This Hyper-Smelling AI Can Sniff Out Counterfeit Sneakers—and That’s Only the Beginning
source: fastcompany.com | image: pexels.com
Osmo, an AI startup focused on mapping scent, has an ambitious plan to use its sensor tech to find everything from fake shoes to tumors growing inside your body.
I want a tricorder,” Alex Wiltschko tells me on a Zoom call. Wiltschko, the founder of the AI company Osmo, is referring to the handheld device used by the Enterprise’s crew in its exploration across the universe. In Stark Trek, the tricorder can tell crew members everything they need to know about an object simply by holding it nearby.
One could say that Wiltschko and his team have created an alpha version of the fantasy device. His team has developed a backpack-sized machine equipped with a smelling sensor that uses artificial intelligence to identify counterfeit products by analyzing their chemical composition. Osmo has recently partnered with sneaker resale platforms to show that the high-tech sniff test is capable of identifying fakes with a high degree of accuracy.
IT’S ALL ABOUT THE MOLECULES
Everything in the world has a smell, from clothes to cars to your body. Those scents are volatile molecules, or chemistry that “flies” off those objects and reaches our nostrils to tell us things. You experience this consciously and clearly when something is new to your nose, like smelling a new car or a pair of sneakers.
But, even while you don’t notice smells, the molecules are always there. Dogs and other animals are much better at detecting scents because they have the organs developed to do so. And now we have sensors that can do the same.
Counterfeit shoes smell different from the real thing. Genuine sneakers and fakes differ not just in materials, but in chemical composition. Until now, companies like StockX have relied on human sniff tests and visual inspection to discern authenticity—a process that is labor intensive and expensive. Osmo’s tech aims to streamline the process.
According to Wiltschko, his team has “trained AI using highly sensitive sensors to distinguish those molecular differences.” This technology will transform how authenticity checks are conducted in industries that traditionally rely on manual inspection and intuition. Osmo aims to digitize that process, adding consistency, speed, and precision. Right now, he says that the Osmo machine takes about 20 seconds to tell fake from real. Soon, he says, it will be just five. And, eventually, it will be almost instantaneous.
Osmo’s technology’s foundation is built on years of laboratory work using highly sensitive sensors that are, as Wiltschko describes, “the size of a dishwasher.” These laboratory-grade sensors are designed to be as sensitive as a dog’s nose, capable of detecting the faintest chemical signatures. “We run these sensors 24/7, constantly collecting data about the chemical makeup of everything from plums and peaches to manufactured products,” Wiltschko explains.
The data collected forms the backbone of their AI training process, which is helping to create a high-resolution understanding of different scents and give them a location in a coordinate system called the Principal Odor Map. If you are familiar with how image colors are encoded in digital images, this works in a similar way. Roughly, the color of a pixel corresponds to a place on an RGB map, a point in a 3D space that has red, green, and blue coordinates. The Principal Odor Map works similarly, except the coordinates in that space predict how particular combinations of molecules will smell in the real world. Wiltschko says this map is Osmo’s secret sauce to make testing possible in portable units with lower resolution sensors that are roughly as sensitive as a human nose.
FROM THE LAB TO EVERYDAY GADGET
Wiltschko says that while the portable sensors are less sensitive than the lab units, the extensive data gathered with the high-resolution sensors makes it possible to perform effective scent detection. Like an image-scaling AI capable of inferring the contents of an image to create a higher resolution version based on billions of images from its trained model, this works the same with smell. This adaptability is crucial for real-world applications, where deploying a lab-sized machine isn’t feasible.
Rohinton Mehta, Osmo’s senior vice president of hardware and manufacturing, points out that the key to the identification process is not so much about the smells we can perceive but the chemical composition of the object, what lies beneath it. “A lot of the things that we want to look for and authenticate may not even have a perceptible odor,” Mehta says. “It’s more like we’re trying to analyze chemical composition.”
He describes the company’s recent pilot test with a major sneaker resale company that yielded a greater than 95% success rate in distinguishing fake shoes from real ones. Osmo’s sensors could even potentially be more accurate than human authenticators, with Mehta hinting that in some cases, the AI caught fakes that the sneaker company itself initially missed.
