Deo Atharva, Lee Jungmin, Gao Dawei, Shenoy Rahul, Haughn Kevin Pt, Rong Zixuan, Hei Yong, Qiao D, Topac Tanay, Chang Fu-Kuo, Inman Daniel J, Chen Yong
Departments of Mechanical and Aerospace Engineering, Electrical and Computer Engineering, Materials Science and Engineering, California NanoSystems Institute, University of California, Los Angeles, CA, 90095, USA.
Department of Aerospace Engineering, University of Michigan, Ann Arbor, MI, 48104, USA.
Commun Eng. 2025 Jun 16;4(1):109. doi: 10.1038/s44172-025-00437-y.
Neurobiological circuits in the brain, operating in Super-Turing mode, process information while simultaneously modifying their synaptic connections through learning, allowing them to dynamically adapt to changes. In contrast, artificial intelligence systems based on computers operate in Turing mode and lack the ability to concurrently infer and learn, making them vulnerable to failure under dynamically changing conditions. Here we show a synaptic resistor circuit that operates in Super-Turing mode, enabling concurrent learning and inference. The circuit controls a morphing wing to reduce its drag-to-lift force ratio and recover from stalls in complex aerodynamic environments. The synaptic resistor circuit demonstrates superior performance, faster learning speeds, enhanced adaptability, and reduced power consumption compared to artificial neural networks and human operators on the same task. By overcoming the fundamental limitations of computers, synaptic resistor circuits offer high-speed concurrent learning and inference, ultra-low power consumption, error correction, and agile adaptability for artificial intelligence systems.
大脑中的神经生物学回路以超图灵模式运行，在处理信息的同时通过学习动态修改其突触连接，使其能够动态适应变化。相比之下，基于计算机的人工智能系统以图灵模式运行，缺乏同时进行推理和学习的能力，这使得它们在动态变化的条件下容易出现故障。在此，我们展示了一种以超图灵模式运行的突触电阻器电路，它能够同时进行学习和推理。该电路控制一个变形机翼，以降低其阻力与升力之比，并在复杂的空气动力学环境中从失速状态恢复。与执行相同任务的人工神经网络和人类操作员相比，突触电阻器电路表现出卓越的性能、更快的学习速度、更强的适应性以及更低的功耗。通过克服计算机的基本局限性，突触电阻器电路为人工智能系统提供了高速并发学习和推理、超低功耗、纠错能力以及敏捷适应性。
