Issue 708|2022/01/02
Opening remarks
This is an era of trials and changes,
This is an era of a closely linked community of common destiny.
In this cold winter when the epidemic continues to rage and warm up;
On the eve of this extreme weather and climate change;
In this wave of virtual reality, surging moments;
We need the power of science and technology, the power of reason,
We are convinced that this is the only thing,
Only then can we break through the prison of reality.
To explore the wider world,
To pursue a fuller self.
The bow and arrow of this era has been opened,
The most powerful technological arrow will fly out!
In 2022, we launched the Opening Year's Great,
Touch the frontiers of technology,
Let's see the future together!
Supercomputers have always been an enduring hot element in science fiction works - the different forms and highly artificial intelligence supercomputing in "Terminator", "Evangelion" and "Superbody", carrying human imagination about the quantum world and the potential of the human brain.
Not long ago, Frank Herbert's epic science fiction novel "Dune" was adapted into a film of the same name, which was screened worldwide. The film presents a detailed desert planet, Arrakis, which feels like you're there.
A group of British scientists studying climate patterns were fascinated by the strange planet and tried to simulate the climate of Erakos. However, the complicated calculation of massive data such as clouds, temperature, precipitation, and wind power can be done by non-human beings. So the scientists called on a supercomputer, which processed data day and night, built models, and finally came up with the climate environment envisioned in Herbert's book—largely in line with real climate models.
Visual demo of the Erakos climate model Image source: Climate Archive
Obviously, the above is only a "corner of the iceberg" of supercomputing forces. From paleoclimate research, earthquake early warning, geological exploration, and star trajectory deduction, supercomputing has shown magical power in many fields, continuously promoting industrial change while constantly expanding the boundaries of human exploration. So, after decades of development iteration, what kind of development form will the future supercomputer show? What are the cutting-edge research areas of brain-computer fusion? What are the shortcomings of China's supercomputing in the development process?
To this end, the Daily Economic News reporter exclusively spoke to Steve Chen, an academician of the National Academy of Engineering, an academician of the American Academy of Arts and Sciences, and a world-renowned supercomputer pioneer. At the same time, Chen Shiqing is also the founder and president of the Third Brain Research Institute in the United States, and the chief scientist of the National Engineering Laboratory of Big Data Application Technology in China's Aerospace Cloud Network Industry.
In the 1980s and 1990s, Chen Shiqing was regarded as a "superstar" in the field of world supercomputing, and Time Magazine once used it as a cover character. He studied under David Kuck, chief systems designer for NASA's early supercomputer ILIAC-IV, and joined the well-known Cray Research in 1979 as chief designer and senior vice president.
Chen Shiqing Image source: Photo by reporter Zhang Jian
As a world-renowned supercomputer expert, Chen Shiqing has achieved four "world firsts": in 1983, he developed the world's first 2-CPU parallel vector supercomputer (Cray XMP/2), which laid the foundation for the world's parallel vector supercomputer system architecture; in 1998, he developed the world's first new supercomputer system architecture and application concept - blade supercomputers .) The world's first supercomputer-based, global networking to form a similar current grid of information networks; the world's first Internet-based, application-to-application, dynamic, low-cost, high-efficiency enterprise collaborative operation intermediate software TONBU, greatly promoted the development of global supercomputing.
"The server of a blade supercomputer is like a 'blade' that can be inserted into a rackmount chassis. A 'blade' is a system board that can either run its own system separately to serve different user groups, or it can be assembled into a service group, sharing resources, and running a single task assigned. It has the characteristics of high parallelism, real-time collaboration, low power consumption, low price, and small space. Chen Shiqing introduced to the "Daily Economic News" reporter.
When talking about the development trend of future supercomputing, Chen Shiqing told reporters, "The current supercomputer has reached a development bottleneck due to too much energy consumption, and the future supercomputing will evolve from centralized to distributed, going deep into communities, streets, industrial parks, universities and research institutions." At the same time, supercomputing and brain science will also move towards deeper integration. ”
"The supercomputing of the future should be distributed, and the whole world should become a net"
NBD: At present, mainstream supercomputers basically adopt the centralized computing power model, what are the shortcomings behind the realization of super computing power output? What will be the evolution of supercomputing in the future?
