Almost all neural networks that support modern artificial intelligence (AI) tools are based on living neuron computing models from the 1960s. However, a new model developed by the Center for Computational Neuroscience (CCN) at the Simmons Foundation Iron Institute in the United States suggests that this decades old approximate model did not capture all the computing power possessed by real neurons, and this older model may hinder the development of AI. The study was published in a new issue of the Proceedings of the National Academy of Sciences in the United States. The developers of the CCN model believe that a single neuron has much greater control over the surrounding environment than previously thought. The updated neural model may ultimately generate stronger artificial neural networks that better capture the power of the human brain. "Neuroscience has made significant progress in the past 60 years, and we now realize that previous neural models were still very primitive." Team leader Dmitry Chiklovsky stated that real neurons are much more complex and "smarter" than this overly simplified model. Artificial neural networks aim to mimic the way the human brain processes information and makes decisions, but the presented approach is still quite simple. These networks are based on neuron models from the 1960s and consist of ordered node layers. The network starts from the input layer node that receives information, then the intermediate layer node that processes information, and finally the output layer node that sends results. Usually, only when the total input received by a node from the previous layer exceeds a certain threshold, will it transmit information to the next layer. When training the current artificial neural network, information can only pass through nodes in one direction, and nodes cannot affect the information they receive from earlier nodes in the chain. In contrast, the new model views neurons as tiny "controllers" (referring to devices that can influence the surrounding environment based on collected information), because human brain cells not only passively transmit input information, but in reality they can also control the state of other neurons. Chiklovsky believes that this more realistic neural controller model may be an important step in improving the performance and efficiency of many machine learning applications. (Lai Xin She)