TopMedTalk AI: Generative AI Basics, Biases and the Future

At the World Congress of Anaesthesiologists (WCA) in Marrakech, Morocco, TopMedTalk co-editors In chief, Kate Leslie and Mike Grocott speak with Sydney anaesthesiologist Alwin Chuan, a member of the joint ESRA-ASRA working group on AI, about strengths and limitations of AI for regional anaesthesia and medicine. Chuan explains how generative AI is trained on vast datasets using transformer architectures, why outputs are probabilistic, and how context affects meaning. They discuss privacy and the possibility of personal or creative work being used as a commodity in training, internet "scraping," and the inevitability of hallucinations, including false facts and fabricated citations. The conversation covers bias in training corpora, the need for human curation and reinforcement learning, and mitigation via prompt engineering such as retrieval-augmented generation (RAG), citation/page verification, and confidence estimates. Chuan predicts future advances and the fusion of imaging AI with generative AI for ultrasound guidance, previewing a follow-up episode. -- Join us at Evidence Based Perioperative Medicine (EBPOM) World Congress 2026 in London. Be part of a global conversation as clinicians from around the world gather between 7-9th July at the British Library in London. Three days of evidence-based perioperative medicine, global insights, and expert debate—featuring speakers including Michael Marmot and Ken Rockwood. Register here - EBPOM World Congress 2026