Project team of the Research on Key Scientific Issues, Key Core Technologies and Their Layout of Brain-Computer Interfaces: Current status and future outlook of the development of brain-computer interface technologies

Abstracts.  This paper reviews the development of brain-computer interface (BCI) technology, from the science fiction period in the 1920s to the technological explosion in the 21st century, and outlines the key advances and achievements of BCI technology at each stage. The paper describes the basic components of BCI technology and how they are categorized, including the different types of BCI systems (output, input, and two-way interactive) and signal acquisition methods (non-invasive, semi-invasive, and invasive), and explores the psychophysiological, technological, industrial, and regulatory challenges that BCI technology faces today. Finally, the article looks at the future direction of BCI technology, which is expected to make significant progress in hardware optimization, software integration, and algorithmic innovation in the coming decades, and may achieve a wide range of applications in a variety of fields, including healthcare, education, entertainment, and even human-computer interaction.

CTR. R318     Literature Identification Number. A

DOI.10.19524/j.cnki.10-1009/g3.2024.03.002

Normally, human beings communicate and interact with the external environment through the three-tier structure of "central nervous system - peripheral nervous system - muscles", while some diseases cause this process to be unable to proceed smoothly. In order to solve this problem, researchers began to study how to directly transform brain signals into human behavior bypassing peripheral nerves and muscles, and brain-computer interface technology was born.

"Brain-computer interface" (brain computer interface, BCI; also known as brain-machine interface, BMI), which is able to bypass peripheral nerves and muscles to establish a new communication and control channel directly between the brain and external devices^[1-3]. It has important potential applications in a variety of areas such as rehabilitation of patients with movement disorders and enhancement of human work capacity on a physical or cognitive level^[4-5]. With the development of brain-computer interface, artificial intelligence, biomedical engineering, neuroengineering and rehabilitation engineering, cognitive neuroscience and psychological science, etc., the connotation and extension of BCI have been enriched.

The concept of brain-computer interface can be traced back to 1924, when German doctor Hans Berger first recorded and named the human electroencephalogram (EEG), which laid the foundation for subsequent brain-computer interface research. During this period, brain-computer interfaces were viewed more as science fiction, with both the technology and the theory in a very preliminary stage^[6-11]Brain signal collection technology, computer science and signal analysis, and neuroscience continue to evolve, creating conditions for the development of brain-computer interface technology.^[12].

In 1973, the term "brain-computer interface" was formally introduced. At this stage, brain-computer interface technology began to gradually move from conceptualization to practical demonstration. For example, in 1977, Jacques J. Vidal developed a brain-computer interface system based on visual event-related potentials, in which the selection of control commands is realized by looking at different locations.^[13-14]; In 1980, German scholars proposed a brain-computer interface system based on cortical slow potentials^[15]. Although these early systems did not make significant progress, they provided an important theoretical foundation and technical reference for subsequent research^[16-17].

Since 1988, brain-computer interface technology has entered a phase of rapid development. A number of key academic researches and technological innovations have led to the gradual movement of brain-computer interfaces from the laboratory to practical applications.^[18].In 1988, researchers proposed a BCI system based on the P300 component, and they developed a speller system that allows users to select characters by observing the flashing letter^[19]In 1997, Dennis McFarland and others proposed a sensory-motor rhythm (SMR)-based BCI system, in which they showed how to control a computer cursor by imagining the movement of a hand or foot^[20]In 1999, a team from Tsinghua University developed a BCI system based on steady-state visual evoked potentials (SSVEPs) and used it to control cursor movement^[21-22]...In the early 21st century, researchers realized a BCI system based on pseudo-random sequence modulation of visual evoked potentials (VEPs), developed a speller system^[23].

