According to the introduction, "Dharma Institute 2022 Top 10 Technology Trends" adopts the analysis method of "quantitative dispersion, qualitative convergence", and the whole analysis process is divided into two parts:
Dharma Institute analyzed 7.7 million public papers and 85,000 patents in 159 fields in the past three years, and excavated the hot areas and key technological breakthroughs, interviewed nearly 100 scientists in depth, and proposed the top 10 technology trends that may come into reality in 2022, covering AI for Science, which is a new paradigm for scientific research.
The Dharma Institute analyzed 7.7 million open papers in 159 fields over the past three years and 85,000 patents, explored hot areas and key technological breakthroughs, conducted in-depth interviews with nearly 100 scientists, and put forward the top 10 technological trends that will likely come to life in 2022, covering areas such as AI, microchips, computing, and communication.
Specifically, these ten technology trends are: AI for Science, co-evolution of large and small models, silicon optical chips, green energy AI, flexible sensory robots, high-precision medical navigation, global privacy computing, star-earth computing, cloud-network-terminal convergence, and XR Internet.
The Dharma Institute believes that the path for computer science to change scientific research is from downstream to upstream. At first, computers were mainly used to analyze and summarize experimental data. Later, scientific computing changed the way of scientific experiments, artificial intelligence combined with high-performance computing, in the experimental cost and difficulty of the field began to use computers to simulate experiments, to verify the assumptions of scientists, to accelerate the output of scientific research results, such as nuclear energy experiments, digital reactor, to reduce the cost of experiments, improve safety, reduce the generation of nuclear waste.
In recent years, AI has been shown to do scientific law discovery, not only in the field of applied science, but also in the field of basic science, such as DeepMind to use AI to help prove or propose new mathematical theorems, assisting mathematicians to form an intuition of complex mathematics.
The Dharma Institute predicts that within the next three years, AI technology will become commonplace in applied sciences, and start to become a research tool in some basic sciences.
Hua Xiansheng, head of the City Brain Lab at the Ali Dharma Institute, said in an interview with InfoQ that using AI to fuel scientific research is mainly based on the two points of data and computation, and that AI capabilities are formed on the basis of data and computing power.
"Essentially, there is not much difference between AI for Science and AI for Industry, AI is also used as a tool to promote the development of the field. It's just that the field is a little bit different in that it has a higher threshold, because it's something that scientists have to do, not something that an ordinary person, a general technical staff member, can do. But in essence, it's also a field where because of the data, algorithms can be designed to tap into the 'mystery' in the data to solve problems in the field."
For practitioners, AI for Science requires AI experts to understand the scientific problems, and scientists to understand the principles of AI. "AI for Industry is actually a single-point technology that is gradually moving towards platformization, and the future of AI for Science, I think, will also gradually move towards platformization. The future of AI for Science, I think, will also gradually move towards platformization. This time is when AI experts combine a certain field, a certain discipline, or even a certain type of problem in a certain discipline with scientists to build a scientific research platform. At this time, scientists may have more freedom, more powerful tools, and can do scientific research in larger quantities to realize richer and more important scientific breakthroughs." Hua Xiansheng said.
Google's BERT, Open AI's GPT-3, Wisdom Source's Wudao, Dharma Institute's M6, and other large-scale pre-training models have made significant progress, and the performance of large models has improved by leaps and bounds, providing a foundation for the development of downstream AI models. However, large model training consumes too much resources, and the performance improvement brought about by the increase in the number of parameters is disproportionate to the increase in consumption, making the efficiency of large models challenged.
Yang Hongxia, a scientist at the Intelligent Computing Laboratory of the Ali Dharma Institute, said in an interview with InfoQ that there are still several topics that need to be broken through in pre-training large models:
The Dharma Institute believes that the development of the parameter scale of the large model will enter a cooling-off period, and that synergy between the large model and the associated small model will be the direction of development in the future. The knowledge and cognitive reasoning ability precipitated by the big model is output to the small model, and the small model superimposes the perception, cognition, decision-making, and execution ability of the vertical scene based on the foundation of the big model, and then feeds back the results of execution and learning to the big model, so that the knowledge and ability of the big model can continue to evolve and form a set of organic cycle of the intelligent system, and the more participants there are, the more beneficiaries there are, and the faster the model can evolve.
