Osmo, an artificial intelligence startup founded by former Google Brain researcher Alex Wiltschko, says its models have reached “human-level olfactory intelligence”—matching the average person at detecting and distinguishing smells and performing like a junior perfume designer when creating them. The company’s technology is already being used by fragrance brands, and Osmo says its model recently outperformed an in-house perfumer at one of the world’s largest consumer goods companies. Wiltschko believes the same technology could eventually support applications such as threat detection and disease diagnosis at airports.
Building an AI dataset for smell
Wiltschko spun Osmo out of Google Brain, Google’s former artificial intelligence research lab, in 2023. Like many AI startups working in other fields, the company is applying large datasets to a difficult problem. In olfaction, however, the main obstacle is the limited amount of useful data.
Before Osmo began its work, Wiltschko says existing databases contained fewer than 200,000 human descriptions of smells. Chemical sensor equipment capable of collecting additional data is expensive, while the fragrance industry has traditionally been secretive about its methods for studying scent.
Osmo has created a dataset containing more than 10 million human labels and 800 million “odor tokens.” These tokens describe the specific molecules and concentrations that produce particular odors. The company operates spectrometers and other chemical sensors around the clock, analyzing the molecular structures of materials ranging from fruits and flowers to food products and organs.
About 70 trained data labelers in the United States and abroad spend their days smelling samples and describing them using a standardized scent vocabulary.
How molecules shape perception
Mapping scents in this way has revealed unexpected similarities. As changing the order of words can alter the meaning of a sentence, changing the proportions of scent molecules can significantly alter the smell perceived by humans.
“The odor tokens that define the smell of a newborn baby overlap significantly with the odor tokens found on a plate of freshly cooked spaghetti,” Wiltschko says. He specifically cited Barilla spaghetti.
Printing custom fragrances
Osmo also creates scents using a scent-printing robot roughly the size of a school bus. The machine combines ingredients according to precise formulas to produce samples for the company’s research and for paying customers.
Customers can design and print fragrances to order using prompts similar to those entered into ChatGPT. During a demonstration in Lisbon at the Web Summit, Wiltschko opened the company’s AI model and was asked to generate a scent resembling a Portuguese custard tart for men aged 18 to 23.
After several minutes, the system produced three potential scent profiles based on a brief describing the target aroma as creamy, coconut-like and fruity. It also identified the molecules needed to create each option. Had the order been completed, the resulting custom fragrance would have been ready in about a week.
Source: www.technologyreview.com


