Dr. Scott Campbell holds degrees from University of Michigan, UCLA, and Columbia University. He grew up in La Jolla, California, excelled academically and athletically, played baseball at Michigan, then pursued a Master’s in Public Health from UCLA before attending Columbia Medical School. His medical career spanned internal medicine, transplant, critical care, and emergency medicine. Recently, he became chief AI officer at Zero Hour Medical, advising organizations including NASA on healthcare AI transformation strategies.
00:05:00Campbell describes his journey from college ambitions through emergency medicine at Kaiser San Francisco’s Permanente Medical Group to focusing on machine learning and AI. He became involved with a Children’s Hospital Orange County group working to establish an American Board of Artificial Intelligence and Medicine. He transitioned from clinical practice to full-time AI work, joining an advisory board and teaching classes to increase AI literacy among physicians, emphasizing the importance of camaraderie and dreaming big throughout his career.
00:10:00Campbell reflects on his path from medical school through successful Northern California practice, highlighting his current enthusiasm about AI’s potential in clinical settings. He emphasizes how positive relationships and camaraderie with emergency medicine colleagues influenced his specialization choice and helped prevent burnout during his career, crediting these professional connections with sustaining his long-term engagement in medicine.
00:15:00Campbell discusses the significance of positive workplace culture in medicine, drawing from his experiences at TPMG and his emergency department where partner selection created harmonious environments focused on patient welfare. The San Francisco Emergency Physicians Association, founded in 1987, provided a collaborative platform enabling emergency physicians across different hospitals to communicate, share experiences, and build unity within the profession through regular interaction and problem-solving.
00:20:00As a long-term president of San Francisco’s emergency medicine community, Campbell addresses significant challenges including overcrowding and extended patient waits. San Francisco implemented a sobering center for intoxicated individuals, reducing emergency department bottlenecks. This initiative, among the first nationally, reportedly freed approximately 60,000 emergency department visits and 6,000 ambulance trips over two decades, with only one documented fatality from the program.
00:25:00Campbell describes creative solutions addressing emergency department overcrowding from behavioral health patients, including a “Dore Center” for individuals likely to receive 5150 psychiatric holds and a medical respite center for homeless individuals. While not entirely successful, these capacity carve-outs demonstrate the value of creative approaches. Campbell emphasizes using AI to identify patients better served outside emergency departments, noting that behavioral health comprises 25-30% of emergency department cases, requiring both human and technological interventions.
00:30:00Campbell explores hospital bed competition and AI’s potential for optimizing bed usage based on real-time data. He notes that current prioritization based on procedure urgency can leave pneumonia patients waiting extended periods. He advocates for stretching hospital operations beyond traditional hours and emphasizes collaboration with administrators. Campbell stresses the importance of real-time situational awareness, referencing pandemic mistakes when data became outdated, highlighting how better information supports improved decision-making.
00:35:00Campbell traces his AI journey from graduate school interests in predictions and data analysis through his medical career, influenced by management consulting background and siloed data structures. He highlights cutting-edge AI applications beyond natural language processing, including image recognition in pathology and radiology, robotics, and robotic process automation. Campbell addresses primary care challenges with patient communication, proposing AI-based solutions for managing clinical inboxes more effectively through intelligent filtering.
00:40:00Campbell discusses AI applications for managing clinician messages and predicting clinical outcomes through process automation, spam filters, and sentiment analysis. He identifies machine learning as the most significant AI application, enabling prediction of numbers, events, or classifications like cancer diagnosis. Campbell cautions that large language models require expensive training and monitoring, presenting resource allocation challenges for healthcare organizations implementing and maintaining these systems effectively.
00:45:00Campbell observes that AI projects increasingly resemble startups, with cultural shifts toward risk-taking potentially creating capital expenditure challenges over the next 5-10 years. He advocates prioritizing simple clinical decision support models as more promising investment areas than large language models, emphasizing the high frequency of healthcare decision-making. Campbell envisions training 16,000 physicians on AI resources while identifying AI talent and incubating small groups to build a clinical decision support network focused on simplicity.
00:50:00Campbell explains medical group AI implementation strategy, emphasizing mastery of logistic regression and classifier building to advance the field. He advocates for healthcare system leadership education regarding AI and the shift from revenue-based to value-based medicine models. Campbell notes large medical groups currently buy or experiment with AI rather than funding development internally due to risk aversion and talent shortages. He encourages clinicians to develop AI fluency, leveraging their domain expertise alongside technological capability, mentioning venture capital and single family offices as significant funding sources.
00:55:00Campbell discusses current AI startup investments and private equity’s challenges funding long-term, unpredictable AI projects. He suggests private equity groups should partner with AI companies earlier in their development cycles. Campbell shares his NASA experience, where AI optimizes astronaut healthcare in space, predicts potential health issues, and enables diagnosis and treatment with limited resources, demonstrating AI’s transformative applications beyond terrestrial medicine.
01:00:00Campbell emphasizes Knowledge Graph databases’ superiority over relational systems, describing them as 10-100 times more powerful through entity relationship leverage. He highlights synthetic data and digital patient twins for addressing limited cases and improving model building. Campbell explains federated learning’s collaborative machine learning approach, enabling multiple hospitals to work on shared problems without centralizing data, reducing time and costs while preserving privacy in healthcare settings.
01:05:00Campbell discusses federated learning, synthetic data, and large language models’ healthcare potential, explaining how federated learning processes data locally without central server transfer, ensuring privacy. He references NASA’s use of synthetic data and federated learning for predicting astronaut health issues, refining models through Earth-based predictions refined in space. Healthcare applications include addressing limited data issues and improving predictions like identifying kidney stone development likelihood. Campbell emphasizes domain experts, data scientists, business developers, and AI engineers collaborating as essential, comparing effective AI development to “building a rock band.”
01:10:00Campbell shares a cautionary EMR company example that lost over a billion dollars on sepsis prediction by editing model features around inaccurately calculated respiratory rates and training on limited regional hospitals. When implemented elsewhere, the model failed. Campbell emphasizes local data importance and domain expert involvement, introducing federated learning where AI systems share data across regions without compromising sensitivity. This approach improves accuracy and effectiveness in healthcare, drug discovery, and other fields.
01:15:00Campbell discusses federated learning’s precision medicine potential on personal devices and population health applications. He references congestive heart failure patients in federated models enabling personalized predictions and treatment plans. Campbell introduces “precision population health” concepts, where large datasets enable novel population groupings and discoveries. He emphasizes continuing AI healthcare advancement momentum without distraction or risking an “AI winter” period of reduced innovation and investment.