Whiskey AI:01 Intelligens
Title | Whiskey AI:01 Intelligens |
Brand | MACKMYRA |
Product/Service | AI01: INTELLIGENS |
Category |
A03. Data-driven Consumer Product |
Entrant
|
FOURKIND Helsinki, FINLAND
|
Idea Creation
|
FOURKIND Helsinki, FINLAND
|
Idea Creation 2
|
MACKMYRA Stockholm, SWEDEN
|
Media Placement
|
FOURKIND Helsinki, FINLAND
|
Media Placement 2
|
MACKMYRA Stockholm, SWEDEN
|
PR
|
FOURKIND Helsinki, FINLAND
|
PR 2
|
MACKMYRA Stockholm, SWEDEN
|
Production
|
FOURKIND Helsinki, FINLAND
|
Production 2
|
MACKMYRA Stockholm, SWEDEN
|
Credits
Jarno Kartela |
Fourkind |
Principal Machine Learning Partner |
Lasse Tammilehto |
Fourkind |
Design Partner |
Niklas Collin |
Fourkind |
Engineering Partner |
Max Pagels |
Fourkind |
Machine Learning Partner |
Mikko Pajulahti |
Fourkind |
Business Design Partner |
Angela D'Orazio |
Mackmyra |
Master Blender & Chief Nose Officer |
Susanne Tedsjö |
Mackmyra |
Sales & Marketing Director |
Why is this work relevant for Creative Data?
It's the world's first complex consumer product that's entirely generated by machine learning, in a field which is deeply traditional and human craftsmanship driven.
Background
Mackmyra has distilled whisky for 20 years in Sweden. They have challenged the conservative trade since its beginning, pushing boundaries on what can be done with whisky. To celebrate their 20-year-old journey, they wanted to deliver something extra special.
They wanted to leave the whisky recipe generation completely to an algorithm to see if it is possible for a machine to do the most complex and and labour-intensive task related to making whisky.
Describe the Creative idea / data solution (20% of vote)
Making the world's first whisky recipe with machine learning and showing the way forward on how machines can be used to make complex consumer products. To do so, we first explored all current generative models but due to poor performance, ended up creating a proprietary generator-discriminator model that was designed to explore new spaces and generate unique recipes, but we also wanted to make the best possible whisky.
We used previous recipe data, tasting notes, ratings of previous recipes, expert reviews, customer reviews and cask information - internal ratings, cask types, filling stages, volumes and alcohol levels to make our model understand what's Mackmyra whisky.
And then we created a framework which can innovate in this space, creating new whiskies that are unique but ultimately taste excellent.
Describe the data driven strategy (30% of vote)
We worked as a team with Mackmyra to gather all previous recipe data in detail to an originating recipe matrix. We scraped online whisky review sites for reviews and ratings for all Mackmyra products. Expert reviews, tasting notes, master blender notes as well as awards in whisky competitions were also added as rating data for the previous recipes.
Then we embarked to the whisky blending process and visited the underground warehouse where whisky is stored in casks, and explored the way how these casks are blended to make new recipes. We then collected all cask data to understand what are the possible ingredient types we are going to use in the generative part. All of this was split into the generative part's data model (previous recipes and their respective metadata) and the discriminate part's data model (previous ratings, awards, tasting notes, cask ratings etc) and then used to make whisky.
Describe the creative use of data, or how the data enhanced the creative output (30% of vote)
Our model's creativity stems from splitting the data models between our generative part and discriminative part. Since the generative part does not receive the scoring data, it is allowed to explore new spaces from the infinite amount of possibilities one could create from Mackmyra's ~100 cask types.
To make it perform well, our generative part takes turns in making selections of cask type & amount and pushes it to our discriminative model which responds if it was better or worse than the previous version. If better, we continue from that recipe and try to further improve it, if worse, we fall back to our previous recipe and continue from that. This is crucial to make the model creative.
Without being able to taste anything, the model can produce new recipes and mimic what master blenders have been doing for centuries - something that was seen impossible by most industry experts.
List the data driven results (20% of vote)
Mackmyra is now in possession of a framework that allows them to create new recipes at a touch of a button. They can use this to make new products, to attract new segments, even to serve new B2B customers by allowing them to make their own bottlings - something which was impossible previously due to the amount of expert work required.
The first batch of our machine learning whisky was sold out in day and we gained media awareness from The Times to Food & Wine, Forbes, Popular Mechanics and The Register to name a few.
Mackmyra's overall sales have improved due to the raised awareness and new batches of the machine learning whisky are being produced. Overall process has gained interest in creative technology throughout the world and the case itself is a sought-after keynote concept.