by Zaara Abbas / May 11, 2024
Scientists have made an exciting leap in developing ultra-miniaturized artificial intelligence (AI) that could revolutionize how we monitor conditions like diabetes. Researchers from the National Institute for Materials Science (NIMS) and Tokyo University of Science have created a new AI device tinier than ever before; one that utilizes the vibrations of just a few molecules to make predictions.
Traditional AI models rely on large hardware and high computing power to crunch data and make forecasts. But this new molecular AI takes a completely different approach, demonstrating that even a scattered cluster of molecules can perform brain-like calculations with incredible efficiency. The key lies in a process called reservoir computing using surface-enhanced Raman scattering. By modifying how hydrogen ions bind to a handful of organic p-mercaptobenzoic acid (pMBA) molecules, the scientists could encode data inputs that alter the molecules' vibrations. These molecular vibration patterns then serve as the "memory" for processing information and making predictions.
To test their completely new technology, the researchers deployed the molecular AI to predict fluctuations in blood glucose levels for diabetic patients, a critical healthcare need. After "learning" 20 hours of data on one patient's levels, the molecular AI could then forecast levels 5 minutes into the future with around 50% less error than traditional AI models. This innovative work demonstrates that even an extremely small quantity of molecules can perform AI computations as good as today's bulky computer hardware.
With continued progress, this molecular reservoir computing could enable the creation of ultra-compact, low-power AI devices that seamlessly integrate sensor data from wearables or implants. The potential impact is immense, especially for conditions like diabetes that require continual monitoring. Molecular AI chips could one day allow continuous tracking and automated insulin response entirely at the patient's side without a conventional computing rig. More broadly, this breakthrough points the way toward common AI embedded into almost any object for a wide range of applications we can scarcely imagine today.
Citations:
National Institute for Materials Science. (2024, May 11). Revolutionary AI device mimics human brain with few-molecule computing. SciTechDaily. https://scitechdaily.com/revolutionary-ai-device-mimics-human-brain-with-few-molecule-computing/
Comments