Episode
How AI Learns to Smell with Alex Wiltschko - #771
- Published
- Jul 8, 2026
- Duration seconds
- 3595
- Processing state
processed- Canonical source
- https://twimlai.com/podcast/twimlai/how-ai-learns-smell
Actions
POST https://stenobird.com/v1/public/podcasts/twiml-ai-podcast/episodes/how-ai-learns-to-smell-with-alex-wiltschko-771/transcription-requests
Idempotently request low-priority transcript generation for this episode.GET https://stenobird.com/podcast/twiml-ai-podcast/how-ai-learns-to-smell-with-alex-wiltschko-771.md
Read the agent-friendly Markdown representation of this episode resource.
Summary
Alex Wilchko, CEO of Osmo, explains how to bridge the gap between chemical structures and digital perception. The discussion focuses on building the 'map' required to digitize smell through graph neural networks and multi-dimensional embedding spaces.
Topics
- Olfactory Intelligence
- Graph Neural Networks
- Molecular Modeling
- Chemical Sensors
- Embedding Spaces
- Chemoinformatics
- Foundation Models
- Digital Scent
Highlights
- Main idea: Digitizing smell requires a 'map' that translates molecular structures into a digital embedding space, similar to how RGB works for vision
- Technical approach: Graph neural networks are used to process molecular macrostructures and propagate information into a 300-dimensional principle odor map
- Practical takeaway: Building a self-perpetuating data loop involves using chemical sensors to feed proprietary datasets that train predictive models
- Failure mode: The 'structure-odor relation problem' remained unsolved for a century because of the difficulty in mapping complex chemical inputs to human perception
- Future potential: Olfactory intelligence could enable early disease detection, emotion sensing, and the creation of entirely new synthetic fragrances
Chapters
1:00The Chemistry of Intelligence: The importance of training AI on the chemical language used by non-human species like bacteria and fungi.5:00The Biological Basis of Smell: Understanding olfactory sensory neurons and the genetic encoding of scent receptors.10:00Solving the Structure-Odor Problem: How neural networks can finally bridge the gap between molecular structure and perceived odor.14:00Graph Neural Networks in Olfaction: Using GNNs to process molecular graphs and capture multi-dimensional scent characteristics.19:00Building Proprietary Datasets: The process of enumerating molecules and creating large-scale olfactory datasets.23:00Aligning Sensors with Human Perception: Using chemical sensors to create data that can be mapped to human sensory labels.28:00The Self-Perpetuating Data Loop: How fragrance design serves as both a business model and a source of continuous training data.