"The co-evolution of large and small models can also better serve more complex new scenarios, such as virtual reality and digital people, which require simultaneous deployment and interaction on the cloud-side end, and the system is also more flexible for protecting user data privacy, as users can maintain their own small models on different ends." Yang Hongxia told InfoQ.
Tang Jie, a professor at Tsinghua University's Department of Computer Science and academic vice president of the Beijing Zhiyuan Artificial Intelligence Research Institute, said that the development of large models, in terms of cognitive intelligence, the model parameters do not rule out the possibility of further increase, but the parameter race is not an end in itself, but to explore the possibility of further performance improvement. Big model research focuses on architectural original innovation at the same time, and further improves the cognitive intelligence capability of trillion-scale models through methods such as continuous learning of models, increased memory mechanisms, and breakthroughs in ternary knowledge representation methods. In terms of the model itself, multimodal, multilingual, programming-oriented new models will also become the focus of research.
The Dharma Institute predicts that in the next three years, pilot explorations of co-evolving intelligent systems based on large-scale pre-trained models will be conducted in individual domains. In the next five years, co-evolutionary intelligence will become the system standard, allowing all of society to easily access and contribute to the capabilities of intelligent systems, and taking another step towards generalized AI.
The development of electronic chips is approaching the limits of Moore's Law, integration technology is saturated with advances, and high-performance computing requires ever-increasing data throughput, which requires technological breakthroughs.
Photonic chips are different from electronic chips in that they are technically a different way to use photons instead of electrons for information transfer, which can carry more information and transmit over longer distances. Photons interfere less with each other and provide two orders of magnitude higher computational density and two orders of magnitude lower energy consumption than electronic chips. Compared to quantum chips, photonic chips do not require a change in binary architecture and can continue the current computer system. The photonic chip needs to be integrated with the mature electronic chip technology, using the advanced manufacturing process and modular technology of electronic chips, and combining the advantages of photonic and electronic silicon optical technology will be the mainstream form in the future.
Zhou Zhiping, professor at Peking University and chief researcher at Shanghai Institute of Optical Mechanics, said that Dharma Institute's choice of "silicon optical chip" as one of the 10 major technology trends in 2022 confirms the great value of this technology in the field of information and communication. The further expansion of silicon optical chip is silicon-based optoelectronic chip, which utilizes the design method and manufacturing process of integrated circuits to heterogeneously integrate micro- and nano-scale photon, electron, and optoelectronic devices on the same silicon substrate, forming a new large-scale optoelectronic integrated chip with comprehensive functions. It more significantly reflects the continuous efforts of human society in nanotechnology and the great interest in smaller devices and more compact systems.
Dharma Academy predicts that optoelectronic fusion is the future development trend of the chip, and that silicon photonics and silicon electronic chips complement each other to give full play to the advantages of both, leading to a continuous improvement in computing power. In the next three years, silicon optical chips will support high-speed information transmission in large data centers; in the next five to ten years, optical computing based on silicon optical chips will gradually replace some of the computing scenarios of electronic chips.
The large-scale development and utilization of green energy has become the main direction of energy development in today's world. Under the trend of high proportion of green energy grid-connected, the traditional power system is difficult to cope with the uncertainty of the power generated by green energy in the weather such as wind, rainstorms, lightning, etc., as well as the ability to cope with the timely response to complex faults.
In the process of operation monitoring, parameter verification and fault monitoring still require a large number of manual participation, and fault characteristics are difficult to extract and difficult to recognize. In view of the various challenges faced by large-scale green energy grid integration in terms of stability, operation and planning, a new generation of information technology based on artificial intelligence will provide technical guarantee and strong support for the efficient and stable operation of the energy system as a whole.
