Mark Megahan
A robot can easily tell the difference between whisky and whiskey. The one who hangs around with Andreas Grasskamp is also a whole lot better than distilling masters at detecting and identifying delicate flavor notes. The human experts are having some job security issues.
The robot nose knows
It’s not surprising that a robot can be more adept “at detecting delicate hints of apple or smoke in a glass of whiskey” than a Whiskey Master. Like the one at Jack Daniels.
The biggest difference between whiskey and whisky, other than the spelling, is the one which doesn’t have an “e” usually comes from Scotland. The other version usually comes from United States. There are exceptions to the rule, Japan and Canada both brew whisky. Ireland distills whiskey.
It’s almost too easy for the robot to tell which is which. There are a few chemical compounds that, when present, nail the distinction down solid.
Artificial Intelligence can do a whole lot better than that. It makes master distillers wonder how much longer they’ll be able to compete against a computer program.
The research team started by training the robot on “the molecular makeup of 16 different whiskies with their odor profiles.”
Their machine learning algorithm “was able to identify the top five flavor notes in each.” Almost every time, the machine named the same flavors as a panel of human experts.
A bouquet of gas molecules
The part we think of as the “bouquet” of a drink “is the product of dozens of gaseous molecules wafting up into the air.” The robot learned to identify them.
That glass of Jim Beam or Pappy Van Winkle in your hand has a unique collection chosen from “more than 40 compounds that create odors.” The chemicals produce a range of tastes and smells “from vanilla to caramel to smokiness.”
The robot takes a sip of the beverage to be tested with a mass spectrometer. That input gives the algorithm a taste of the sample’s molecular makeup. It’s learned to sort out the ones humans get excited over.
According to Andreas Grasskamp and his colleagues, at the Fraunhofer Institute for Process Engineering and Packaging IVV in Freising, Germany, “getting from that makeup to the subtler impression of an array of odors has proven difficult.” They did it. It’s never going to get burned in a unicorn re-bottling scheme.
It turns out that “individual molecules can have different odors depending on the medium they’re in — air, water, oil — and different odors stand up against one another in complex ways.” The robot can sort that all out with ease. The team “combined two algorithms: one a statistical computer model that distinguishes samples based on the detected molecules, and the other a neural network trained to predict identifiable scents based on the detected molecules.”
They used that to determine the top five flavor notes in each sample. The robot nailed exactly what the humans reported after tasting the same 16 products. Seven American and nine Scotch.
