AI System Outperforms Humans in Odor Identification
A recently developed computer-trained scent model has surpassed human capabilities in odor recognition. Furthermore, when it came to examining 500,000 potential odor molecules that had never been synthesized before, it efficiently accomplished a task that would typically require 70 person-years of effort.
The Complex Journey to Replicate Smell
While machines have made significant strides in replicating human sight and taste, they have faced challenges in developing a comparable sense of smell. Electronic noses have been able to detect cancer in blood cells and analyze the air around wastewater treatment plants, but achieving a genuine computer-based sense of smell has proven elusive.
This could be attributed to the fact that our noses possess 400 olfactory receptors, a far greater number than the four receptors we utilize for vision and the roughly 40 receptors employed for taste.
Collaborative Research to Enhance Computerized Perception
In an effort to level the playing field in terms of computerized sensory perception, a team of researchers at the Monell Chemical Senses Center at the University of Pennsylvania, in collaboration with Osmo (a spin-out of Google DeepMind), conducted a study.
They developed a neural network-based system capable of analyzing an odor molecule and providing a human-like description of how that molecule should smell. This AI system led to the creation of what the researchers have coined as a Principal Odor Map (POM).
Joel Mainland, a senior research co-author from Monell, noted, “In olfaction research, the mystery of what physical characteristics give an airborne molecule its perceived odor to the brain has persisted. However, if a computer can discern the connection between the shape of molecules and how we ultimately perceive their scents, scientists could use this knowledge to advance our understanding of how our brains and noses collaborate.”
Enhancing Products with Newfound Understanding
This understanding could assist researchers in improving products like mosquito repellents or deodorizers, among other potential uses.
To teach the system, the research group provided it with the molecular makeup of 5,000 odorants along with a set of descriptions characterizing odors like “minty” or “musty.”
Additionally, they enlisted 15 individuals to evaluate 400 different odors and assigned them a vocabulary of 55 words to describe each scent.
In testing, the AI system outperformed the panelists marginally. However, an even more remarkable outcome emerged.
Read the original article on: New Atlas
Read more: AI: 2023’s Dominant Tech