The deep integration of artificial intelligence and energy power will promote large-scale new energy generation, grid connection, transmission, consumption and safe operation, and complete the upgrading of the energy system.
The chief system architect of the Chinese Academy of Electricity Sciences, Mr. Zhou Jiuzi, believes that the new power system to achieve intelligent regulation and control, operation and deduction will be inseparable from the AI technology, under the auspices of the AI technology to build a number of physical power grid and IT applications interacting with the digital twins, and each digital twin to solve a certain scenario or a certain aspect of the power grid operation problems. In this way, when there are enough twins to form a grid control digital twin system to solve all aspects of grid operation problems, intelligent control can be realized.
Dharma Academy predicts that in the next three years, artificial intelligence technology will help the power system to achieve large-scale green energy consumption, energy supply in the time and space dimension can be interconnected and mutually beneficial, the network source of coordinated development, elasticity of scheduling, and the realization of the power system's security, high efficiency, and stable operation.
Robotics is a collection of technology, in the past hardware, network, artificial intelligence, cloud computing under the convergence of the development of technology maturity has made a leap forward, the robot towards multi-tasking, self-adaptive, collaborative route to development.
Flexible robot is an important breakthrough representative, with soft and flexible, programmable, scalable and other characteristics, combined with flexible electronics, force sensing and control technology, can adapt to a variety of working environments, and adjusted in different tasks. In recent years, flexible robots combined with artificial intelligence technology have enabled robots with sensing capabilities, improved versatility and autonomy, and reduced reliance on pre-programming.
Flexible perceptual robots increase the ability to perceive the environment (including force, vision, sound, etc.), have increased ability to migrate to tasks, no longer needing to exhaust possibilities like traditional robots, and can perform tasks that rely on perception (e.g., medical surgeries), expanding the applicability of the robot's scenarios. Another advantage is the ability to adapt to the task, responding to sudden changes, accurately completing the task and avoiding problems.
Dharma Academy predicts that in the next five years, flexible robots will be fully integrated with the intelligent perception capabilities brought by deep learning, and will be able to face a wide range of scenarios, gradually replacing the traditional industrial robots, and becoming the main equipment on the production line. At the same time in the field of service robots to achieve commercialization, in the scene, experience, cost has the advantage, began to scale the application.
Traditional medical care relies on the experience of doctors, as if the artificial pathfinding, the effect is uneven. The deep integration of artificial intelligence and precision medicine, the organic combination of expert experience and new auxiliary diagnostic technology will become a high-precision navigation system for clinical medicine, providing automatic guidance for doctors, helping to make medical decisions faster and more accurate, and realizing the quantifiable, calculable, predictable, and preventable major diseases.
It is expected that in the next three years, human-centered precision medicine will become the main direction, and AI will fully penetrate all aspects of disease prevention and diagnosis, and become a high-precision navigation synergy for disease prevention and diagnosis. And with the further development of causal reasoning, interpretability is expected to achieve breakthroughs, artificial intelligence will provide strong technical support for disease prevention and early diagnosis and treatment.
Data security protection and data circulation is a dilemma in the digital age, and the way to crack it is privacy computing. In the past, constrained by performance bottlenecks, lack of technical trust, and non-uniform standards, privacy computing can only be applied to a small number of data scenarios. With the development of special chips, encryption algorithms, white-boxing, data trust and other technology integration, privacy computing is expected to cross to the massive data protection, data sources will be extended to the whole domain, stimulating the new productivity of the digital era.
Ren Kui, a professor at Zhejiang University and dean of the Zhejiang University Cyberspace Security Institute, said that privacy computing is not a single technology, but a grand unification of the name, including the earliest security multi-party computing proposed in 1982, to the later homomorphic encryption, trusted computing, differential privacy, and so on. But privacy computing earlier did not have much practical value, like full homomorphic encryption theory is very good, but the performance overhead is too large, the actual use is very difficult. Now with hardware acceleration and software innovation, we are gradually seeing a trend towards practicality, but of course there is still a process.
