Google designed an AI engine that can recognize scentsLast updated on Oct 25, 2019, 01:53 pm
For years, scientists have struggled to recognize the connection between molecules and odors.
They can determine the color of light using its wavelength, but there's nothing on smell yet. No one knows how exactly the structure of a molecule affects its odor.
However, to tackle this problem, Google Brain Team is now using an artificial intelligent engine, training it to recognize scents.
AI trained to recognize smells
Google's researchers took as many as 5,000 molecules identified by perfumers to be associated with smells like 'tropical, buttery, and weedy'.
Then, they employed more than half of that data to help its AI system, a graph neural network, learn the way to associate molecules with their possible scents or descriptors.
Once the system was trained, they put it to test.
What happened in the test?
In the test, the team fed the remaining molecules - about one-third of the original 5,000 - into the AI system, without their descriptors.
It processed the data and was able to use the molecules' structures to associate them with their respective scents.
Google describes this as a major first step, which could ultimately bolster our understanding of chemistry, sensory neuroscience, and fragrances.
However, there's still a long way to go
While Google has an AI that can recognize smells from molecules, it is important to note that it's just a basic development.
The composition of smells is way too complicated and Google would need to address a number of problems to make its system more sophisticated.
For instance, what happens when two scents are combined or two perfumers describe the same scent differently.
Other projects revolving around AI and smell
Along with Google, many other teams are working to leverage AI for scents and fragrances.
A team, for instance, is working on a way to transfer smells digitally, while another used machine learning to create the scent of an extinct flower.
Meanwhile, scientists in Russia are using AI to detect a mixture of gases that could be potentially dangerous to humans.