Startup Osmo Claims AI Smell Model Matches Human Olfactory Intelligence
· GlobeWire Daily Editorial Desk · Source: www.technologyreview.com

Image: MIT Technology Review · Photo source · Resized and letterboxed to 16:9; no AI generation
An artificial intelligence startup named Osmo reports that its newly developed foundation model for smell has achieved human-level olfactory intelligence, according to founder Alex Wiltschko. The company states that the technology is capable of detecting and discriminating between smells comparably to an average person, and designing scents on par with a junior perfume designer.
Building the Scent Database
- Training Data: Osmo created a proprietary dataset featuring more than 10 million human labels and 800 million scent tokens, describing exact molecules and concentrations for various odors.
- Data Collection: The startup utilizes spectrometers and chemical sensors running continuously alongside teams of trained human data labelers in the US and internationally.
- Origins: Founder Alex Wiltschko spun Osmo out of Google Brain in 2023 to address the historical lack of large-scale olfactory data.
Industry Applications and Testing
Osmo reports that its technology has been utilized within the fragrance industry to assist brands in designing and manufacturing new scents. The company states that its AI model outperformed in-house perfumers at a major consumer goods company and has worked with Procter & Gamble. Additionally, the Bill & Melinda Gates Foundation awarded Osmo a grant to assess molecules for repelling disease-carrying insects, leading to the identification of eight substances reported to be more repellent than DEET. The firm has also collaborated with StockX to explore identifying counterfeit sneakers by analyzing material chemical compositions.
Critiques and Limitations
Independent experts have raised questions regarding the technology:
- Skepticism: Smell scientist Avery Gilbert has criticized Osmo's marketing claims and patent filings on his Substack, noting they still rely on human sensory panels.
- Context Challenges: Michael Schmuker, a former neural computation professor who has consulted for Osmo, points out that human olfaction relies heavily on context. For example, the same molecule can evoke very different associations depending on the environment, such as the shared chemical properties between certain cheeses and used diapers.
- Future Goals: While Wiltschko envisions potential future applications in disease diagnosis and airport threat detection, experts note that sensor technology and models require further advancement to handle such complex tasks securely.
Readers can verify official announcements and learn more about scent-detection research directly through Osmo's published materials or independent scientific literature.
AI-assisted reporting; source claims are attributed and have not been independently verified.



