Mind, blown! AI has got quite good at seeing and listening, what with recent advancements in cameras and mics. However, when it comes to smells, this is one human sense that the technology has not got around as much.
This may change if Google has their way.
Researchers at the search engine giant are trying to develop a neural network that helps an artificial intelligence system identify the smell characteristics of a molecule.
The Google Brain Team created a data set of some 5,000 molecules that were identified by perfumers. These molecules were then labelled with rather colorful descriptions ranging from “buttery” to “tropical” and “weedy.”
About two-thirds of the dataset was used to train the AI to associate molecules with the descriptors they often received. And the researchers then used the remaining scents to test the artificial intelligence system.
And guess what? It passed the test.
The algorithms able to accurately predict the smells of the molecules based on their structures. The researchers used what is called a graph neural networks (GNNs), a deep learning model that takes graphs as inputs.
Of course, there were a few caveats here, which is what make the science of small so tricky. The most obvious one is that people might describe the same scent differently. Things get even more complicated when scents are combined.
But the team is hopeful of the application of their research in the future. They want to create solutions for digitalized scent creations, and perhaps even build solutions for those without a sense of smell.
Oh, the sweet smell of success!