Humans can detect one trillion different smells, no wonder it is often considered one of the most powerful senses of the body.
A group of researchers at Google want to harness that supernatural power through AI. The Google Brain Team is now training AI to recognize the smell of items based on their molecular structure.
Previously, scientists have struggled to find the correlation between the smell of a molecule’s structure and its scent. While they can just go a little close and smell the substance, the goal has been to identify the odor just by looking at it.
In the paper published in Arxiv, researchers at Google Brain explain how they are achieving the unachievable. The researchers asked perfumers to identify over 5000 molecules and fed two-thirds of the data set to a graph neural network (GNN).
The team labeled known terms with the associated smell such as buttery, tropical, weedy, etc. Later, Google researchers used the remaining scents to test the AI. After a few tests and trials, the algorithm successfully managed to predict molecules’ smells based on their structures, something scientists have been trying to achieve for years.
However, the algorithm is not even close to ready. As the Wired points out, GNN’s biggest let down is with chiral pairs; two molecules having the same number of atoms and bonds but arranged as mirror images, therefore having different smells.
Another key area missed by the GNN is what happens when two or more molecules are combined to create an entirely different smell. The human perception is also a key area where GNN fails; for instance, two molecules might smell different, however, people would point out only one odor.
Despite the few caveats with the AI, Alexei Koulakov, a researcher at Cold Spring Harbor Laboratory points out that this development, “could form the basis for improvements of this and other algorithms in the future.”