AI Champions Before AI Tools: Teaching Medicine to Build Without Code
Recommend the wrong movie and no harm is done. Lead a clinician down the wrong path and “you could hurt people or kill people.” Dr. Hooman Rashidi, Associate Dean of AI in Medicine at the University of Pittsburgh School of Medicine and a practising hematopathologist, argues that health systems need AI champions before they need more AI tools. He explains why his team runs a hybrid strategy (“do you actually need a 10-trillion-parameter model to write a letter of recommendation? Usually no”), how Pitt-GPT+ treats a custom chatbot like a laboratory-developed test, and how Field Care GPT runs fully off-grid on a laptop for combat medicine. The heart of the episode is a patient-safety warning: in tumor boards, we question every physician’s limitations, but nobody questions the AI’s output. His answer is no-code AI education, so every clinician can ask about sensitivity, specificity and F1, and push back on a vendor pitch. Co-hosts DR GPT™ Harvey Castro and Edward Marx bring the cost and governance questions, and the conversation closes on keeping the human in the loop, and engaged.
What you'll discover
- wrong movie vs. wrong patient (Dr. Hooman Rashidi)
- from electronic medical records to AI medical records (Junaid)
- From bioinformatics to medicine: “nobody cared about machine learning” (Dr. Hooman Rashidi)
- Still on service: why practising medicine keeps AI honest (Dr. Hooman Rashidi)
- Do you need a 10-trillion-parameter model? The hybrid strategy (Dr. Hooman Rashidi)
- Pitt-GPT+: a private chatbot built like a lab test (Dr. Hooman Rashidi)
- Field Care GPT: AI for combat medicine with no Wi-Fi (Dr. Hooman Rashidi)
- Edge AI: big models condensed onto your phone (DR GPT™ Harvey Castro)
- Will AI raise hospital costs? Learn from the 5% that got it right (DR GPT™ Harvey Castro)
- AI literacy is a patient-safety issue (Dr. Hooman Rashidi)
- Tumor boards: nobody questions the AI (Dr. Hooman Rashidi)
- AI champions who can push back on a vendor pitch (Dr. Hooman Rashidi)
- Who is responsible when AI gets it wrong? Educate first, then govern (Dr. Hooman Rashidi)
- Ed Marx joins: governance boards that say no because they don’t know (Ed Marx)
- Keep the human in the loop, and keep the human engaged (Dr. Hooman Rashidi)
- Are new doctors losing the skills AI now does for them? (Junaid)
- Put yourself in your patients’ shoes: AI is not magic (Dr. Hooman Rashidi)
- Thank you and close
Referenced in the episode
- Pitt-AI-cademy: no-code, conceptual AI education for medicine and the health sciences: https://ai-cademy.hs.pitt.edu
- Pitt-GPT+: a custom, private chatbot-generator framework, validated like a laboratory-developed test (negative and positive controls, model cards): https://ai-cademy.hs.pitt.edu/
- Field Care GPT (FCGPT): a fully off-grid derivative of Pitt-GPT+ for combat and disaster care, built with Dr. Nam Tran, Dr. Ron Poropatich, trauma surgeons and ED physicians: https://www.path.pitt.edu/people/hooman-h-rashidi-md-ms-fcap
- CPACE, the Computational Pathology & AI Center of Excellence (Pitt and UPMC): https://cpace.pitt.edu/
- STNG (Synthetic Tabular Neural Generator): Rashidi HH, Albahra S, Rubin BP, Hu B. “A novel and fully automated platform for synthetic tabular data generation and validation.” Scientific Reports 2024;14:23312: https://www.nature.com/articles/s41598-024-73608-0
- Nebulon-GPT, CPACE’s open-source, locally hosted generative-AI platform: https://github.com/pitt-cpace/NebulonGPT
- Hybrid AI strategy: vendor partnerships plus home-built, open-source frameworks when the ROI isn’t there
- The widely quoted statistic that only about 5% of AI deployments show real return on investment
- Tumor boards and multidisciplinary conferences; sensitivity, specificity and F1 as the questions to ask of a predictive model
- LM Studio (local LLM app), referenced as a comparison for Pitt-GPT+
- Dr. Rashidi’s faculty profile: https://www.path.pitt.edu/people/hooman-h-rashidi-md-ms-fcap
Medical disclaimer. This podcast is for general informational purposes and is not medical advice. Listening does not create a doctor and patient relationship. Views expressed by the guests and co-hosts are their own and do not represent their employers. No specific vendor or product is endorsed. No PHI is shared.