Researchers have shown how to maintain a network nanowires in a state right at the edge of chaos. This is a breakthrough that could be used for artificial intelligence (AI), which acts like the human brain.On a simulation of a nanowire, the team experimented with different levels of electricity to find a balance between too much and too little. The outputs of the network were not useful if it was too low. If it was too high, they were useless."We discovered that the network does the same thing over again if it is pushed too slow without learning or developing. Joel Hochstetter, a physicist at the University of Sydney and lead author of the study, says that if we push it too hard or fast the network becomes unpredictable and erratic.Scientists report that simulations kept within the range of these extremes yielded the best results. These findings suggest that nanowire networks could eventually produce brain-like dynamics.Conceptual image showing randomly connected switches. (Alon Loeffler).Zdenka Kuncic, a physicist at the University of Sydney, Australia says that some theories in neuroscience suggest that the human mind might operate at the edge of chaos or what is known as the critical state. "Some neuroscientists believe it is in this state that we achieve maximum brain performance."Nanowires of 10 micrometers in length and no more than 500 nanometers thick were used for the simulations. They were placed randomly on a 2-dimensional plane. For comparison, human hairs can measure up to 100,000 nanometers in width.The problem that the network was faced with in this instance was to transform a simple waveform into something more complex. In order to solve the problem, the network had to adjust the frequency and amplitude of the electrical signals to find the best state.Nanowire networks combine two systems to manage memory (the equivalent in computer RAM) as well as operations (the equivalent in a computer CPU). They are memristors, able to recall past signals and change their output in response to them.Hochstetter says, "Where wires overlap, they create an electrochemical junction like the synapses among neurons."The algorithms usually train the network to find the best routes, but this time the network did it all on its own.Hochstetter says that electrical signals sent through the network find the best route to transmit information. This architecture also allows the network's to'remember" previous routes through the system.This could lead to significant energy savings as the networks are able to train themselves using the most efficient processes. It will become increasingly important to ensure that artificial intelligence networks are as efficient and lean as possible as they grow.Scientists have demonstrated that nanowire networks can solve problems right at the edge of order and chaos. This is a step closer towards AI that thinks like our brain.Kuncic says, "What's so fascinating about this result? It suggests that these types nanowire networks can also be tuned into regimes that have diverse, brain-like collective dynamics which can be used to optimise information processing."Nature Communications published the research.
