The Evolution of Understanding Odor
2024
Unveiling the Mysteries of Smell
In the realm of senses, we've mastered the art of splitting light into colors and sounds into tones. Yet, the world of odor has long remained an enigma. Is it too complex, too personal to map? Surprisingly, the answer is no.
Recent advancements have revolutionized our understanding of smell, drawing on collaborations between neuroscientists, mathematicians, and AI experts. Unlike our intuitive grasp of colors and sounds, the world of smells has eluded easy categorization. But now, a groundbreaking 'odor map' published in Science has changed the game.
This map isn't just a catalog of smells; it's a set of rules for understanding them. Just as a geographical map tells you that Buffalo is closer to Detroit than to Boston, the odor map reveals that the smell of lily is closer to grape than to cabbage. More remarkably, it allows us to pinpoint any chemical's location on the map, predicting how it smells based on its properties. It's akin to a formula that, given a city's population size and soil composition, can precisely locate Philadelphia's coordinates.
The Evolution of Odor Perception
But how do our noses create this 'odor space'? Unlike Newton's study of light or the analysis of pitch, smell defies simple tools like tuning forks. Early attempts to categorize odors, like Linnaeus' and Haller's schemes, lacked empirical rigor. They were more about intuition than data.
One bold attempt, by Hans Henning in 1916, proposed an 'odour prism' with six vertices corresponding to primary odors. While Henning's theory was flawed, it sparked a quest for the underlying principles of smell. Later efforts, like Susan Schiffman's odour maps in the 1970s, provided valuable insights but fell short of a complete solution.
The Rise of AI in Decoding Odors
Enter the age of AI. In 2017, the DREAM challenge brought AI into the fold, leading to models that could predict odors with impressive accuracy. These 'random forests' of AI models can be complex, mimicking human judgment in convoluted ways. They can predict that a chemical smells like rose based on a multitude of factors, not just its structural properties.
The Osmo Revolution: Giving Computers a Sense of Smell
Osmo, a startup born from Google Brain's digital olfaction group, is at the forefront of this revolution. Led by Alex Wiltschko, Osmo is training AI models to understand smells using simplified molecular graphs. These models, inspired by the brain's processing, can compute distances and angles in 'odour space', predicting how a chemical will smell based on its relationship to others.
The Future of Odor Science
The odour space isn't a simple geometric shape like a circle or prism. It's more like a rugged landscape of chemical continents, each representing a different aspect of human ecology. Two chemicals might smell alike not because they're structurally similar, but because they play similar roles in nature.
In conclusion, the study of smell has evolved from introspective musings to data-driven AI models. While we're far from fully understanding the geometry of odor, these advancements have brought us closer than ever. Perhaps smell has been the last great sensory mystery because its mathematics are the most esoteric. But with the ongoing work of researchers like those at Osmo, we're unlocking the secrets of scent, revealing a world rich in meaning and possibility.