Brain-computer interface (BCI) technology has made remarkable progress since the beginning of the 21st century. In terms of algorithmic research, advanced EEG signal processing and machine learning algorithms have been applied to early BCI paradigms, significantly improving the performance of the system, especially reducing the training time required by subjects; the development of publicly available datasets has opened up opportunities for the development of novel BCI target recognition algorithms. On the hardware side, the increasing maturity of miniaturized multichannel amplifier technology and the emergence of new generation sensors such as dry electrodes and hydrogel electrodes have further enhanced the utility of BCI systems with these recording hardware. In terms of interaction paradigms, many new paradigms such as hybrid brain-computer interfaces, emotional brain-computer interfaces, cognitive brain-computer interfaces, etc. are emerging and becoming more mature.^[24]. In terms of system applications, a variety of BCI application systems have emerged in recent years, and assistive/rehabilitative BCIs designed for communication and motor rehabilitation have been widely used in experimental studies and clinical trials and validated as being suitable for patients with amyotrophic lateral sclerosis, stroke, spinal cord injuries, and disorders of consciousness^[25-33].

Brain-computer interfaces are "emerging"; in 2004, Brown University developed a brain-computer interface system called BrainGate, which was implanted to allow a woman paralyzed by stroke to take her first sip of coffee with a robotic arm; in April 2021, Neurolutions' IpsiHand Upper Limb Rehabilitation System was approved by the U.S. Food and Drug Administration (FDA) as the first non-invasive brain-computer interface system for rehabilitation. Neurolutions' IpsiHand Upper Extremity Rehabilitation System receives approval from the U.S. Food and Drug Administration (FDA) as the first non-invasive brain-computer interface robot for rehabilitation. 2023. A Stanford University research team develops an intracortical brain-computer interface (iBCI) system that converts brain signals from ALS patients into text in real time. 2024, a team from Tsinghua University and Xuanwu Hospital of Capital University of Medical Sciences successfully conducts the first wireless minimally invasive brain-computer interface clinical trial, allowing paraplegics to have "brain-controlled drinking water. On January 29, 2024, a team from Tsinghua University and Xuanwu Hospital of Capital Medical University successfully conducted the first clinical trial of wireless minimally invasive brain-computer interface, and a paraplegic patient realized "brain-controlled water drinking". The system realizes unmanned control of robots by cultivating the "brain" for tasks such as obstacle avoidance, tracking, and grasping.2024 On August 22, Neuralink, a brain-computer interface company, said that its second human brain-computer implantation patient was able to design 3D objects. In addition, there has been a steady growth in non-medical applications using BCI technology for healthy populations, expanding into the military, education, entertainment, industrial and transportation safety, and many other areas.^[34-39].

Overall, the development of brain-computer interface technology has gone through the process from science fiction to scientific proof to technology explosion. Currently, the technology is in a stage of rapid development and shows broad application prospects in many fields such as healthcare, industry, education, entertainment and so on.^[12]In the future, brain-computer interface is expected to realize more breakthroughs with the further improvement of technology and application. In the future, with the further improvement of the technology and further application, brain-computer interface is expected to realize more breakthroughs and bring more possibilities for human beings.

In recent years, with the development of brain-computer interface technology, the research activity in this field has increased significantly. The emergence of more and more academic papers, patent applications and funding for research projects fully reflects this trend. Whether it is basic research or application development, brain-computer interfaces are receiving extensive attention from research institutions and enterprises around the world, and governments and industries have also increased their investment to promote rapid iteration and innovation of the technology. In this section, we will further discuss the research progress, market scale and its development trend in the field of brain-computer interface in recent years.

From the data of related thesis publication and patent application of brain-computer interface included in the literature library of the National Science and Technology Library and Documentation Center, the field has entered into a stage of rapid growth after 2000, with the number of related theses growing from 222 in 2000-2002 to 6958 in 2021-2023 (see Fig. 1), and the number of related number of patent applications grows from 1,387 in 2000-2002 to 10,121 in 2021-2023 (see Figure 2).

