Scientists have made significant progress towards human-like intelligence in machines a single artificial neuron capable of performing the functions of multiple brain areas. These developments could enable robots to perceive, learn, and act with adaptive and responsive abilities previously thought to be unique to living brains.
The device, called a transneuron, can switch roles between brain cells involved in vision, planning and movement. Developed by an international team led by Loughborough University in collaboration with the Salk Institute and the University of Southern California, the transneuron represents a major step forward in the field of neuromorphic computing – a technology designed to replicate the performance and flexibility of the brain in hardware.
Traditional artificial neurons typically perform a single, narrowly defined function, requiring large networks to handle even basic tasks. The new transneuron overcomes this limitation.
By precisely adjusting electrical settings, a single unit can reproduce neuronal firing patterns from three different areas of the brain, achieving 70-100% accuracy. These ranged from steady bursts to fast bursts, closely mirroring the variability of biological neurons.
In addition to imitating neural activity, the transneuron performs basic computational functions. The device changes its firing frequency based on input signals and responds differently when two signals reach each other or are out of sync – an ability known as time coding. Typically, replicating this requires the cooperation of many artificial neurons.
This possibility is made possible by a nano-scale element called a memristor. Silver atoms in the memristor change as current flows, creating and breaking conductive bridges that allow the device to retain memory of past signals and adapt its response, much like synaptic plasticity in the brain. Changes in voltage, resistance, or temperature further adjust the neuron's behavior without software intervention.
The next step involves integrating a network of transneurons to create a “cortex on a chip.” Such systems could form the basis of artificial nervous systems in robots, enabling perception, adaptation and learning in real time. These networks provide continuous, energy-efficient learning and dynamic responses that overcome the limitations of current artificial intelligence systems.
The technology could eventually connect directly to the human nervous system, offering new tools to study neural communication, treat neurological disorders, and even improve brain function. Transneurons can serve as experimental platforms to study neural communication or the emergence of consciousness in controlled environments.
Published research signals a shift in artificial intelligence from software that simulates brain functioning to hardware that behaves similarly. With the ability to adapt, compute and change roles on demand, the transneuron could become the building block of future self-learning robots and next-generation computing systems that operate with the efficiency and flexibility of biological brains.
