A non-biological system based on a network of silver nanowires is capable of learning and remembering in a way similar to the human brain, according to a new study. The research focuses on a system that uses a network of nanowires to mimic neurons and synapses in the brain: tiny wires one-thousandth the width of a human hair, made of silver and encased in plastic. are surrounded. Artificial networks “remember” past electrical signals for at least seven steps, which is the average number of items humans can keep in working memory at a time.
In new research led by the University of Sydney in Australia, a team of experts led by researcher Alon Loeffler succeeded in designing an “artificial brain” based on silver nanowires that developed learning and memory capabilities similar to those of the human brain. does. A new study recently published in the journal Science Advances describes the innovation and the results obtained.
Artificial systems are more “human” than AI
According to an article published in The Conversation by the lead authors of the research, the scientists chose an approach toward non-biological systems that are more like the human brain than artificial intelligence (AI). The reason is that while AI systems are highly capable, for example, vastly outperforming humans in large-scale data pattern recognition tasks, at the same time they are not as smart, at least in some areas, as of now.
In the Australian researchers’ view, AI systems are not structured like our brains and do not learn in the same way. Also, they expend vast amounts of energy and resources for training, their ability to adapt and function in dynamic or difficult-to-predict environments is worse than ours, and they do not have the same memory capacity as humans.
Because of this, the new “artificial brain” is based on a system that uses a network of nanowires to mimic the neurons and synapses of the human brain. It is a non-biological system made up of tiny wires, about one-thousandth the width of a human hair. They are made of a highly conductive metal, such as silver, and covered with an insulating material such as plastic.
The nanowires self-assemble to form a network structure similar to a biological neural network. Like neurons, which have an insulating membrane, each metal nanowire is coated with a thin insulating layer. By stimulating the nanowires with electrical signals, ions move through the insulating layer and into a neighboring nanowire, just as neurotransmitters do through synapses in our brains.
learning and memory
The study shows that it is possible to selectively strengthen or weaken synaptic pathways in nanowire networks. This mechanism is similar to learning in the human brain: in the process, the output of the synapse is compared to a desired outcome. In this way, the synapse is strengthened if the signal orients to the desired outcome or shortened if the signal is not expected.
This is the kind of structure in which humans learn: when we acquire knowledge, the brain rewards us and motivates us to seek new knowledge. Conversely, when we take the wrong path, error marks us and forces us to try again. Under this criterion, the researchers verified that the artificial network responded to dynamics of being “rewarded” or “punished” depending on its outcome.
At the same time, self-organizing networks of tiny silver wires manage to remember in the same or very similar way to the human brain. For example, by applying the same test to this system that is used to measure working memory in humans based on a series of electrical stimuli, the scientists found that the network could “remember previous cues for at least seven steps”. remembers”. It is precisely this number of steps that is considered the average number of elements that humans can simultaneously hold in working memory, that is, the number that is used to solve habitual tasks.
Finally, the research shows that it is possible to implement the characteristics required for intelligence, such as learning and memory, in physical, non-biological hardware. Furthermore, these nanowire networks are different from the artificial neural networks used in AI, although they may still lead to so-called “synthetic intelligence”. As a result, at some point they may be able to develop interaction and communication strategies that are more “real” from a human perspective than artificial intelligence systems.
Reference
Neuromorphic learning, working memory, and metaplasticity in nanowire networks. Alon Loeffler et al. Science Advances (2023). DOI:
