A WHIFF OF THE FUTURE

TEACHING AI TO SMELL FOR HEALTH, SAFETY— AND SHOPPING

- by Sarah Gottlieb

An Al-generated illustration of an AI model that can smell

Alex Wiltschko, in a long, white lab coat and protective glasses, picks up a Nike shoe box and slips a sensor into a tiny hole in the side. Inside is a pair of Air Jordans complete with Nike’s iconic swoosh logos. They look authentic, but do they smell authentic?

He watches the screen where the computer is plotting the scent. “Here comes the scent from the fake,” he says. “It’s much, much bigger and different than the signature for the real shoe.” He takes out one of the sneakers and sticks his nose deep into the high-top and confirms what the sensor had detected: These Jordans are fakes! They’re much more pungent and have a stronger chemical scent than real ones. Today, “what people [who are really into sneakers] are doing is they’re sniffing them to make sure that they’re real,” he says. “And it turns out that you can give that ability to computers.”

Not a Nike Air Jordan but a counterfeit sneaker!

The sensor Wiltschko was using was built by Osmo, an artificial intelligence (AI) company he founded in 2022. Spun out of Google DeepMind (Google’s AI research lab), Osmo has been on a mission to crack the code of digitizing smell.

What is a smell or scent or odor made up of? “It’s molecules that have come off of something, usually that’s alive,” Wiltschko explains. These particles “have traveled through the air and made it into our nose and touched a very sensitive, small piece of tissue in our nose that is sensitive to molecules. And our brain knows how to interpret that thing as an apple or milk, or a bad apple or spoiled milk.”

Scent Chemistry