A First Look into Artificial Neurons

Despite the rapidly advancing artificial intelligence (AI) industry, there is one part of the body that still remains unique to humans: the brain. The day when our brain’s ability to retain memories and perform various cognitive functions in a matter of seconds is replicated synthetically has been thought of as a distant reality. A recent prototype of an artificial neuron developed by a team of French researchers has revealed that the possibility of a synthetic brain is more tangible than previously thought.

The objective of the study was to reproduce the brain’s efficiency into artificial systems. Though the brain is responsible for carrying out complex tasks on demand, it actually needs very little energy to do so. This efficiency can be credited to neurons, brain cells that make use of the ion channels in their porous membrane to communicate information. These channels open and close in response to stimuli from surrounding neurons, and the resulting ion flow produces an electric current. From this current, action potentials are released, which then enable the neurons to communicate with each other. While AI can match the brain’s performance of complex tasks, it requires a great deal of energy to do so. Therefore, researchers have been trying to develop artificial networks that mimic the brain’s use of ions, rather than electrons, to be as energy-efficient as the brain.

Interestingly, the most promising approach to achieving this goal lies in the field of nanofluidics, which studies the behavior of fluids within structures less than 100nm wide. A study by a team of researchers from ENS Laboratoire de Physique of the University of Paris built a prototype that uses a single layer of water molecules within thin graphene splits to mimic the functions of a neuron. This model parallels features of the neurons in our brain: the graphene incisions represent the ion channels where the ion flow is observed.

Furthermore, scientists have observed that when a neuron is exposed to the electric field, ions of the water layer arrange themselves into clusters and exhibit a property called the memristor effect, which is when the clusters are able to retain parts of the stimuli they received previously. When the voltage was removed, the clusters were able to retain some of the stimuli. This is especially significant because it raises the possibility of artificial memories, yet another way that AI can be advanced to further mimic our traits.

This prototype was largely made possible by advancements in the field of nanofluidics that allowed water to exist in a single layer. These monolayer electrolytes open the door to attaining bioinspired functionalities through ion-based transport systems. Next, scientists hope to apply these findings in their understanding of deep neural networks, which use artificial neurons connected by artificial synapses to recreate the functions of the brain. This would allow them to produce systems that can recognize and process certain stimuli. Not only is this promising for future biological implementations in AI, but it also demonstrates the cooperation between different fields needed to make something like a synthetic neuron possible.

While this is exciting news for the AI industry, it is only the beginning. Scientists will continue their research on electronic memories and collaborate with a team from the University of Manchester, UK, to rule out the efficiency of ion-based systems. The implementation of the artificial neuron in a larger artificial network or basic machine learning algorithms is also continuing to be explored. While AI has always been a step behind humans in energy efficiency, the possibility of the artificial neuron can allow researchers to focus on adapting other aspects of AI to match humans, such as the ability to store memories. The future of AI remains uncertain, but as AI is being developed to become increasingly comparable to humans, it is likely we will see some familiar characteristics.