Smar NetworkSmar.inSmargrid.comNewsInsuranceRoomsGamesMeasurementsCooking

SMAR Daily Industry News

Daily news across AI, technology, finance, insurance, healthcare, automotive, real estate, and government

Facebook Telegram
← Back to AI & Technology

Osmo AI Claims Human-Level Smell Detection

📅 2026-11-18 | 📰 MIT Technology Review
#ai smell #osmo #olfactory ai #fragrance design #scent tokens

In a Newark Airport-adjacent lab, AI startup Osmo is training a foundation model for smell using a dataset that blends over 10 million human scent descriptions with 800 million molecular “scent tokens.” Founder Alex Wiltschko, who spun the company out of Google Brain in 2023, says the model now performs at “human-level olfactory intelligence,” meaning it detects and discriminates odors as well as an average person and creates new scents at the skill level of a junior perfume designer.

Osmo’s approach relies on two parallel data streams. Spectrometers and other chemical sensors continuously analyze the molecular composition of items ranging from fruit and flowers to pantry staples and organs. Simultaneously, a team of about 70 trained labelers in the United States and overseas sniff the same samples and record standardized descriptors such as “nutty,” “peachy,” and “marine.” The combined data feed a large-scale AI model that can map molecular patterns to human-perceived aromas.

The company reports that its model has already outperformed in-house perfumers at one of the world’s largest consumer-goods firms and is being used by Procter & Gamble to design retail fragrances. Osmo also operates a smell-printing robot—roughly the size of a school bus—that mixes ingredients on demand. Users can submit ChatGPT-style prompts, and the system generates molecular formulas for custom scents, as demonstrated with a Portuguese custard-tart fragrance aimed at 18- to 23-year-old men.

Wiltschko envisions broader applications beyond fragrance creation. He cites potential deployment for threat detection at airports and disease diagnosis, leveraging the model’s ability to recognize complex odor signatures. The startup’s progress addresses a historic data gap: before Osmo, fewer than 200,000 human smell descriptions existed in public databases, and chemical sensor hardware was costly and the fragrance industry secretive.

Osmo’s dataset also reveals surprising overlaps in scent chemistry. For example, the molecular tokens defining a newborn’s smell share substantial similarity with those of a fresh plate of Barilla spaghetti, illustrating how rearranging molecule ratios can dramatically shift perception.

Why it matters

A reliable AI sense of smell could transform product design, safety screening, and health diagnostics for industries that rely on scent.

FAQ

What does Osmo define as human-level olfactory intelligence?
The model is about as good as the average person at detecting and discriminating smells and as good as a junior perfume designer at creating them.

How large is Osmo’s scent dataset?
More than 10 million human labels and 800 million scent tokens describing molecules and concentrations.

Which major companies have used Osmo’s technology?
Procter & Gamble and a major consumer-goods firm where Osmo’s model outperformed in-house perfumers.

What future uses does Wiltschko anticipate for the AI smell model?
Threat detection at airports and diagnosing diseases, among other applications.

Source: MIT Technology Review. See the original source for full details.