Fig. 1 Growth in the number of published papers related to brain-computer interfaces over the years

Fig. 2 Growth in the number of brain-computer interface-related patent applications over the years

Although brain-computer interface research began half a century ago, private investment did not begin to emerge until the early 21st century. From 2005 through 2014, funding in the field was limited, with an average of only about $24 million raised annually through three, deals. The sector began to accelerate in 2016 and has experienced multiple large-scale funding rounds in excess of $100 million in the years since (see Figure 3). 2021 was the peak year, with a total of $750 million raised across all sectors. the largest funding round in 2021 came from Elon Reeve Musk-founded neuroprosthesis and neuroimplant company Neuralink, which raised $205 million in a Series C round. Other startups that received significant investments in the same year included Petal, a user interface software company, and MindMaze and Ceribell, two non-invasive wearables companies. After a slight dip in 2022, the industry is once again approaching its 2021 funding record in 2023, largely due to Neuralink's $323 million in Series D funding. Despite the small number of companies, the neuroprosthesis and neuroimplants sector saw the highest amount of investment between 2015-2023, at $1.3 billion, more than half of which was contributed by Neuralink.

Figure 3 Changes in Global Investment in Brain-Computer Interfaces

Organizations differ in their predictions for the future market size of brain-computer interfaces. According to Precedence Statistics, the global brain-computer interface market size will be $2.35 billion in 2023 and is expected to reach $10.8 billion in 2033 (see Figure 4). Other research firms such as the China National Institute of Standardization of Electronic Technology predict that the global brain-computer interface market will reach $3.7 billion by 2027, while a report by IMARC Groupe expects it to reach $3.3 billion by 2027. In addition, some more optimistic forecasts suggest that the global brain-computer interface market size will reach $5.340 billion by 2030.

Figure 4 Global Brain-Computer Interface Market Size and Forecast Situation

From the perspective of national strategies, all countries have recognized the strategic importance of brain science research and promoted it as a key project at the national level. Whether it is the U.S. Brain Program 2.0, the European Union's Human Brain Program (HBP), China's Science and Technology Innovation 2030-"Brain Science and Brain-Like Research", or South Korea's Digital Bio-Innovation Strategy and Japan's Long-Term Research Program, countries are trying to deepen the understanding of the work of the brain and ultimately benefit human health through massive financial investment and technological research and development. All countries are trying to deepen the understanding of the brain's work and ultimately benefit human health through massive investment and technological research and development.

The U.S. Brain Program places special emphasis on the military and medical applications of brain-computer interface technology, reflecting the U.S. emphasis on the potential value of technological innovation in the areas of national security and health care; the European Union's Human Brain Program (HBP) is dedicated to understanding and exploring the functioning mechanisms of the human brain through simulation, and although the HBP program stopped funding at the end of 2023, it has shifted to focus on the next phase of the mission, namely, the use of personalized brain models to advance drug discovery and improve the treatment of brain diseases; China's Brain Science and Brain-Like Research Program, although a late starter, has shown strong determination in funding, with the first batch of funding reaching RMB 3 billion and subsequent funding expected to reach tens of billions of RMB, showing China's confidence in its desire to make major breakthroughs in the field; South Korea is focusing on brain-computer interaction technology, brain function visualization and digital therapies, showing Korea's innovative thinking and practical direction in this field; Japan's projects are not as large as those of the other countries, but it focuses on establishing animal models to study the occurrence and development of brain diseases, with a view to accelerating the understanding of human brain diseases (see Table 1).

Table 1 Main countries and territories

Brain-computer interface funding

With the rapid development of brain-computer interface technology in academia and industry, the related research results and market potential have been continuously explored. The growth in the number of papers, the emergence of patents, and the continuous investment of research funds have not only promoted the technological breakthroughs in this field, but also contributed to the expansion of the market scale year by year. Against this background, the current mainstream technology of brain-computer interfaces is gradually taking shape and has made significant progress in multiple directions. Next, the basic components and classification of brain-computer interfaces will be highlighted.

