Our artificial intelligence, in reality, lacks a “brain.” It understands nothing, does not engage in thinking, and cannot perceive; even the meaning of text and images eludes it. Current AI systems are trained using vast amounts of data, with each training iteration adjusting the neural network’s data reception and output at every layer to provide more accurate answers. All AI accomplishes is generating responses resembling AI human brain “thinking” based on extensive training data.
For instance, ChatGPT possesses no geographical knowledge but can inform you about the capital of France. Of course, this is contingent upon the inclusion of the words “France,” “capital,” and “Paris” in its training data. When responding to such queries, it merely analyzes the relationships between words—France, capital, and Paris often co-occur.
In light of this, scientists envision simulating the mechanisms of human brain information processing, developing computational chips and corresponding hardware systems that enable computers and AI to think like humans—a true form of “artificial intelligence.”
Make AI more like the human brain
A research team at the University of Science and Technology of China has developed brain-like neuron devices using vanadium dioxide phase-change films, known as neuromorphic chips. These chips simulate dynamic neural synapse connections when exposed to external stimuli. The simulation showcases the brain-like neural system’s ability for multi-channel signal transmission and processing.
If these neuromorphic chips are interconnected to form an artificial neural network in the future, they could emulate the workings of the human brain’s neural network. Computers and AI equipped with such neural networks would then think like humans.
What is a Brain-inspired Chip?
Inspired by the human brain’s information processing methods, these devices simulate the brain’s neural system, creating a virtual brain. This could lead to the fusion of machines and living organisms through brain-machine interfaces, culminating in the creation of a superbrain combining artificial and biological brains. In essence, neuromorphic research involves the use of devices, such as neuromorphic chips, to simulate the neural system of the human brain.
Neuromorphic chips replicate the human brain. One might ask: why simulate the human brain with these chips? Neuromorphic research has achieved applications in diverse fields. These include multimodal target tracking, robot spatial positioning, multiscale brain simulation, and collective control of robot groups.
Neuromorphic research is progressing, and with increasingly intricate applications, scientists are directing their attention to enhancing the intelligence of computers and AI to align more closely with the human brain. A proposed solution to this is the use of pulse neural networks. These networks, unlike conventional neuromorphic research methods, closely emulate the functioning of the human brain. They bring benefits in terms of efficiency, computational capacity, and applicability. The construction of pulse neural networks relies on the foundation provided by neuromorphic chips.
Synaptic Connections of the human brain
As early as 2005, the United States proposed a neuromorphic research plan. Now, China, Japan, and European countries have also embarked on neuromorphic research. Despite the current popularity of neuromorphic research based on neuromorphic chips and some achievements, the path to large-scale commercial operation of neuromorphic chips is still long.
Silicon-based artificial intelligence earns its designation due to the prevalent utilization of silicon in chip production. Simulating the neural system of the human brain, particularly synaptic connections, using such silicon-based chips is challenging.
The University of Science and Technology of China’s team has developed a novel neuromorphic chip using vanadium dioxide phase-change films. Vanadium dioxide undergoes a reversible conductivity transition between insulators and metals near room temperature, making it an ideal material for neuromorphic chips simulating synaptic connections. Scientists in this study used a novel method to directly simulate synaptic connections between human brain neurons, overcoming production bottlenecks. The research was published in the October 4, 2023, issue of “Science Advances.
Read Also: ChatGPT: The AI Chatbot That Can Build Software in Minutes
