🤓 Artificial Intelligence is Now Able to "Sniff"

Osmo, a startup founded by former Google Research team members, has developed an AI model that describes smells as accurately as humans.

The AI was tasked with characterizing about 5,000 known odorants. The data input was the odorants' molecular structure, and the output predicted which words best describe their smell.

The model was labeled "The Principal Odor Map." For example, it described the compound "aldehyde C-14" with the words "peach," "apricot," "fruity," and "sweet."

🧪The result was a diagram of the atoms and molecules of many scents. The model could identify dozens of pairs of structurally dissimilar molecules with intuitively similar odors and predict the smell properties for 500,000 additional molecules without needing to synthesize them first.

The neural network not only understands the relationship between odors and their formulas, but also organizes smells into groups. Floral scents are in one part of the map, meat scents in another; lavender is closer to jasmine than meat scents on the map.

The developers checked the accuracy of the AI-based "sniffer" by comparing it with human olfaction. They asked a panel of trained volunteers to describe 400 odorants. The AI-generated descriptions mostly matched the panel's and, in many cases, were more accurate than the human assessment.

❓ Why is This Useful?

• Digitizing scents is handy in cosmetics, food, and medicine. For example, the model can help test and restore the sense of smell in COVID-19 survivors.

• Technologists will be able to synthesize new fragrances. The startup Osmo received a $3.5 million grant from the Bill & Melinda Gates Foundation.

• The model can warn of leaks or accidents in factories dealing with hazardous chemicals.

"Computers have been able to digitize vision and hearing, but not smell – our deepest and oldest sense," says Alex Wiltschko, founder of Osmo.

#startup @hiaimediaen