The BCI system is not just a mere collection of technologies, it involves a comprehensive intersection of multiple types of technologies and disciplines, and is a complex science and engineering system.The BCI system integrates knowledge from a variety of fields such as neuroscience, signal processing, computer science, artificial intelligence, and electrical engineering.The BCI system mainly consists of the user (brain), brain signal acquisition, brain signal processing and decoding, control interfaces, robots, and other peripherals, and neurofeedback (see Figure 5). Among them, the most central ones are brain signal acquisition, brain signal processing and decoding, and control interface, which are the three basic components of the BCI system.

Figure 5 Brain-computer interface system components

(1) Brain Signal Acquisition

Brain signal acquisition is the capture of electrophysiologic signals from specific activities of the brain (e.g., motor, speech, auditory, and visual) by electronic devices. The electrophysiological signals processed by most BCI systems can be categorized as follows: electroencephalography (EEG), the electrical activity of the brain measured by electrodes placed on the scalp; electrocorticography (ECoG), the EEG signals measured directly by electrodes placed on the surgically exposed cerebral cortex; local field potentials (LFPs), the electrical potentials measured in the extracellular space of a neuron; neuronal action potentials ( AP or Spike), a rapid temporary change in neuronal membrane potential. The captured brain signals are filtered, amplified, and digitized before being transmitted to the next BCI component.The overall performance of the BCI system is highly dependent on the quality of the acquired brain signals (signal-to-noise ratio)^[40-46].

(2) Brain Signal Processing and Decoding

Feature extraction: a BCI system extracts key electrophysiological features from the acquired signals to define brain activity and thus encode the user's intention; electrophysiological features can be categorized into frequency-domain features and time-domain features. For example, in a BCI system based on motor imagery, the focus is usually on power changes in the β and μ frequency bands (approximately 8-30 Hz). In contrast, in BCI systems based on P300 event-related potentials, the focus is more on detecting peak responses at specific time points. Conventional feature extraction methods include the commonly used power spectral density (PSD), autoregressive modeling (AR), covariance matrix analysis^[47]. The advantages of traditional methods are good real-time performance, low computational complexity, and suitability for low-power and embedded application scenarios. However, the disadvantages are the inability to effectively deal with the complexity and high-dimensional features of brain signals, and the limited ability to deal with noise and artifacts^[48]. Before designing a BCI system, the desired feature classes should be identified.

In recent years, feature extraction methods incorporating AI techniques have gradually emerged, especially the application of deep learning algorithms has significantly improved the performance of brain signal processing. Deep neural networks (DNNs) are able to automatically extract multi-level features from raw EEG signals, avoiding the limitations of manual feature engineering^[49].. Convolutional neural networks (CNNs) are good at extracting spatio-temporal features, while recurrent neural networks (RNNs) such as long short-term memory networks (LSTMs) are able to capture time series dependencies in EEG signals^[50]. Instead of predefined features, these methods train models with large-scale data to capture complex neural activity patterns, which are far superior to traditional methods in terms of feature extraction accuracy and classification performance^[51].

Feature Classification: extracted features represent the brain activity used to anticipate an action, and feature classification is responsible for converting the extracted features into actionable commands or outputs. In this process, BCI systems utilize machine learning algorithms or other classification methods to differentiate between different patterns of brain activity and thus infer the user's intent. Traditional classification methods such as Support Vector Machine (SVM), Linear Discriminant Analysis (LDA), and Plain Bayes are widely used in BCIs^[52]. These algorithms perform well in low-dimensional feature spaces and are suitable for resource-constrained real-time applications. However, traditional classifiers usually rely on hand-extracted low-dimensional features and have low classification accuracy in the face of high-dimensional, nonlinear, noisy signals^[53].

Artificial intelligence techniques have introduced more sophisticated classification models such as Convolutional Neural Networks (CNN), Generative Adversarial Networks (GAN), etc.^[54]. These deep learning methods can capture more complex patterns in brain signals by automatically learning features. For example, convolutional neural networks excel at extracting spatio-temporal features in 2D and even 3D EEG images, while recurrent neural networks (RNNs) excel on long time series data. Further, hybrid models such as Convolutional Recurrent Networks (CRNN) combine the advantages of the former two and show excellent performance in BCI decoding^[55].. Although AI models outperform traditional methods in classification accuracy, their high computing power and large data requirements make real-time and resource consumption a key bottleneck. In addition, the training of deep learning models often requires large amounts of labeled data, which may be difficult to obtain in BCI research.

