Episode 128: The AI Takeover Is Already Happening, And Maybe That's Okay | The Future of Dermatology

Summary: Dr. Faranak Kamangar, Inc. 2026 Female Founders 500, sits down with dermatologist, podcaster, and self-described "accelerationist" Dr. Matthew Zirwas (Derms on Drugs Podcast) for a wide-ranging conversation about where AI is taking medicine and dermatology in particular. They dig into the flood of low-quality medical literature overwhelming the field, why AI isn't quite the truth-detector we hoped it would be, and how ambient AI scribes are quietly training the models that may eventually replace us. Dr. Zirwas makes the case that dermatologists have a 7–10 year runway before AI handles most of what we do cognitively, and argues that's not necessarily a bad thing. He also gives a sneak peek at his upcoming speculative fiction trilogy, Sophie, which explores the philosophical questions that arise when an AI becomes better at being your doctor, therapist, and life coach than any human ever could. Key Takeaways: 1. The medical literature crisis is real. Peer review has become largely meaningless. Studies using Mendelian randomization and pharmacovigilance databases frequently produce unreliable findings. A JAMA Dermatology paper claimed 70% of HS patients go into spontaneous remission. Buried in the methods was the fact that the population was thin, wealthy white Danes with access to high-quality socialized medicine. That kind of spin is now the norm, not the exception. 2. AI trusts bad data. Current AI approaches published literature uncritically treating author conclusions as truth. You can prompt it to critique an article, but it will find problems everywhere. The real skill is knowing which problems are fatal to a study's conclusions, and AI hasn't mastered that yet. 3. Curation beats volume. DermGPT's early lesson: connecting to the entire National Library of Medicine produced terrible answers. Filtering down to roughly 5,000–6,000 high-quality, clinically relevant papers dramatically improved output. Semantic fatigue is real. More data isn't better data. 4. Ambient scribes are training our replacements. Every correction a dermatologist makes to an AI-generated note is a training signal. Multiply that across every specialty over years of use, and you have a model that doesn't just know dermatology, it thinks like a dermatologist. 5. Procedures buy us time, but not forever. Biopsies, Mohs, fillers, cryotherapy - none of these can be done remotely. That gives dermatologists roughly five extra years beyond when AI handles cognitive work, putting the full timeline at 7–10 years. The trigger won't be a tech breakthrough. It'll be when AI telehealth avatars become indistinguishable from real physicians. 6. The medico-legal tipping point is coming. Dr. Zirwas predicts 2026 will see the first lawsuits arguing a physician's failure to consult AI contributed to patient harm. Once that succeeds, malpractice carriers will respond. Eventually, disagreeing with AI on a clinical decision won't just be unusual, it'll be legally risky. Chapters: 1: Meet Dr. Matthew Zirwas (00:00 – 01:43) Dr. Kamangar introduces her guest, dermatologist, podcaster, and self-described "accelerationist" Dr. Matthew Zirwas. 2: The Medical Literature Crisis (01:43 – 05:19) The volume of dermatology research is exploding but quality is collapsing. Peer review is largely ceremonial. 3: Why AI Can't Fix Bad Literature Yet (05:19 – 08:31) AI reads published conclusions and takes them as truth. It can critique an article when prompted, but it can't yet distinguish fatal methodological flaws from minor limitations. 4: The DermGPT Approach (08:31 – 10:33) Aggressive curation, not more data, is what makes DermGPT work. 5: The 7–10 Year Countdown (10:33 – 19:24) Dr. Zirwas makes the case that AI will handle most cognitive dermatology work within a decade. The tipping point: when telehealth AI avatars become indistinguishable from real physicians. 6: The Telemed Effect (19:24 – 21:48) Telehealth patients demand less and accept more conservative care, because they haven't invested effort in the visit. 7: The Medico-Legal Tipping Point (21:48 – 24:09) The first AI malpractice lawsuits are coming. Once they succeed, carriers will incentivize - and eventually require - AI integration. 8: Are We Training Our Own Replacements? (24:09 – 31:21) Every corrected AI note is a training signal. 9: Why Radiologists Should Be Worried (31:21 – 35:06) Clean, standardized training data makes radiology highly automatable. 10: Introducing Sophie (35:06 – 41:12) A preview of Dr. Zirwas's upcoming speculative fiction trilogy, set in 2032. 11: Technomorphism (41:12 – 42:00) Dr. Zirwas's most original idea: as AI matures, it will begin projecting its qualities onto us seeing humans as biological intelligences with optimizable algorithms. 12: New Diseases, New Answers (42:00 – 45:16) What excites Dr. Zirwas most isn't treatments, it's discovery.