A novel scent model (AI system), trained by a computer, has outperformed humans in the identification of odors. In an analysis encompassing 500,000 potential odor molecules that had never been synthesized previously, this computer-driven model efficiently completed work that would have taken approximately 70 person-years to accomplish.
To level the playing field between human and computerized sensory input, a study led by researchers from the University of Pennsylvania’s Monell Chemical Senses Center, in collaboration with colleagues from Osmo (a spin-out of Google DeepMind), devised a neural-network-based system capable of evaluating an odor molecule and articulating, in human terms, how that molecule should smell. This AI system led to the creation of what the researchers have termed a “Principal Odor Map” (POM).
The AI system can discern the relationship between how molecules
Joel Mainland, a senior research co-author from Monell, stated, “In olfaction research, the question of what physical properties make an airborne molecule smell the way it does to the brain has remained an enigma. But if a computer can discern the relationship between how molecules are shaped and how we ultimately perceive their odors, scientists could use that knowledge to advance the understanding of how our brains and noses work together.”
As machines have made remarkable progress in emulating human senses like sight and taste, they have faced a certain lag in the development of a sense of smell. While electronic noses have emerged, capable of detecting cancer in blood cells and analyzing the air near wastewater treatment facilities, achieving a genuine computer-driven sense of smell has proven to be challenging. This difficulty may be attributed to the fact that our noses possess a staggering 400 olfactory receptors, a vast contrast to the mere four receptors we employ for vision and the approximately 40 receptors dedicated to taste.
AI system exhibited slightly superior performance compared to the panelists
This understanding has the potential to assist researchers in enhancing mosquito repellents, deodorizing products, and various other potential applications.
To instruct the system, the research team supplied it with the molecular makeup of 5,000 odorants and a set of descriptors characterizing odors, such as “minty” or “musty.” Additionally, 15 panelists were engaged in smelling 400 odors and were provided with a lexicon of 55 words to articulate each scent. During the tests, the AI system exhibited slightly superior performance compared to the panelists. However, an even more remarkable outcome emerged.
The model’s success at olfactory tasks it was not trained to perform is the most unexpected result, according to Mainland. “What opened my eyes was that even though we never taught it to learn odor strength, it could still predict outcomes accurately.” The method was then utilized by the scientists to map 500,000 odor molecules that have never been created; according to the team, this effort would take a person 70 years of smelling to complete.
According to the researchers, “new maps of the world supported by neural circuitry are frequently created and discovered as a measure of progress in neuroscience.” Each is only possible because researchers first built a map of the outside environment, and then they examined how brain activity changed depending on where a stimulus was placed.