Feature conversion: in this signal processing stage, the classified features are converted and output as actual commands to operate external devices (BCI applications). An important attribute of the feature conversion algorithm is adaptability: the conversion algorithm is able to adaptively self-adjust to changes in the features to generate the commands needed by the user.

(3) Control interface

In a BCI system, commands generated after feature extraction and conversion are used to control a variety of external devices to realize different functions. For example, controlling the cursor on a computer screen to help the user select letters or text; controlling the movement of a wheelchair, or the operation of a robotic arm; the control signals generated by BCI can also be fed back to the brain or other organs of the body to promote rehabilitation or replace lost functions^[56].

Based on the structural and functional characteristics of brain-computer interface systems, BCIs can be broadly categorized into three types: output BCIs (Output BCIs), input BCIs (Input BCIs), and bidirectional BCIs (Bidirectional BCIs). Output BCIs are primarily concerned with obtaining signals from the brain and converting these signals into control commands for the external world. These types of BCIs are designed to help individuals who are unable to perform regular physical movements due to a physical disability by enabling them to control external devices, such as controlling a wheelchair or prosthetic limb, a computer cursor, or an input device, through brain signals. Input BCIs, on the other hand, are designed to receive external stimuli and translate them into perception or cognition within the brain. These types of BCIs are designed to enhance or restore the user's sensory experience and are often used to restore impaired sensory abilities, commonly visual prostheses (e.g., microchips implanted under the retina), auditory prostheses (e.g., cochlear implants), and others. Bidirectional interactive BCIs combine the functions of output BCIs and input BCIs to further enhance user experience and control precision.

BCIs can be categorized as invasive, semi-invasive, and non-invasive based on the method and technology of signal acquisition^[29-30].. The most common technique for non-invasive BCI is electroencephalography (EEG), and in recent years, functional near-infrared spectroscopy (fNIRS)^[31], Magnetoencephalography (MEG)^[32]Functional Magnetic Resonance Imaging (fMRI), Positron Emission Tomography (PET)^[33]and functional transcranial Doppler ultrasound^[34-36]It is also utilized. Semi-invasive BCIs include electrocorticography (ECoG or ECog), subdural electrodes (Subdural Electrodes), and others. In the field of invasive BCI, microelectrode arrays (MEA)^[37]and neuronal discharge assays (SUA) are the most common^[38].. Invasive BCI provides very high-quality signals but requires surgical implantation and carries some risk; noninvasive BCI is easy to use and less risky, but the signal quality is usually not as good as invasive methods.

BCIs can be categorized into Active BCIs, Reactive BCIs, and Passive BCIs based on the way they use brain signals. Active BCIs require the user to be trained to produce explicit control signals and are commonly used to control external devices or interact with computers, such as maneuvering a wheelchair or performing spelling tasks. Reactive BCI utilizes the user's natural response to specific stimuli (e.g., sound, image, touch, etc.). When the user responds to these stimuli, the BCI system is able to detect changes in brain activity and performs the appropriate actions accordingly, e.g., by detecting the P300 wave to select letters or shapes on the screen for spelling tasks. Passive BCI provides information about the user's state (e.g., fatigue, attention level, etc.) by analyzing the user's natural brain activity without the user's active participation, and is often used to monitor the user's cognitive state to optimize work or learning efficiency, for example, to monitor the fatigue level of employees in the workplace to improve safety.

In addition, there are other categorization methods, such as according to the purpose of the BCI system, it can be divided into assistive technology BCI, enhanced BCI, entertainment BCI, etc.^[57].

Although brain-computer interface (BCI) technology shows great potential for development, it still faces many bottlenecks and challenges in the process of its wide application. In the following section, we will analyze these key issues that hinder the wide application and further breakthrough of BMI technology from four aspects: psychophysiological bottlenecks, technological challenges, industrial challenges, and regulatory challenges.

