Over the last few years, one of the biggest talking points in a wide range of industries was the potential impact artificial intelligence could have, and one surprising pioneer in this was the world of fragrance.
Many businesses want a different smell in their retail scent diffuser every season, and with that in mind, IBM Research and Symrise collaborated on the development of Philyra in 2018, who worked with cosmetic company O Boticário to create the first AI scent that was brought to market.
Philya works through vast amounts of machine learning and data processing. Symrise has a collection of 1.7m formulas, as well as demographic information about which smells are popular with which groups of people.
Finally, after processing this information, Philyra is given an instruction to create a fragrance
recipe for a smell that will work with a specially chosen target audience.
It was formulated both on its own and with adjustments made by a professional perfumer and was sold as Eggeo ON.
The responses were somewhat mixed, to say the least, and a lot of the concerns relate to a common issue with exactly how AI develops its fragrances.
Given that it has no sense of smell, it works by processing all of the smells that worked well before to find profiles and patterns within them, which is not the same process that a perfumer would use.
Perfumers train their noses for years to build up the requisite experience of how to evoke emotions through their scents, which is why perfuming is sometimes treated more like an art than traditional product development.
The bigger issue is that the algorithm does not understand why certain smells are preferable, just that they are and as a result a perfume AI needs to be carefully monitored by an expert to ensure that it is learning the right lessons from its training data.
This is a problem all AI systems have, but it is far more noticeable with fragrances.
