Xavier Amatriain, Cofounder & CTO, Curai
summary
Healthcare systems still give clinicians 15 minutes per visit, and medical errors cause about 400,000 deaths annually, so the approach combines ML models (RNNs, autoencoders, deep vision), knowledge bases derived from expert encodings (2–4 weeks and 50–250 papers per disease), and EHR/NLP pipelines to map text to ontologies. A stealth startup called Curai builds a patient-facing conversational system, uses GANs for synthetic EHRs to augment privacy-safe training data, and faces challenges in causality, error-cost tradeoffs, and merging expert and learned models.
AI-written summary · about
chapters (29)
- 0:00Intro
- 0:17Issues with medicine
- 1:55Medical diagnosis
- 2:53Healthcare is a disaster
- 3:41Improving medical care
- 4:13The ideal AI medical system
- 4:38What is medical decision support
- 5:03Medical research
- 6:40Source of data
- 7:46Machine learning AI
- 10:04Conversational medicine
- 10:43Ontology
- 11:28NLP
- 12:33Knowledge vault
- 12:54Data
- 15:15Vision
- 16:19Machine learning
- 16:47Personalization
- 17:21Precision medicine
- 18:19Personalized medicine
- 19:02Building personalized systems
- 19:34Health recommendation systems
- 19:45What is different
- 19:59Selfdriving car paradigm
- 22:53About Curai
- 23:18About the team
- 24:23Challenges
- 25:43References
- 26:23Hiring
related talks
25mAI for healthcare: Scaling Access and Quality of Care for Everyone
28mThe Future of Healthcare – AI’s Promise and the Challenges Ahead
29mAI for COVID-19: An online virtual care approach
45mGenerative AI in medicine: Improving patient & provider experiences
05mHow generative AI can support medical staff
09mHow AI is changing the healthcare system
59mArtificial intelligence in Healthcare
36mResponsible AI