The method only works for high volume objects, for now. The tech can’t authenticate very rare things where only three were ever made, according to Mehta. This is because, as Wiltschko tells me, their AI learns using data. For it to learn the smell of a specific new model of shoe, you need to give it about 10 pairs of real sneakers. Sometimes, the scent signature is so faint that it will need 50 authentic sneakers for it to learn the new model.
Wiltschko envisions a future where the system doesn’t just authenticate products after they’ve entered the marketplace but actively prevents counterfeiting at the source.
Osmo’s lab doesn’t only smell things already made by others, but also creates new scents in-house using the same AI and robotic systems. The company’s scientists showed how this works in a practical way during an experiment they called the Scent Teleportation Project. They captured a scent using Gas Chromatography-Mass Spectrometry (GCMS), which breaks it down into its molecular components and uploads the data to the cloud. This captured data became a coordinate on the Principal Odor Map. Once mapped, a Formulation Robot in another location is then instructed to mix different elements according to the scent recipe, effectively re-creating the original smell.
Using this same scent manufacturing technology, Wiltschko imagines Osmo could embed odorless molecules directly into products as unique identifiers, creating an invisible signature that counterfeiters would have no way of detecting or replicating. Think about this as an invisible seal of authenticity. Osmo is developing these unique markers to be embedded in materials like glue or even in fabric itself, providing a covert yet unmistakable indicator of authenticity.
There is a huge opportunity here. As Wiltschko tells me, the sports industry is a multibillion-dollar market, with Nike alone reporting $60 billion in revenue last year. Yet, counterfeit versions of their products circulate widely, with a reported $20 billion in counterfeit goods cutting into those revenues. The U.S. Customs and Border Protection seized only $1 billion worth of counterfeit goods last year across all industry sectors, not just sports goods. Clearly, this scent technology could become a crucial weapon for battling fakes, particularly in the toughest cases where conventional methods, like examining visual markers, fall short.
SMELL IS A KEY TO THE FUTURE
Wiltschko sees the system as part of a broader strategy to digitize the sense of smell—a concept he began working on at Google Research. The foundation of the system lies in a concept called the Structure-Odor Relation (SOR). SOR is about predicting what a molecule will smell like based on its chemical structure, and the key to solving this was the use of graph neural networks.
These advanced machine learning algorithms allowed Wiltschko’s team to build a complex odor map capable of interpreting and predicting scents in a way that computers could understand. “We built new molecules no one had ever smelled before and predicted them with superhuman accuracy,” Wiltschko says.
Today, Osmo’s gadget is larger and not quite as magical or omniscient as Star Trek‘s tricorder. But Wiltschko imagines that someday it could be the size of a matchbox and detect everything from a toxic substance in your home to the freshness of food in your fridge to a cancerous tumor growing inside you.
On a practical level, Wiltschko believes there’s opportunity to use portable sensors for household safety, like detecting toxic chemicals in the air or identifying harmful gasses that could be present in a kitchen or garage. Right now we have rudimentary carbon monoxide detectors, but there are many more molecules that modern sensors can register. Eventually, Wiltschko predicts, professional-grade environmental sensing will find their way into everyday homes, giving families tools to monitor and manage the safety of their environments.
The medical potential of this technology is equally transformative. Wiltschko envisions the system being used to detect diseases early—such as cancer, diabetes, or even neurological conditions like Parkinson’s—by analyzing the subtle changes in body odor that often precede symptoms. “The human nose can detect some of these changes, like when a loved one is sick, but we’re working on making a system that can quantify and analyze these changes far earlier and with much more accuracy,” Wiltschko explains.
He’s cautious about when this advancement might happen, saying that scientists must first identify the molecular markers of these scents before the machine can detect different illnesses. The company is already working with researchers, but there’s a long path ahead. That said, the implications for preventive healthcare are vast, potentially enabling earlier diagnosis and better outcomes for patients.
Despite starting with something as mundane and small as sneakers, the goal is anything but. His dream of a compact tricorder-like device is still in the future, yes, but he’s convinced we will get there sooner than later.