Use #braubeviale #myBeviale

28 - 30 November 2023 / Nuremberg, Germany

BrauBeviale Newsroom

Artisans versus artificial intelligence

For the world's first AI gin, an algorithm selected the botanicals to use
For the world's first AI gin, an algorithm selected the botanicals to use // © Circumstance Distillery

With the world’s first AI gin and the first whisky blended by a machine, we asked what artificial intelligence can bring to beverage production – and what it can’t.

AI in beverage production

Full speed into the future. Whereas last year there was talk of digitalisation in nearly every conceivable area of our lives, the buzzword now is artificial intelligence. Artificial intelligence (AI) describes the process by which a computer makes independent decisions – based on the knowledge used in training the computer. This, in turn, is known as machine learning. And that’s another buzzword.

In fact, humanity is still in its infancy when it comes to the development of artificial intelligence. All you have to do is listen to IT visionaries to find out what – theoretically – will be possible in the future. Currently, we all use a form of artificial intelligence when we use Google Search. Search engines are forms of AI. Other functions that rely on AI include Google Translate, and the face recognition feature on Facebook and Siri. AI is also used in medicine, computer games and autonomous weapons – AI is terribly versatile.

But AI has nothing to do with soft drinks, beer, schnapps or wine. Not so far, at least.

AI enters the spirits industry

In 2019, artificial intelligence experienced two firsts in the spirits industry. First, Swedish whisky maker Mackmyra introduced ‘Intelligens’, an elegant, golden Swedish single malt whisky. For this whisky, Mackmyra's master blender Angela D'Orazio worked closely with tech giant Microsoft and also Fourkind, a Finnish IT company focusing on machine learning.

A short time after that, Monker’s Garkel, the first AI-created gin, came onto the market. The Circumstance distillery in Bristol joined forces with creative agency Tiny Giant and digital consultancy firm Rewrite Digital to produce a gin. The fruit of this collaboration combines coriander, raspberry leaves, angelica, gooseberry, prune, mandarin and orange peel, marigold and, of course, juniper. These are the botanicals that the AI believes make the best gin.

We spoke to the people behind these two projects. What drives a master blender and a distiller to leave their own work to a computer programme? How does that even work? And above all, what does this mean for the future of the spirits industry, and possibly the entire beverages industry? Will smart machines in the future brew and distil whatever we like to drink?

Spoiler: no, they won't. At least not in the foreseeable future. Artificial intelligence will not be taking over large parts of beverage production (at least not for the time being), but there are some approaches hidden here that could be useful. Anyway, it’s bad to close your eyes to the future for too long; it’s worthwhile to at least not categorically rule out the use of artificial intelligence in distilleries or breweries. Plus, it’s good for a successful PR stunt.

Monker's Garkel: the AI gin

Liam Hirt, founder of Circumstance Distillery in Bristol, UK, launched Monker’s Garkel in autumn 2019 – the world’s first gin made with the help of artificial intelligence.

Mr Hirt, how did you come to bring gin and AI together?

Liam Hirt: As far as gin is concerned, I'm self-taught. Before founding Circumstance distillery, I was distilling in my cellar as a hobby. That was in my previous life, when I was a cardiologist. One of my last research projects was related to machine learning in medical research. Somehow the crossover came from that. As we shifted our focus as a distillery increasingly towards gin, conducting a lot of recipe development on behalf of customers, the idea came to me to use an AI to speed up the creative process.

You’ll have to explain a bit more about how exactly that works.

Hirt: The AI helps with quickly and accurately selecting various botanicals. We work with a so-called recurrent neural network. This basically works like the neurons in the human brain.

Except the machine can’t smell or taste, so how can it tell which botanicals are good?

Hirt: We have taught the machine many things. There is a large database in a recurrent neural network. We have stored thousands of possible ingredients and described the individual botanicals. When we ask the AI to write a recipe, it selects botanicals from this database based on the information stored within it and then it makes a recipe suggestion. We, as human experts, read this suggestion and respond with a ‘yes’ or ‘no’. If we say ‘no’, the AI tries again, but it remembers what we said – what didn’t work – when making the next suggestion. Then we make a decision on its next suggestion. The machine learns with every recipe rejection.

That sounds like a lot of effort and not really an acceleration of the creative process.

Hirt: It all happened within two months. The most laborious part was the creation of the database. Now that the machine is trained, it only suggests useful recipes. We had it write five recipes, and we produced samples of the best three of these – the resulting gins were all good.

Does that mean that, from now on, the AI will always write your new gin recipes?

Hirt: No! We only used the AI for producing Monker’s Garkel, and haven’t used it since. We don't have any concrete plans for using it. Nevertheless, the potential isn’t to be scoffed at: the way we used artificial intelligence in our distillery, for product development, as a creative member of our team, was very useful. In brainstorming sessions, the AI can provide valuable input with interesting ideas. But of course, in the end, the actual production of a successful new product only works with real people.

The master blender and the AI whisky

Angela D'Orazio, master blender at Swedish whisky-maker Mackmyra, teamed up with AI experts from Microsoft and the Finnish machine learning company Fourkind in the spring of 2019 to create a whisky using a computer programme – the first ever AI whisky.

Ms Orazio, please tell us a little about yourself.

Orazio: I’ve been working in the whisky industry for 27 years, and specifically as a whisky blender since 2005. It is a marvellous and very creative job. You can use all your senses. I also have to have an overview of the whole distillery, knowing every single cask at all times. There’s a lot of planning work involved in keeping everything organised, knowing exactly when each whisky will be however old, and so on.

So, your job requires experience, creativity and all your senses – skills that I would assume a machine would never have.

Orazio: That's right. We had to feed the machine with data first. For months, we fed the computer programme with a number of parameters: barrel types, age of the whiskies inside, finishes, different types of distillation. As a Swedish company, we are not as bound by tradition as our Scottish colleagues, so we can be very creative in making whisky. We sometimes work with Swedish oak, sometimes with American oak, sometimes with our own yeasts, etc. We also entered reviews and prices into the programme, as well as all the opinions people had about our whiskies.

So, the machine has developed taste, so to speak?

Orazio: Exactly.

But isn't a whisky going to be automatically more mainstream if the AI's taste is based on what has already proven to be particularly popular?

Orazio: Well, the AI only makes suggestions. There has to be a person there, as mentor and supervisor. The person makes the decisions, not the AI. The machine can process questions, but at the end of the day it is a tool used by the person who controls it. So, my experience and my senses were important throughout the process. The AI was my tool, my assistant.

So, honestly, does it bring a great benefit?

Orazio: I can only speculate here, but I suppose if I were to use this programme regularly – which I don't – and someone could predefine the parameters that are important to me, then the AI would probably be able to provide reliable help to me and speed up the process of creating a new product.