The Dharma Institute predicts that over the next three years, full-domain privacy computing technology will have new breakthroughs in performance and interpretability, or there will be data trusts to provide privacy computing-based data **** enjoyment services.
Digital services based on terrestrial networks and computing are limited to densely populated areas, with no-man's land such as deep space, oceans, and deserts remaining a service gap. High and low orbit satellite communications and ground mobile communications will be seamlessly connected to form an integrated three-dimensional network of air, space and sea. As the calculation moves with the network, the star-earth computing will integrate the satellite system, air network, ground communications and cloud computing, becoming an emerging computing architecture, expanding the space of digital services.
Zhang Ming, head of the XG Lab at Ali Dharma Institute, believes that star-earth computing still involves a lot of breakthroughs in core technologies to achieve successful commercialization and large-scale development.
Taking the low-orbit satellite terminal as an example, one should be oriented to the scene demand and commercial value, and the other is the need to design high-performance, low-cost, adaptable to the scene of many commercial products from the perspective of technological breakthroughs and solving engineering problems. For example, in terms of key technologies, how to design a new type of millimeter-wave phased array antenna, as well as the corresponding beam fouling control algorithm, to meet the performance index requirements in a low-cost way; how to design a new type of star-ground communication protocol, to meet the requirements of satellite Internet multiuser, mobility, and complex dynamic business; in addition, in the integration and optimization of the terminal, there are still a lot of engineering problems that need to be broken through and solved, so as to meet the multi-directional needs of the sea, land and air in different scenarios. In addition, there are many engineering problems that need to be solved in terms of terminal integration and optimization, so as to satisfy the multi-dimensional needs of different scenes.
Dharma Academy predicts that in the next three years, the number of low-orbit satellites will usher in explosive growth, and high-orbit satellites **** with the composition of the satellite Internet. In the next five years, the satellite Internet and terrestrial networks will seamlessly combine to form a ubiquitous Internet that integrates heaven and earth, and the satellite and its terrestrial system will become a new type of computing node that will play a role in all kinds of digital scenarios.
The development of new network technologies will drive cloud computing towards a new computing system that integrates the cloud network end, and realizes the professional division of labor between the cloud and the network end: the cloud will act as a brain, responsible for centralized computing and global data processing; the network will act as a connection, integrating multiple network forms through the cloud to form a low-latency, wide-coverage network; and the end will act as an interactive interface, presenting a multifaceted form that provides thin, lightweight, long-lasting, and immersive data storage. The end as the interactive interface presents multiple forms, which can provide thin, light, long-lasting and immersive experience. The cloud-network-terminal convergence will promote the birth of new applications such as high-precision industrial simulation, real-time industrial quality inspection, and virtual reality integration space.
Dharma Institute predicts that in the next two years, there will be a large number of application scenarios running on the cloud-network-terminal convergence system, accompanied by more new devices based on the cloud, which will bring a more extreme and richer user experience.
With the development of end-to-end cloud collaborative computing, network communications, digital twins and other technologies, the XR (future virtual-reality fusion) Internet centered on immersive experiences will meet an explosive period. The glasses are expected to become the new human-computer interaction interface, promoting the formation of XR Internet, which is different from the flat Internet, and giving rise to a new industrial ecosystem from components, devices, operating systems to applications. XR Internet will reshape the shape of the digital applications, changing the interaction mode of entertainment, socialization, work, shopping, education, medical and other scenes.
Dharma Institute predicts that the next three years will produce a new generation of XR glasses, a blend of AR and VR technology, the use of end-to-end cloud computing, optics, perspective and other technologies will make the shape and weight close to ordinary glasses, XR glasses to become a key entry point to the Internet, a wide range of popularity.