From a psychophysiological perspective, BCI performance is influenced by a variety of psychophysiological factors that lead to significant individual differences. Psychological factors such as attention, memory load, fatigue, and competing cognitive processes affect brain transient dynamics^[5]. For example, less empathetic individuals are less emotionally engaged when using the P300-BCI paradigm, yet are able to produce higher P300 amplitudes than more empathetic individuals. In addition, the user's personal characteristics, such as lifestyle, gender, and age, can also affect BCI performance^[44-46].. Physiological parameters, such as frequency-domain characteristics of resting heart rate variability (HRV), were similarly correlated with BCI performance^[58]. Human resting-state networks (RSNs) are dynamically changing and are able to instantly adjust the cortical features that need to be read by any BCI system^[59]The term "BCI-illiterate" is used to describe individuals who are unable to produce strong enough brain signals to operate the BCI system. Individuals of approximately 15%-30% are referred to as "BCI illiterate" due to their inability to generate brain signals strong enough to operate the BCI system.^[60]. Therefore, an efficient BCI system must be robust to such intrinsic physiological fluctuations and also requires a deeper understanding of how the nervous system works as well as the development of new materials and techniques to improve the interface between the device and the brain tissue.

The main technical challenges facing brain-computer interface technologies include the accuracy of signal perception and the complexity of data processing. Both invasive and non-invasive technologies need to address the accuracy of signal perception, especially in terms of increasing the speed of signal transmission. Long-term stability and biocompatibility are also important issues, as invasive devices may trigger an immune response or lose signal quality over time.^[61]. In addition, the complexity of data processing is a key issue, especially when dealing with a large number of EEG signals. The understanding of the human brain is still relatively limited, and the research on brain mechanisms such as neuronal mechanisms needs to be deepened. In terms of hardware, domestically produced EEG acquisition equipment has a low share in the market of scientific research instruments, and most of the research uses still high-precision, multi-lead EEG acquisition equipment that is large in size, heavy in weight, and poor in portability. Therefore, there is an urgent need to develop portable, high-performance EEG acquisition and recording hardware to support technology translation.

The industrialization of brain-computer interface technology faces the problem of balancing benefits and risks, costs and returns. As the brain-computer interface technology market is still at an early stage, the industrial scale is unclear, product compliance is debatable, and there is a lack of relevant legal basis, making it difficult to achieve complete industrialization. Therefore, the focus of industrialization should be placed on solving the immediate needs of society, such as disease treatment and rehabilitation, as a basis for gradually promoting the industry to the ground, and subsequently attracting market capital to follow up to ensure the healthy and sustainable development of the industry. At the same time, it is also necessary to strengthen the popularization of science to the public, to avoid unnecessary speculation and panic caused by exaggerated propaganda.

From a security and ethical point of view, brain-computer interface technology faces the risk of privacy leakage such as hacking and data theft, and especially invasive devices also have the problem of possible trauma and infection of brain tissue during implantation in the human body. Therefore, the issues of equipment safety, personal privacy and security, informed consent, autonomy and responsibility, as well as social equity and justice that may be triggered by the use of brain-computer interface devices need to be emphasized. At present, there is no unified basic theoretical framework for brain-computer interfaces, and there is a lack of standards for scientific evaluation of the performance of brain-computer interface systems. Therefore, from a regulatory point of view, it is necessary to formulate corresponding policies and regulations to meet the above challenges in order to standardize the development of brain-computer interface technology and industry.

In the face of the bottlenecks and challenges in the current development of brain-computer interface technology, solving these problems will become the key to promote further breakthroughs in the technology. Looking into the future on this basis, the progress of brain-computer interface technology will be characterized by stages. Next, the future development direction of brain-computer interface and possible technological breakthroughs will be explored in three phases: short-term, medium-term and long-term.

