Die Ähnlichkeit zwischen dem digitalen Zwillingsgehirn und dem menschlichen Gehirn beträgt bereits fast 60 Prozent.

In July last year, a patient with motor dysfunction had a 256-channel flexible brain-computer interface designed by Brain Tiger Technology implanted in the top of the cerebral cortex. After two weeks of training, the patient could operate social media applications and control a smart wheelchair with their mind. In December of the same year, a patient with language dysfunction had the same device implanted. After training, the patient was able to communicate in Mandarin at a speed of 50 characters per minute (about one-third of the average speaking speed of ordinary people), with a delay of less than 100 milliseconds.

These two cases mark the first real-time decoding of Mandarin Chinese internationally, filling a gap in the relevant domestic field.

The human brain, a complex and precise system, holds the ultimate code for intelligent operation and is also the core prototype for many scientists to explore digital twin brains and brain-inspired intelligence.

What are the differences between twin brains and brain-inspired intelligence? What is the current research progress? What is the significance of twin brains and brain-inspired intelligence to humans? What are the application scenarios? At the 2025 World Artificial Intelligence Conference (WAIC 2025), a symposium on brain science was held. Experts from research institutions such as the Chinese Academy of Sciences, Fudan University, and Zhejiang University discussed these topics, presenting a blueprint for the future of brain science research full of infinite possibilities.

For humans, this not only means that the diagnosis and treatment of brain diseases may ushers in a new paradigm of "digital experiments" and precise intervention. For example, optimizing treatment plans by simulating pathological states through twin brains. It also indicates that brain-inspired intelligence will serve all aspects of life in a more efficient and biologically intelligent way.

Twin Brain: Dividing the "Brain" into 500,000 Parts

Although the human brain weighs only about three pounds and consumes only one ten-thousandth of the energy of a supercomputer, it is composed of 86 billion neurons, and each neuron is connected to more than 1,000 other neurons. It can be regarded as the most complex and mysterious nervous system.

"What I cannot create, I do not understand." This famous quote by the American physicist Richard Feynman also applies to the field of brain science research.

After seven years of research, the research team led by Zheng Qibao, a distinguished professor and doctoral supervisor at Fudan University and an expert in information and communication, built a digital twin brain (DTB) simulation platform at the scale of the whole human brain.

The digital twin brain can help researchers understand how information is transmitted and processed in the brain. In Zheng Qibao's view, the essence of the digital twin brain is to simulate the human brain, mainly to promote the development of artificial intelligence and improve the diagnosis and treatment of brain diseases.

How is a digital twin brain formed?

Zheng Qibao introduced that first, the most detailed available brain imaging data is used to build a mesoscale brain model. The so-called mesoscale means dividing the brain image into 23,000 blocks, each about a cube with a side length of 3 millimeters. Then, the brain imaging data obtained by magnetic resonance imaging technology is used to clarify the connections between these "small blocks".

These data alone are not enough. Each cube contains hundreds of thousands or even millions of neurons, and it is still difficult to fully obtain the subtle connections and connection strengths between them with current technology. So the research team used the data assimilation method in mathematics to solve this problem to a certain extent and promoted the construction of the digital twin brain.

Previously, the team conducted visual and auditory experiments. The Pearson correlation coefficient (used to quantify the similarity between the digital twin brain and human brain activities, with a higher value indicating a closer activity pattern) between the digital twin brain and human brain activities reached 0.63 in the visual experiment and 0.57 in the auditory experiment. This means that the activity pattern of the digital twin brain is highly similar to that of the human brain, and its processing responses to visual and auditory information are already somewhat close to the real brain's activity state.

Zheng Qibao revealed that the team has started the research and development of the "version 3.0" and plans to improve the cutting accuracy of the brain image to 1 millimeter. "The previous 23,000 blocks may become more than 500,000 blocks. The finer the division, the better the simulation will be in terms of cognitive performance. On the other hand, it is to improve the dynamic mechanism. Although the current simulation has significant brain functions, for example, when an image is input into the visual cortex of the digital brain, a time series similar to that of the biological brain can be obtained. But without stimulation, this digital brain is 'dead'." Zheng Qibao said that if one day the entire brain can perceive the world through the body's feedback, combined with vision, hearing, touch, etc., that will be a truly complete digital twin brain.

Brain-Inspired Research: Low Energy Consumption and High Intelligence like the Human Brain

Imagine that when you say "It's a bit cold" to a computer, it understands that you want to turn on the heater; when you spray insecticide at it, it responds "Toxic gas". This is when a computer has a "thinking ability" similar to that of the human brain.

