A start-up claims its systems can distinguish counterfeit sneakers from the genuine articles – by their smell. In short, the two things smell different because they are different, at the molecular level. The trick is to map the molecules and compare the maps.

Both the founder, Alex Wiltschko, and the team at Osmo AI hail from Google Research, where they worked in machine learning. Since then they have used graph neural networks to predict a substance’s smell by its molecular structure, created substances with novel molecular structures and predicted their smell, and moved to a practical application: the molecular design of insect repellent, one that smells bad to mosquitoes.

Osmo AI

The company is backed by Lux and Google Ventures (GV), along with the Bill & Melinda Gates Foundation and several others. Its payroll includes a master perfumer, a neuroscientist, specialists in analytical and synthetic chemistry, and specialists in machine learning.

According to Fast Company (FC), years of work in the lab have produced sensors with the size of a dishwasher and the olfactory acuity of a dog. The machines run round the clock, “constantly collecting data about the chemical makeup of everything from plums and peaches to manufactured products,” as Wiltschko tells FC, and laying out the data in what Osmo calls a Principle Odor Map. FC likens the map to the three-dimensional RGB (red, green, blue) map of digital colors, where every color has its own set of three coordinates. The number of dimensions in Osmo’s map is no doubt higher but unclear.

In any case, the map promises to make smell tests conclusive even with smaller, less sensitive sensors. The collected data’s strength can, with some AI-powered inferences, overcome the hardware’s shortcomings. “A lot of the things that we want to look for and authenticate may not even have a perceptible odor,” Rohinton Mehta, Osmo’s Senior Vice President of Hardware and Manufacturing, tells FC. “It’s more like we’re trying to analyze chemical composition.” Apparently, a recent test with a sneaker reseller yielded an accuracy of 95 percent.

Earlier this year Osmo succeeded with a related experiment, which it calls scent teleportation. If you can map the formula for a scent in one place, then in principle you can send the map somewhere else and, with the right machines, formulate the same scent there. This is what Osmo has done, with gas chromatographer / mass spectrometer, a digital cloud and formulation robots. “The process,” according to the company, “requires no human intervention at this point, except at the input and output level.”

Nothing in the process is novel in itself. Any olfactory expert, as the company say, “will likely be familiar with most of the technologies.” The novelty, as with the atoms of a scent, lies in their combination.

Back to the real world

We’re speculating here, but if it turns out to be fast, simple, cheap and accurate, Osmo’s system or systems like it could perhaps shift the authentication business back to physical articles.

This year the EU will begin to require “nearly all products” sold on its territory to carry a digital product passport (DPP), with “comprehensive information” on “origin, materials, environmental impact, and disposal recommendations.” It will have a “unique product identifier, compliance documentation, and information on substances of concern.” It will have manuals and safety instructions. Companies like Certilogo already specialize in this.

And a DPP will no doubt serve as an indirect guarantee of authenticity. There’ll be a QR code right on the product’s label, ready for scanning and document download.

A similar system exists at companies like StockX, the marketplace for collectibles, notably sneakers. As we’ve reported, StockX maintains a physical vault where investors can keep their assets, just as a middleman will keep physical metals for the shareholders of certain exchange-traded funds (ETFs). What these StockX traders trade is NFTs, which flit around on a blockchain ledger. Like labels, these NFTs are assigned to the physical assets. In other words, StockX has set up a secondary market.

In all such cases it is the “label,” the asset’s representative or symbol, that undergoes scrutiny, and so the owner could always fall victim to the old switcheroo – unless, of course, we return to the examination of assets in themselves.