[ netdynamic // tech news ]

Osmo’s AI Nose Aims to Rival Human Smell

In a laboratory near Newark Airport in New Jersey, a team of roughly experienced human sniffers is methodically working through odors from around the world. Their careful notes, describing everything from nutty to peachy to marine, are cross-referenced with chemical analysis to build a rich training dataset for Osmo, an AI startup that calls its project the world’s first foundation model for smell. Founder Alex Wiltschko says the goal is to give computers a sense of smell. He argues that scent molecules are effectively the only language most species on Earth can speak, since many animals cannot communicate through sound or sight. Smell is also deeply tied to memory and emotion, he notes, because a single scent can instantly transport someone back decades.

Wiltschko claims that Osmo’s model has reached what the company defines as human-level olfactory intelligence. In practical terms, the system performs about as well as the average person at detecting and distinguishing smells, and roughly on par with a junior perfume designer at creating them. The technology is already being used in the fragrance industry to help brands develop and manufacture new scents. According to the company, its model recently outperformed in-house perfumers at one of the world’s largest consumer goods companies. Wiltschko suggests that future applications could include threat detection at airports or even disease diagnosis.

The startup grew out of Wiltschko’s work at Google Brain and follows a familiar AI strategy: gather large volumes of data and see what emerges. That approach was difficult in this domain because high-quality smell data has been scarce. Osmo has built its own dataset of more than a million human labels and millions of scent tokens that describe the exact molecules and concentrations behind specific odors, using spectrometers and chemical sensors that run continuously. Osmo’s technology has been applied in retail fragrances, a grant-supported project from the Gates Foundation that identified mosquito-repelling compounds, and work with StockX to detect counterfeit sneakers by chemical composition. Still, some experts remain skeptical. Smell scientist Avery Gilbert has criticized the company’s undocumented marketing claims and its continued reliance on human sensory panels, while former researcher Michael Schmuker calls the work an achievement but cautions that context strongly shapes how humans perceive smell, a factor that data alone may not fully capture.


Source: This smell startup says its AI nose is as good as a human’s via MIT Technology Review