If the digital twin brain aims to be "brain-like", then brain-inspired intelligence focuses more on "using the wisdom of the brain". However, current artificial intelligence systems generally face the problem of high energy consumption, often requiring tens of thousands of GPUs to operate. In contrast, the biological brain can achieve complex cognitive functions with low energy consumption. This huge gap is the key direction for brain-inspired intelligence to break through.

Since 2012, Pan Gang, a professor at the School of Computer Science at Zhejiang University and the director of the National Key Laboratory of Brain-Machine Intelligence, has led his team in the research of brain-inspired computers. They have successively developed the Darwin I and Darwin II brain-inspired chips. In 2023, they further developed the Darwin III brain-inspired chip, which has the largest number of neurons per chip internationally.

In his view, some mechanisms of the biological brain may not be the core of intelligence. For example, 90% of the energy consumption in the brain is for resting-state activities. If completely replicated, it will only increase system redundancy. The key to brain-inspired intelligence is to extract the efficient characteristics of the biological brain, such as sparse connections and dynamic activation, rather than mechanical imitation. "Just like when the human brain processes information, only a few neurons are activated. This 'work-on-demand' sparse model ensures efficiency and reduces energy consumption. Brain-inspired intelligence should learn from this idea, break free from the blind replication of biological brain details, and focus on extracting the core mechanisms that truly support intelligence, so as to achieve a better balance between energy consumption and performance in artificial intelligence systems."

Large-scale simulations require a large concentration of computing resources and may not necessarily produce the best results. However, biological evolution is amazing. The biological brain has very low energy consumption but strong capabilities. This makes people think about whether to change the existing computing methods and whether there is a more efficient computing system.

Li Guoqi, a recipient of the National Science Fund for Distinguished Young Scholars at the Institute of Automation of the Chinese Academy of Sciences, also believes that brain-inspired intelligence needs to absorb the structural and functional characteristics of the brain to build a more efficient model, but not all aspects of the brain can be used.

"The neurons in the biological brain have diversity and dynamic characteristics, while the computing units in existing large models are highly homogeneous." Li Guoqi's team is trying to design a network with "pulse communication" characteristics, allowing only a small number of neurons to be activated, thus significantly reducing energy consumption.

Not Only "Simulating a Living Brain" but also "Evolving in Interaction"

Currently, cross-species brain research is becoming an important breakthrough in brain science research. Exploring the mechanisms of different biological brains provides more "evolutionary references" for intelligent models. Although the brain structures of lower organisms are relatively simple, their neural circuit mechanisms can provide a "simplified reference" for human brain models. For example, the reasoning ability of zebrafish and the snake's self-awareness of its body both hide the basic logic of intelligence.

"True embodied intelligence should not stop at motion control but should, like living organisms, interact with the environment through body perception to achieve the dynamic reconstruction of the system itself." Zheng Qibao's team has started research on the zebrafish digital twin brain.

Ultimately, the goal of both the digital twin brain and brain-inspired intelligence is to help humans better understand the mysteries of the brain, break through the existing bottlenecks in artificial intelligence technology, and better serve fields such as medicine and scientific research.

Zheng Qibao said that the digital twin brain has shown potential in precision medicine. For example, for patients with Parkinson's disease, his team can simulate the specific pathological beta waves (a type of brain wave generated by brain neuron activities) in the basal ganglia (a group of gray matter nuclei deep in the brain) by inputting patient data. These are abnormal brain waves. Then, treatment plans such as deep brain stimulation can be tested in the digital space to provide target references for clinical precise intervention. "This is like creating a 'digital laboratory' for the treatment of brain diseases, which can significantly reduce the cost of clinical trial and error and enable more targeted treatment." Zheng Qibao said.

In the view of the participating experts, the digital twin brain is regarded as a bridge connecting biological intelligence and artificial intelligence. The next goal is to achieve a "living simulation", enabling the digital twin brain to be independent of external stimuli and have the ability to think autonomously. The development of brain-inspired intelligence emphasizes more on "interactive evolution" with the environment.

During the Spring and Autumn and Warring States periods, bronze tools evolved into iron tools, and the innovation of tools promoted the leapfrog development of agriculture and handicrafts. "Now, the development of brain science is in a similar stage to the Spring and Autumn and Warring States periods and will also bring about great social progress, reshaping the way we understand the world and ourselves." Li Guoqi said.

This article is from the WeChat official account "IT Times" (ID: vittimes). Author: Pan Shaoying. Editors: Hao Junhui and Sun Yan. Republished by 36Kr with permission.