Intel's Neuromorphic Chip Mastering 'Smell': Detection of 10 Hazardous Chemicals

Researchers Advance AI's Sense of Smell with Neuromorphic Chips

Though scent is one of the most challenging senses for AI to comprehend, researchers are making notable strides. Recently, a team from Intel and Cornell University trained a neuromorphic chip, Intel's Loihi, to identify the scents of ten hazardous chemicals. This breakthrough could pave the way for "electronic noses" and robotic systems capable of detecting weapons, explosives, narcotics, and diseases.

The researchers developed an algorithm inspired by the brain's olfactory circuit. When we inhale, scent molecules activate olfactory cells in our noses, which relay signals to the brain's olfactory system, generating electrical pulses. The team successfully mirrored this process in Loihi's silicon circuits.

Intel reports that the chip can accurately recognize ten odors, including acetone, ammonia, and methane, even in the presence of overpowering scents. Remarkably, Loihi mastered each scent with just a single sample—significantly more efficient than conventional deep learning methods, which typically require up to 3,000 samples for similar accuracy.

In a recent article in Nature Machine Intelligence, Nabil Imam, a senior research scientist in Intel Labs' neuromorphic computing group, highlighted this research as a prime example of the intersection of neuroscience and artificial intelligence.

Intel and Cornell are not alone in this endeavor; the Google Brain Team collaborates with perfumers to map scent molecules to their perceived aromas. Meanwhile, Russian researchers employ AI to identify dangerous gas mixtures, and some scientists have aimed to replicate the aroma of an extinct flower using machine learning.

"Understanding how the brain's neural circuits navigate complex computational tasks will provide vital insights for developing efficient and powerful machine intelligence," Imam states. By unlocking the intricacies of how the brain processes scents, we may revolutionize the design of AI technology.