In the short term, the application of brain-computer interface technology will mainly focus on the medical and rehabilitation fields, especially in restoring patients' mobility and realizing the output BCI in the medical field. Through neurorehabilitation applications, BCI can help stroke or spinal cord injury patients rebuild their connection with the outside world and improve their quality of life and independence^[62]. In addition, BCI technology will show innovative potential in precision treatment and monitoring, early diagnosis of diseases, and facilitating the rehabilitation process. As BCI technology spreads, there will be an exponential growth in the data collected by BCI devices, which will further drive the iterative updating of the technology.

As technology advances, brain-computer interface technology will become more accurate, portable, reliable, and easy to use. Hardware-wise, the accuracy of brain-computer interfaces in reading and writing brain signals will continue to improve, enabling more accurate information transmission and finer control of brain functions. With the development of micro-nano processing technology and material science, the electrodes and chips for collecting EEG signals will develop in the direction of flexibility, miniaturization, high throughput, and integration. For example, in September 2024, the Swiss Federal Institute of Technology (SFIT) developed a miniature brain-computer interface with a size of only 2.46 mm2 (MiBMI )^[63]Neuralink's communication speed is currently 10 bits per second, and its future goal is to reach the Megabit level. Software and algorithms, brain-computer interface technology will also incorporate more and more advanced artificial intelligence technologies to achieve better signal processing, feature extraction and pattern recognition, thus improving the performance and intelligence of the system. The researchers also released Image Decoder, a new deep learning application based on Meta's open-source model DINOv2, which can decode a subject's brain activity in real time^[64]; In August 2024, Synchron, a brain-computer interface company, combined its own brain-computer interface device with the ChatGPT and Vision Pro to give users a new experience in using the device. In the coming years, more breakthroughs are expected in the development of safer and invasive electrode systems, improved efficiency in neural coding and decoding, and so on^[65]. The brain-computer interface devices will also utilize advanced energy technologies and efficient algorithms to achieve greater energy efficiency and endurance to ensure long-term operation and large-scale deployment.

In the medium term, output BCI will further expand from the medical field to other fields such as education, entertainment, military, and industry, while input BCI as well as bi-directional interactive BCI will be realized in the medical field.

As output BCI technology advances and costs decrease, brain-computer interface technology will gradually become pervasive and an integral part of people's lives and work. At this stage, a variety of commercial BCIs will also continue to emerge, including autonomous driving and brain-controlled robots^[66]The following are some of the most important areas of interest to the industry: web browsing and virtual world navigation, education and learning, gaming and entertainment, alertness monitoring, memory and cognitive enhancement, polygraphy, security and identification, brain-controlled art, and more.^[67-71]. In terms of remote control technology, BCI will make the operation of equipment more convenient and realize remote control of robots, thus enhancing work efficiency. Against the backdrop of other biometric technologies being gradually cracked by AI, BCI can provide a new type of biometric identification means.^[66]BCI technology can be used to monitor and evaluate students' engagement, concentration, and cognitive load levels when listening to lectures or completing exercises in real time. At the same time, BCI will also revolutionize the gaming experience by realizing a new mode of social interaction through virtual reality, bringing users a brand new entertainment experience. In addition, in terms of input-type BCI and two-way interactive BCI, Neuralink and other BCI companies have indicated that they will develop BCIs that act on human visual nerves, enabling blind people to see again.

