What's Your Problem? Teaching Computers to Smell
15 snips
May 22, 2025 
Alex Wiltschko, co-founder and CEO of Osmo, started out as an AI researcher at Google before combining his childhood passion for perfume with cutting-edge technology. He discusses how Osmo is teaching computers to smell, with potential applications in diagnosing diseases and enhancing security. The conversation also delves into the exciting fusion of AI and fragrance creation, highlighting the innovative perfume line, Generation. Wiltschko shares personal scent experiences and the emotional intricacies linked to our olfactory preferences.
AI Snips
Chapters
Transcript
Episode notes
AI Predicts Smell from Molecules
- AI models trained on curated fragrance ingredient datasets can predict what molecules smell like.
- Professional perfumers' scent ratings form a crucial, specialized dataset for training AI in smell prediction.
Validate AI Models with Human Panels
- Validate AI smell models by comparing predictions with human scent perceptions on new molecules.
- Achieving accuracy better than an average person proves a significant AI breakthrough in scent prediction.
Challenge of Identifying Real-World Molecules
- Identifying molecules in real-world smells is complex and involves chemical sensing challenges.
- Automated detection of molecule composition in air is essential for digitizing and replaying scents.