He walks down the street and passes an ad for a perfume. For this poster, an entire visual universe has been imagined to match the scent of the perfume with the colors and thus arouse the customer’s curiosity. Intrigued, you go to a perfumery to smell it. Before that, you discover the packaging of the perfume, its packaging, its bottle… Last step, you smell it. And you are not so surprised by the smell because it matches the visual universe that was presented to you earlier.
All this is not anecdotal since upstream a work of coherence has been established. But the most amazing thing is that this correspondence can be done by artificial intelligence. This is all the work done by Muriel Jacquot, head of an artificial intelligence tool within Givaudan, one of the world’s largest perfume manufacturers. The idea? An artificial intelligence (AI) capable of translating all kinds of smells into colors and emotions.
Two years ago, Givaudan approached Muriel Jacquot about her AI tool called Myrissi. “I don’t come from that world at all,” he likes to clarify with Tech & Co.
In 2012, its technology developed at the University of Lorraine was patented. First a start-up, Myrissi is now a fully integrated system within Givaudan. “Now we can explore the entire Givaudan fragrance catalog in terms of colors and emotions,” explains Muriel Jacquot.
The objective is to predict and model what a consumer will feel when smelling a perfume, translate it into colors, and thus guide brands in the construction of the complete visual universe around the new perfume” and investigate in terms of emotions, what emotional territory we can explore. “adds Muriel Jacquot.
“An emotional AI”
Technically, the basis of the Myrissi system is based on more than 25,000 consumer tests. “We asked a panel of consumers to make them smell odors. Instead of asking them if they liked that smell, they were told to translate their perceptions, what they feel, into a color. Color and smell interact with a large potential, here we have a kind of emotional AI ”, he specifies.
Thus, in the base, a perfume component corresponds to 5 or 6 colors. “We need to have a proposal that encompasses a maximum of responses from consumers.” Specifically, the brands with which Givaudan works present the different components of its new perfume. “With the help of Myrissi, we present a color palette with the most associated shades.”
Today in charge of everything related to neuroscience at Givaudan, Muriel Jacquot works with the research team, develops synesthesia activities, that is, the interaction between the senses, and ensures the correct integration of her tool in the company. “We had to adapt our AI to be compatible with Givaudan’s systems. The company came looking for innovation because we were the only ones with this approach and had collected enough data.”
“Maximizes olfactory performance”
From now on, any new ingredients coming to Givaudan will be integrated into the tool. Therefore, AI seeks to satisfy the largest number of consumers. To this, Muriel Jacquot rétorque that the but is ahead of “repondre de façon plus pertinente à la demande du client, d’augmenter la satisfaction du consommateur et d’anticiper de manière fine sa réaction afin que tous les signaux sensoriels aillent dans le same sense”. Givaudan’s neuroscience team is currently working on a cultural approach because consumer responses will be different in different regions of the world.
Givaudan integrated AI into its creation process in 2017 with a dedicated team of 25 people, including 10 data scientists. The multinational is also betting on AI to test new combinations of scents. Carto is a tool developed by the company in 2019, around the Odor value map to “maximize the olfactory performance of the final formula,” according to Givaudan. Therefore, a touch screen, which is based on data visualization and cognitive insight, makes it possible to explore all of Givaudan’s palettes.
A robot extracts from a database the 288 most used ingredients in perfumery and gives the possibility to try new combinations, all in a few seconds. Carto is thus able to suggest combinations and estimate their potential success. After this magic trick, a sample comes out with the new combination created with this machine.
invent new smells
In Cambridge, Massachusetts, Osmo, a Google start-up, is also using artificial intelligence, but this time to intervene directly on odors. Its founder, Alex Witschko, is a doctor in neuroscience and joined Google Research six years ago. “I’ve always been fascinated by neuroscience and deep learning,” he told Tech & Co.
Alex Witschko and his team have developed a system to predict the smell of molecules based on their structure and thus invent new ones. Twelve people are working on the project, and “this figure should double this year,” the founder rejoices. And the company does not hide its intentions: to create the next generation of molecules for perfumes, but also shampoos or candles.
To achieve this objective, Osmo has designed an odor map whose database is based on 5,000 molecules from perfume catalogues. And it is thanks to the AI that this map was built. The idea is to categorize odors so that molecules that smell the same are grouped together.
From this set, the software learned to recognize associations between the chemical structure of each molecule and the way a human would describe it, that is, the perception of smell. For example, is the smell fruity, buttery, woody? In this way, the software has already been able to predict the smell of 400 molecules based solely on their structure, including one that mixes the smell of watermelon with that of the ocean. At the moment, the idea is to recreate perfumes that exist in nature, but are still scarce in the form of molecules.
For example, to recreate the scent of lily of the valley, Osmo predicts its scent using AI before analyzing it. Osmo asks the AI to find the molecules that smell like lily of the valley. The system will propose molecules that are close to this smell to recreate it.
End of human creation?
But mixing two existing molecules to recreate or create a new scent can be risky because chemical bonding or the number of carbon atoms in a molecule can affect its smell and quickly sour it. And there are many failures.
Although the global perfumery industry weighs in at 30 billion dollars, the other issue that Osmo wants to respond to is environmental. His software would thus avoid picking up the source of the scent sought at the other end of the world. They also want their new molecules to be biodegradable. In the longer term, Alex Wiltschko would like to digitize the scents and be able to send them from his mobile phone.
Asked if these technologies will replace the professions of perfumery creators, Muriel Jacquot and Alex Witschko reply that it is above all a tool, but that it will develop more and more in the field of perfumery “at the service of humans”. ”.
Source: BFM TV
