The researchers created a dataset of nearly 5,000 molecules identified by perfumers, who characterized the molecules with descriptions ranging from “buttery” to “tropical” and “weedy”. The team used about two-thirds of the data set to train its AI (a graphic neural network or GNN) to associate molecules with the descriptors they often receive. The researchers then used the remaining scents to test the AI - and they passed. The algorithms were able to predict the odors of the molecules based on their structures.
As Wired points out, there are a few limitations that make scent science so difficult. For starters, two people may describe the same fragrance differently, for example “woody” or “earthy”. Sometimes molecules have the same atoms and bonds, but they are mirrored and have completely different smells. These are called chiral pairs. Caraway and spearmint are just one example. It gets even more complicated when you combine fragrances.
Still, Google researchers believe that training AI to associate certain molecules with their scents is an important first step. This could have implications for chemistry, our understanding of human nutrition, sensory neuroscience, and the production of synthetic fragrances.
Google is not alone. At an AI show in London’s Barbican Center, scientists used machine learning to recreate the smell of an extinct flower. In Russia, AI is being used to detect potentially deadly gas mixtures, and IBM is experimenting with AI-generated perfumes. Some have even used our sense of smell to rethink how we design machine learning algorithms.