Take the Whisky Test: How IBM Tech Can Help Root Out Fake Scotch
With the festive season just days away, it’s not just turkey and cranberry sauce flying off the supermarket shelves. It’s also champagne, wine, gin… and, of course, whisky.
And it better be genuine.
IBM researchers are working on technologies to help you make sure the whisky you buy is indeed what it says on the label. Whisky is a counterfeiting target all over the world, be it in the developed or developing markets. Recently, for instance, Scottich scientists found that more than a third of 55 bottles of vintage Scotch tested in a lab were fake. And in general, about €2.8 billion gets lost annually in Europe because of counterfeiting in the wine and spirits industry, according to the European Union Intellectual Property Office.
And while many counterfeit products can be quickly identified using simple analytical processes, spotting “more sophisticated examples of the counterfeiter’s craft is way trickier,” explains Ian Goodall, senior scientist at Scotch Whisky Research Institute (SWRI) in Edinburgh.
Three teams — in Daresbury lab near Manchester in the UK, in Zurich, Switzerland and in Yorktown Heights half an hour drive from New York — have developed AI-powered technologies to rapidly and accurately differentiate between whisky samples.
And it’s not just about spotting fake whisky like, say, vodka mixed with some coloring, says Clyde Fare, a computer scientist at IBM Research based in Daresbury, UK. It’s also about making sure that it’s Scotch whisky and not Irish or American whiskey (note the spelling — only Scotch whisky is spelled with a ‘y’, everything else is spelled with ‘ey’ at the end) or anything else that is simply not, ahem, Scotch. Scotch whisky is mostly made from wholly-malted, peat-smoked barley and aged in oak barrels for three years or more. American whiskey or bourbon, on the other hand, is distilled from corn, while Irish whiskey is made from kiln-dried, raw and malted barley.
For manufacturers, making sure the drink is genuine comes down to brand protection — they have to make sure that a specific whisky brand remains the same from batch to batch and from year to year, despite any seasonal variations in the starting materials — ensuring brand loyalty.
Scotch whisky specifically is one of the largest contributors to the UK economy. “Identification, eradication and legal prosecution of fake and adulterated whisky is key,” says Geraint Morgan, a chemist at Open University that works with IBM on testing the technology. Besides brand protection, “fake distilled spirits can contain harmful compounds and, in some cases, may even result in death of customers,” Morgan adds.
That’s why SWRI asked IBM (which is working together with the UK’s Science and Technology Facilities Council) and Open University to develop a method to accurately distinguish Scotch whisky from anything else. As counterfeit products get ever more sophisticated, it’s crucial “to identify unknown markers for counterfeit detection as well as inconsistent levels of known whisky compounds,” says Goodall. That’s a challenge to analytical methodology, and an even greater challenge is being able to automate the processing of the large data files created.
To develop the technology, first SWRI delivered about a hundred whisky samples to Morgan’s team, half from various Scotch whisky brands and the rest — either non-Scotch whisky samples or some totally fake stuff. Every sample has different molecular composition, and to analyze it, Morgan and colleagues have been using chromatography and mass spectrometry. First, a tube-like tool separates molecules in a sample by shape. Then the mass spectrometer breaks them into fragments — charged particles (ions) and measures the mass and electric charge ratio using electric and magnetic fields. The output is the liquid’s chemical fingerprint — the exact molecular composition.
This data, in digital form as a graph with several peaks, gets sent to IBM Research. Each peak corresponds to a fragment of a molecule. Typically, it’s human experts — mass spectrometrists — who interpret the graph. Computer programs do it better, though — and IBM’s software compares the output to the graphs of known Scotch whisky chemical fingerprints to get the best match. “Human experts understand that this kind of molecular feature will appear as that particular spectral pattern, but they might miss small though ultimately significant variations in the data. Machine learning methods are better able to flag up these kinds of small differences,” says Fare.
Two real whiskies will be different, and they will also be different from fake whiskey. “Our job is to write algorithms that will identify the kinds of patterns present in the real ones versus those seen in the fake ones,” says Fare. The software has about 90 percent accuracy, he adds, and is getting better with more data. The technology can also be used to identify the authenticity of other spirits — as well as for other food product testing and even anti-doping screening in sports.
But that’s not the only tool at IBM that can reliably spot counterfeit whisky. A team in Zurich led by scientist Patrick Ruch has developed an “electronic tongue” — a device called Hypertaste that complements the human sense of taste. An array of minitaturized electrochemical sensors on a device that looks like a lemon slice detects chemical information in a liquid by measuring changes in voltage signals.
Those signals are then evaluated by machine learning models in the cloud, which determine the similarity of the tested substance to reference substances, and sends the result to asmartphone app. This way, the technology provides chemical fingerprints within just a minute — for complex water-based liquids, such as beer, wine — or whisky (and whiskEy, too).
And across the Atlantic, in Yorktown Heights, researchers have built what they call the IBM Crypto Anchor Verifier — a tiny tool that works with smartphones to detect whether a product is genuine or not.
It consists of a small optical device that gets attached to an smartphone. Point the device at something, and the Verifier scans it, capturing its unique wavelength signature and microscopic features. Every material has a different molecular structure, so as light gets reflected, it carries with it a unique wavelength signature. As it’s captured by the instrument, the AI in the Verifier app installed on the phone analyzes the signature and immediately gives an output — a graph that tells the user whether the material is authentic. “We’ve used it on clothing and drugs, as well as wine, whiskey and vodka, cooking oils, diesel fuels, metal parts, paper products, and so on,” says Donna Dillenberger, an IBM Fellow and the lead researcher on the Verifier project.
The optical device can capture features with a one-micron resolution — and there’s 25,400 microns in one inch. “We use microscopic features and wavelength data to authenticate clothes, and to tell the difference between luggage. For paper, you could use this to tell whether a government certificate is counterfeit or not. We’ve used to tell whether signed documents are signed with ink or they’ve just been Xeroxed,” says Dillenberger.
Several companies are already working with the Crypto Anchor Verifier, and it can also be used together with blockchain, to help make sure that products stay authentic as they move through the entire supply chain — for instance, to assess the origin of diamonds.