologic sections of the tissue reveal a moderately cellular neoplasm composed of pleomorphic round to spindle-shaped cells arranged in sheets, cords, and clusters. The neoplastic cells exhibit varying degrees of anisocytosis and anisokaryosis. There is a prominent infiltrate of eosinophils within and around the neoplastic cells. Additionally, scattered mast cells with characteristic metachromatic granules are identified throughout the tumor.
Mitotic Index: [Number of mitotic figures per 10 high-power fields (HPF)].
Histologic Grade: [Low, Intermediate, High], based on cellular and nuclear features, mitotic activity, and presence of necrosis.
Tumor Margins: Tumor margins are [Clear/Close/Involved] [Specify Margins]. Tumor invasion into adjacent tissues is [Present/Absent].
Immunohistochemistry: Immunohistochemical staining is performed to assess the expression of various markers, including c-Kit (CD117) and tryptase. Results show strong positive staining for c-Kit (CD117) in the neoplastic cells, confirming their mast cell lineage. Tryptase staining is also positive, further supporting the diagnosis.
Diagnosis: Canine Mast Cell Tumor (Grade [Grade]), [Anatomic Location]
Tumor Size: [Size in cm]
Tumor Margins: [Clear/Close/Involved]
Mitotic Index: [Number of mitotic figures per 10 HPF]
Invasion into Adjacent Tissues: [Present/Absent]
Discussion: Canine mast cell tumors are one of the most common cutaneous neoplasms in dogs. They arise from mast cells, which are normal components of the immune system. These tumors can exhibit a wide range of biological behaviors, from benign to locally invasive to metastatic. The histologic grade of the tumor, as well as the presence of certain molecular markers, can provide valuable prognostic information.
Prognosis: The prognosis for this case is influenced by factors such as the histologic grade of the tumor, presence of mitotic activity, and extent of tumor invasion. Close follow-up and potential further diagnostic evaluations are recommended to monitor for local recurrence or metastasis.
Recommendations:
Consideration of wide surgical excision with appropriate margins.
Monitoring for local recurrence and metastasis through regular physical examinations and imaging studies.
Consultation with an oncologist to discuss potential adjuvant therapies, if indicated.
Client education regarding the importance of ongoing surveillance and prompt reporting of any changes.
This report is based on microscopic examination and immunohistochemical analysis of the submitted tissue. It is intended to provide diagnostic and prognostic information to the referring veterinarian. Further clinical correlation and additional testing may be warranted.
Again, this isn’t exactly how I would write such a report, and there are some things I’d tweak, but this is a very good first draft. It contains many of the key variables in MCT grading schemes, and provides a progressive example of a synoptic pathology report. Not bad for a few seconds from a computer!
Creating Study Aids
There are so many things to learn in med school that students spend a lot of time trying to figure out memorization tricks. ChatGPT makes this a snap, and in seconds can create a ton of novel mnemonic devices on command:
Prompt:
Can you come up with a med school mnemonic study aid to remember the causes of hypercalcemia?
Output:
Of course! Here's a mnemonic study aid to remember the causes of hypercalcemia for medical school:
"MACHO-CaL"
M - Malignancy
A - Addison's Disease
C - Calcium Supplements (Excessive)
H - Hyperparathyroidism (Primary or Secondary)
O - Other (Granulomatous diseases, e.g., Sarcoidosis)
Ca - Chronic Kidney Disease
L - Lithium (medication-induced hypercalcemia)
Remember, mnemonics are meant to help you recall information easily. Feel free to modify this mnemonic to better suit your learning style or to include additional details as needed.
Prompt:
Oh, cool! Can you create three more unique ones?
Output:
Absolutely, here are three more mnemonic study aids to help you remember the causes of hypercalcemia:
"CALCIUM DAZE"
C - Calcium Supplements (Excessive) A - Addison's Disease L - Lithium (medication-induced hypercalcemia) C - Chronic Kidney Disease I - Immobilization (Prolonged) U - Ultra-High Vitamin D (Intoxication) M - Malignancy D - Diuretics (Thiazides) A - Acromegaly Z - Zollinger-Ellison Syndrome E - Excessive Vitamin A (Hypervitaminosis A)
"HYPERCAL"
H - Hyperparathyroidism (Primary or Secondary) Y - Yamaguchi Syndrome (Familial Hypocalciuric Hypercalcemia) P - Paget's Disease of Bone E - Excess Vitamin D (Intoxication) R - Renal Failure (Chronic Kidney Disease) C - Cancer (Malignancy) A - Addison's Disease L - Lithium (medication-induced hypercalcemia)
"CALC-HOPE"
C - Calcium Supplements (Excessive) A - Acromegaly L - Lithium (medication-induced hypercalcemia) C - Chronic Kidney Disease H - Hyperparathyroidism (Primary or Secondary) O - Other (Granulomatous diseases, e.g., Sarcoidosis) P - Paget's Disease of Bone E - Excess Vitamin D (Intoxication)
Feel free to use or modify these mnemonics as needed to aid in your study of hypercalcemia causes. Remember, the key is to create associations that are memorable for you personally.
These are impressive not only because of their breadth, but because the acronyms themselves link back to the core idea of diseases that cause elevated calcium. Many of the mnemonics I learned in vet school have nothing to do with the actual concepts to memorize, which makes it trickier to retain.
Conclusion
The examples above represent a tiny fraction of ideas for implementing generative AI into the classroom! There are so many ways it can be used to enhance, rather than hurt, student learning, and we’ve likely only scratched the surface of what is possible. In the coming months and years, there will likely be more competitor AI products and apps that integrate with them to add even more functionality.
I recommend sitting down with ChatGPT or another LLM tool and simply start playing around with it. Teachers, experiment with how it might help you create your next lecture or quiz. Students, see how it might turbocharge your education!
How have you used or seen others use AI in the classroom? Leave a comment below 👇
All examples in this article are based on the free version of GPT-3.5. Presumably, the newer and much larger GPT-4 model that requires paid access is even more powerful.