| If anyone understands the power of scent, it's Patrick Süskind, German author of the 1985 historical novel Perfume: The Story of a Murderer. The novel presented readers with a poignantly depicted world, dominated by explorations of this relatively ignored sense. "The persuasive power of an odor cannot be fended off, it enters into us like breath into our lungs, it fills us up, imbues us totally," he wrote. Scent triggers strong feelings of affection or contempt, disgust or lust, love or hate, such that "he who ruled scent ruled the hearts of men."
You don't need to be in Süskind's 18th-century France to know this power. A whiff of chalk can transport you to second grade. Walking by a bakery can put you in your grandmother's kitchen. A spritz of perfume can conjure an old friend and bring tears to your eyes.
Smell — the ability to sense chemicals in the air — is intimately tied to our memories and emotions, which makes it perhaps our most powerful sense. It's also our most ancient. But relative to other senses, such as vision and hearing, little is known about it. That's partly because it isn't traditionally viewed as important to our survival. Unlike hedgehogs or rodents, humans can do fine without it. But, in fact, it is deeply tied to our experience of the world: the taste of food, navigating our social environments, threats drifting through the air as smoke.
And we certainly miss it when it's gone. Conditions including Covid-19, which is known to steal people's ability to smell, have brought more attention to the sense. But scents are abstract and ephemeral, without clear definition. We often have trouble describing them, at least in English, and lean hard on generalizations; something might smell like "Christmas" or a "sewer." A scent is likely made up of not just one molecule, but many.
Those many molecules waft in, like fleeting and intangible bouquets, and click onto about 400 olfactory receptors in the nose. Different patterns of interactions help the brain discern more than a million different smells. If scientists succeed in further decoding this sense, they may gain insight into how the modern brain carries out such an ancient protocol, and may also learn something about memories, language and emotions. Eventually, this could help them develop a digital nose — a device that might help in detecting threats, diagnosing disease or aiding those who are hard of smelling.
What's New and Noteworthy It was only a few years ago that researchers first glimpsed, at a molecular level, what it looks like when a receptor binds to an odor molecule, creating a flexible structure. But their findings were made in an insect called the jumping bristletail. It was a few more years before they determined the three-dimensional structure of a human olfactory receptor bound to an odor molecule. Researchers say that seeing this structure is an important step toward understanding how our sense of smell guides our experiences.
We rarely experience or perceive the scent of a single molecule, which makes it really hard to predict what a single compound would smell like by itself. A few years ago, a group of researchers at Google figured out a way to use machine learning to compute single-molecule smells by looking at their structures. They found that molecules related to each other biologically tend to smell similar, even if their structures differ. Those that are often found together in nature, say, on the inside of an orange, also tend to smell similar to one another.
Not only are scientists using artificial intelligence to decode our sense of smell, but they are also using our sense of smell to inspire AI. I recently covered this theme in an article about the relationship between AI and the human brain. Some of today's highly successful artificial neural networks were inspired by the visual system, which enables them to process information in increasingly complex ways. But they have limitations, namely that they aren't very good at working in unstructured and changing environments. While things we see are very structured, things we smell are not. By designing AI that better mimics how the brain processes a different, more abstract sense, we might be able to improve the way these systems work, researchers say. | 