In the long term, brain-computer interfaces will shift from the current focus on replacing and restoring patients' lost functions to improving, supplementing and enhancing functions. From the perspective of information flow, the current EEG signal is mainly the output BCI of "from brain to machine", and in the future, it will experience the input BCI of "from machine to brain" and "from brain to brain" as well as the two-way interactive BCI. In the future, we will experience "from machine to brain", "from brain to brain" input BCI and two-way interactive BCI, in which EEG signals flow and interact in both directions between the brain and the machine, and ultimately realize brain-machine fusion intelligence.^[72]People may be able to experience some perceptions that cannot be experienced only by human natural senses through external devices of brain-computer interface. People may be able to experience some perceptions through external devices of brain-computer interfaces that cannot be experienced only by the natural human senses, such as perceiving ultraviolet and infrared light, ultrasound and infrasound, etc., and the threshold of human perceptions will be expanded tremendously!^[73]. In addition, BCI technology can also be used to achieve memory and cognitive amplification, as Theodore W Berger and Sam A Deadwyler et al. have demonstrated experimentally as early as 2011 that brain implanted devices placed in rats can restore lost memories and enhance the memory of new knowledge^[74]In the future, the brain and the machine will adapt to each other and work together. In the future, the brain and the machine will both adapt to each other, work together, and ultimately achieve mutual cooperation at multiple levels of information perception, information processing, decision-making judgment, and even memory and intent.

Of course, as these technologies advance, issues of ethics, privacy and security will become increasingly prominent^[75-76]. How to ensure the security of personal data and prevent abuse, as well as how to define the attribution of responsibility between humans and machines, will become problems that need to be faced and solved by the society together. Therefore, the establishment of a corresponding legal framework and technical standards to ensure the healthy development of the technology and the general acceptance of the society will be the key to promote the wide application of BCI technology.

Since the concept of BCI was first proposed in the 1920s, this field has undergone a transformation from science fiction to real-world application. From the initial theoretical conception to the technological explosion in the 21st century, BCI technology has not only gained initial validation during the scientific demonstration period and accumulated a large number of research results during the academic development period, but also realized many breakthroughs during the technological explosion period. These advances cover the improvement of brain signal acquisition technology, the optimization of signal processing algorithms, and the precision of external device control. Currently, mainstream BCI technologies can be categorized into non-invasive, semi-invasive, and invasive, each with its own unique application scenarios. For example, non-invasive EEG techniques are widely used in assistive technology and enhanced BCI due to their ease of use and low risk, while invasive microelectrode arrays play an important role in scenarios requiring high accuracy due to their ability to provide high-quality signals. In addition, the categorization of BCI technologies includes output, input, and bidirectional interaction, as well as passive, active, and reactive BCI, which reflect the flexibility of the technologies in different application areas. Although BCI technology has made significant progress, it still faces some major challenges. These challenges mainly include psychophysiological bottlenecks, technological challenges, industrial challenges, and regulatory challenges. The psychophysiological bottleneck involves inter-individual differences and the dynamics of brain activity; the technical challenge involves the accuracy of signal perception and the complexity of data processing; the industrial challenge includes market acceptance, compliance, and financial support; and the regulatory challenge focuses on the safety of the device, privacy protection, and ethical issues. The future development of brain-computer interface technology will be divided into three phases: short-term, medium-term, and long-term. In the short term, the technology focus will be on medical rehabilitation, especially helping stroke and spinal cord injury patients regain mobility and improving their quality of life through neurorehabilitation applications. Technological advances will make devices more portable, accurate and easy to use, while data growth will drive technology iteration. In the medium term, output BCI technology will expand into education, entertainment, military and industry, while input and bi-directional interactive BCI will be further utilized in healthcare. As the cost of the technology decreases, BCI devices will gradually become more popular and play an important role in daily life and work, including autonomous driving, virtual reality gaming, and educational assistance. In the long run, BCI will shift to function improvement and enhancement, realizing two-way interaction from "brain-to-machine" to "machine-to-brain" and even "brain-to-brain", and ultimately achieving brain-computer fusion intelligence. In addition, BCI is expected to expand human perceptual ability and enhance memory and cognitive functions. However, with the advancement of technology comes concerns about ethics, privacy and security, and the establishment of a reasonable legal framework and technical standards is essential to ensure the healthy development of the technology.

With the continuous progress and improvement of the technology, we have reason to believe that BCI technology will bring far-reaching impact on human life and society in the future, but a series of challenges still need to be overcome to fully realize its potential. Continuing multidisciplinary research, strengthening international cooperation, and improving laws and regulations will be indispensable key steps in promoting the healthy development of BCI technology.