The Center for Data Innovation spoke with Frederik Duerinck, founder of Algorithmic Perfumery, a startup based in the Netherlands that uses AI to create bespoke perfumes. Duerinck discussed how psychological and sociological data are used to create one-of-a-kind scents and how technology is helping to make the time-honored craft of perfumery more accessible.
Hodan Omaar: What inspired you to start Algorithmic Perfumery?
Frederik Duerinck: I had the idea in 2017 when I was working on an art exhibition in New York on the future of multisensory design with International Flavor and Frangrances (IFF), one of the world’s largest fragrance and flavor companies. The central question of the exhibition was: How can we challenge the status quo in perfumery?
One of the pieces I worked on for the exhibition was a collaboration between an electronic musician who uses computer-generated sounds and a perfumer who controls every decision in the creative process. During the exhibition I had the idea—what if there was a system that uses technology to create bespoke perfume and people could watch that creation happening in front of their eyes? I started working out how much funding I would need right then and there, quit my job, and started pitching Algorithmic Perfumery in the United States a year later in 2018.
Omaar: How does it work?
Duerinck: Currently, visitors can go to one of our locations in London, Dubai, Vienna, Breda in the Netherlands, or they can go to our website. They are invited to fill in a short questionnaire on their phones and our AI-enabled machine will create three one-of-a-kind scents based on their answers.
We have three systems at work. One system creates the perfume. It’s made up of 46 scent building blocks, or “chords” as they are called, and each chord is made up of five to eight ingredients. It uses machine learning to understand how to combine these chords into a new vial of perfume and to understand what combinations will and won’t work. The second system interprets the data the customer has supplied and creates a unique profile for them. The third system matches their unique profile to the machine to create a one-of-a-kind scent for them.
Omaar: What kinds of data do you need from customers to create their profiles?
Duerinck: Scent is profound; fragrance evokes an immediate physiological response, often associated with deep emotions, memories, and sensory experiences so we collect physiological data and sociological data, as well data on stylistic preferences. Your preferences are really determined by where you come from, so we factor that in as well as implicit data like the amount of time it takes to answer a question on the questionnaire. We are continuously iterating on the questions we ask. As a society, we know very little about how smell works compared to say, how sight works, so there is a lot of room for improvement and it is a fascinating space!
The other important data we collect is feedback from customers. Users can give feedback on their perfume formulas and we feed that information back into our algorithm so that their particular fragrances can be tweaked over time to better suit their preferences.
Omaar: The famed French perfumer Jean Carles once said perfumery is an art, not a science. What do you make of critiques that say AI dehumanizes perfumery?
Duerinck: Algorithmic Perfumery is about empowering individuals to craft distinct scents that deeply resonate. I think it is fascinating that AI can help us create things that evoke genuine emotional reactions and also challenges us as a species to question what it means to be emotionally touched.
Secondly, we believe every person knows what is the best scent for them. Part of what we are doing is taking the art of perfumery out of the ivory tower and using AI to help people safely experiment for themselves, so rather than dehumanizing perfumery I think what we are doing is empowering human agency.
Omaar: What are the biggest challenges you are working on to improve Algorithmic Perfumery?
Duerinck: We are working on how we can improve how our customers interact with our systems. Right now it’s a questionnaire, but we are exploring new technologies that we can incorporate, new data types we could collect, and building up our infrastructure to support the collection of new insights.