KI lernt, zwischen Aromen von US-amerikanischen und schottischen Whiskys zu unterscheiden | Ein Algorithmus identifizierte die fünf stärksten Noten in jedem Getränk genauer als jeder andere einer Expertengruppe
https://www.theguardian.com/technology/2024/dec/19/ai-learns-to-distinguish-between-aromas-of-us-and-scottish-whiskies
7 Comments
From the article: Notch up another win for artificial intelligence. Researchers have used the technology to predict the notes that waft off whisky and determine whether a dram was made in the US or Scotland.
The work is a step towards automated systems that can predict the complex aroma of whisky from its molecular makeup. Expert panels usually assess woody, smoky, buttery or caramel aromas, which can help to ensure they don’t vary substantially between batches of the same product.
“The beautiful thing about the AI is that it is very consistent,” said Dr Andreas Grasskamp, who led the research at the Fraunhofer Institute for Process Engineering and Packaging in Freising, Germany.
“You have this subjectivity still in trained experts. We are not replacing the human nose with this, but we are really supporting it through efficiency and consistency.”
Nailing down a whisky’s aroma is no simple business. Most of the strongest notes in the spirit are a complex mixture of chemicals that interact in the nose and even mask one another to create a particular aromatic impression. The interactions make it extremely difficult to predict how the whisky will smell from its chemical signature.
For the latest work, the researchers obtained the chemical makeup of 16 US whiskeys and Scottish whiskies, including Jack Daniel’s, Maker’s Mark, Laphroaig and Talisker, and details of their aromas from an 11-strong expert panel. The information was used to train AI algorithms to predict the five major aromas and origin of the drinks from their molecular constituents.
One algorithm was more than 90% accurate at distinguishing the US from Scottish spirits, though the performance would be likely to drop against tipples it had not been trained on. On average, it identified the five strongest notes in each whisky more accurately and consistently than any individual on the expert panel. The details have been [published](https://www.nature.com/articles/s42004-024-01373-2) in Communications Chemistry.
Wow a molecular detector and machine learning algorithm could find patterns?
In any other era this wouldn’t be called AI.
If I had a quantitative chemical detector instead of a nose I think I’d be a lot more accurate as well.
The big takeaway here is that human noses/taste buds are not as good as mass spectroscopy at determining chemical makeup (obviously).
The actual AI used here is ancient tech. We’ve been using Convolutional Neural Networks since 1969. They’ve been good enough to do this type of classification task for at least a decade.
Underwhelming. My nose is also less accurate then a detector. They can “sniff” things I don’t even have receptors for.
Hence why I own CO detector. It’s more accurate then me 100% of the time.
That being said I can see the equipment and software being useful for mass producers who aim for consistency in product smell/taste.
I’d love to see reverse-engineering the ingredients/recipe for Coca Cola to be a “Turing test” for this sort of thing going forward.
Remove the fact it’s tested Whisky and the charm is lost to pure aromatic detection. Impressive technically but I’ll stick with a human nose to describe how whisky REALLY smells!
Smellavision is just around the corner guys. Get ready for the end of sports TV. And porn