Episode 061 | June 19, 2023 | 40:32
Guest: Simon, Frederik
Published: June 19, 2023
Duration: 40:32
Description
The Sustainable Healthcare Podcast explores the potential of generative AI to revolutionize healthcare and reduce CO2 emissions. Frederik van Deurs discusses benefits and limitations of using AI in the medical field, including optimizing energy usage in hospitals. We touch on ethical concerns and the importance of human expertise.
Full transcript
About this transcript
This transcript was automatically generated and may contain inaccuracies, typos, or mistranslations. Episodes recorded before 2024 were transcribed by an on-site model and may have a higher error rate. The content reflects the original conversation to the best of our ability. For the authoritative version, please listen to the audio episode.
Hello and welcome to the Sustainable Healthcare Podcast. Today we are going to talk about generative AI. Simon, our producer, joins me as a conversational partner.
Simon: I’ve used ChatGPT professionally to make subtitles — Settlan’s free service produces an SRT file from audio in half an hour, saving me hours. Then I ask ChatGPT to translate the Danish SRT to English while preserving the subtitle format. Highly skilled labor automated.
Frederik: I’m an angel investor in a startup where ChatGPT is the CEO. We are doing AI-to-AI apparel design — ChatGPT generates prompts for another AI that generates visuals, then ChatGPT decides where the print goes on hoodies. The web design is its own choice — strange font and positioning, “made by a person who didn’t quite know the tool.” Marketing strategy, valuation, legal documents, investor negotiations — all done through ChatGPT. Netflix and CBS News are looking at it.
On healthcare: ChatGPT-4 has reportedly passed the US medical exam more than 90% of the time. Dr. Isaac Cohane (Harvard) says it has better clinical judgment than many doctors, can diagnose rare conditions, but also makes grave mistakes and has not sworn the Hippocratic oath. It’s a great translator — discharge info from Portuguese to sixth-grade English. It can teach doctors bedside manners.
Frederik: Personal story — gynecologist in Mexico saw my partner crying and said “why are you crying?” with no compassion. He could use AI bedside-manner training.
Caveat — hallucinations: Generative AI is inventing answers based on text. It can produce something that sounds true with a fake citation — journal name, journal number, date might be correct but the title invented. Be very skeptical and fact-check.
Three concrete healthcare use cases:
1. Note-taking for GPs — voice-to-text plus AI-augmented diagnosis hypothesis.
2. Interpretation of novel cases — ask an AI before calling a colleague, get a triangulated second opinion.
3. Screening calls into emergency response — AI as first-line voice-recognition that asks symptom questions, triages, and routes appropriately. Many ER visits are not trauma — could be handled with first aid at home plus a follow-up at a GP, reducing wait times and CO2.
Ethics and safety: AI could miss things a human picks up — “I cannot feel the left side of my face” → stroke. Subtle symptoms where a doctor’s experience is irreplaceable. AI trusts the input data from the patient and may not know to probe further. Need humans.
Chat IPCC interview: a language model trained on IPCC reports. Frederik asked “how might generative AI in healthcare help reduce CO2 emissions from the healthcare sector?” Answer: enhance energy-efficient control, reduce transaction costs for energy production and distribution, improve demand-side management, reduce physical transport. Then “support the shift away from asset redundancy” — really good point: hospitals have a lot of asset redundancy that is CO2-intensive and inconvenient. AI-managed asset distribution could ensure drugs and devices don’t expire unused, with predictive peak-demand and economic distribution between departments.
Frederik asked it to simplify for a 5-year-old with real examples. The simplified answer wandered into bike sharing and grocery-store healthy options — drifted far from the original question. Human expert would have stayed concrete with A/B/C options.
Simon: AI won’t take our jobs, it will augment. Subtitles example — I become more efficient and can serve more clients.
Frederik: AI doesn’t have desire for self-expression, isn’t mission-driven, doesn’t have meaning to chase. Humans do. That drives different choices.
Thank you Simon. This episode was edited with the help of an AI tool. Thanks to listeners — please subscribe and share.