Shiqing Chen: From the 1980s to the 1990s, I was only doing one thing – inventing and building the world's fastest supercomputer. On the basis of the system architecture I designed, centralized supercomputers began to develop rapidly, widely used in defense, petroleum, aerospace, aviation, automotive, nuclear power, meteorology, chemicals, materials, biology, pharmaceuticals and other fields, such as the United States AT&T, Boeing, General Motors, Mercedes-Benz and so on are my customers.
Of all the larger supercomputers in the world, 80 or 90 percent of them use the system architecture and concepts I created at that time, from single to dual, one to two, two to four, four to eight, evolving into thousands of parallel computing.
But as supercomputers get bigger and faster, their energy consumption is getting higher and higher. The world's top ten supercomputers currently need a power station of at least 20 megawatts to support when running, consuming more than 20,000 kWh of electricity per hour, equivalent to the daily consumption of a small town.
Although I created a centralized supercomputer architecture, the huge energy consumption problem was always unsolvable. Without the financial support of the state or large enterprises, this model is difficult to sustain and widely applied to the new scenarios of the future digital economy. At the national level, such large centralized supercomputers still have huge demand in fields such as lunar exploration, meteorological monitoring, and oil exploration. However, from the market level, due to its huge energy consumption, many small and medium-sized enterprises are difficult to afford.
So, in the early 2000s, I began to rethink another system architecture—from centralized to distributed. Its core purpose is to make supercomputers consume less energy while being faster. It's a completely different field.
NBD: How do you understand that all the supercomputers of the future should be a huge "distributed network"?
Chen Shiqing: I think that in the future, all supercomputing should be distributed, supplemented by artificial intelligence, big data and other technologies, and the world will become a network. The edge/bottom layer of this network is the community, street, industrial park, etc., where there is big data for all applications, and supercomputing can be accessed at any time. This is a completely new idea.
At present, we are still in the pilot stage of developing grid format intelligent supercomputing. For example, divide a city into 5,000 or 50,000 grids, set up a small supercomputer for each street or each neighborhood, and all local data can be calculated, transformed, and served locally in real time. Then all the communities come together to form a huge distributed grid. Similar to the current power grid, in the distributed network of supercomputing, the future computing power, computing capacity, and algorithm are also distributed and can complement each other.
For example, we can design the underlying grid format intelligent supercomputing designed in Shanghai Jiading to upload the big data of various communities in a centralized manner; we can also build a grid format intelligent supercomputing in hangzhou's West Lake District, digitize all urban changes, dig deep into and integrate data on the same platform, and then use it for intelligent medical, education, transportation, manufacturing, cultural and creative, energy, finance, environmental protection, security and other terminal services.
NBD: As a top scientist spanning multiple eras, how have you observed the evolution of the concept of human-computer interaction, from minicomputers and giant machines to supercomputers to cloud computers?
Chen Shiqing: The concept of human-computer interaction has evolved over a long period of time. The earliest human-computer interactions were centralized and mechanical, throwing everything there and letting the supercomputer help calculate. Nowadays, the concept of human-computer interaction has transitioned from the previous mechanical and centralized to a more intelligent and personalized model.
The future of supercomputing should be closer to you and more personal. For example, where you live now, there is a small distributed intelligent supercomputer next to your community. It needs data from this community, and its complete application scenarios are also in this community, and serve the people of the entire community.
Suppose one day, a heart patient suddenly falls on the road. How do we know immediately what type of heart disease this patient has? Which hospital has a specialist in this type of heart disease and is closest? How to dispatch the rescue force as quickly as possible? These require the use of intelligent supercomputing to quickly calculate all data, including patient conditions, location, road conditions, hospital conditions, etc. This is one of the important directions and application scenarios for the future development of intelligent supercomputing.
"The future of supercomputing should be developed in the direction of brain-like"
NBD: In addition to the trend of distributed and gridding, how will supercomputers evolve in the future?
Chen Shiqing: I think that the future supercomputer should also develop in the direction of "brain-like", with small size, fast computing speed and low energy consumption.
In fact, the daily consumption of the human brain is only 25 watts, but it can command about 100 billion neurons and process huge amounts of information every day. The human brain, as the most delicate intelligent organ of the human body, has stronger "performance" than supercomputers. At present, the development of the human brain is only 5%, if you can use brain science to enhance the development of the human brain and combine it with supercomputing, it will create a very valuable application scenario.
NBD: You prepared to establish the Third Brain Research Institute in Silicon Valley in 2013, and in 2019, it was officially established to focus on brain-computer fusion research. So what do you think is the core idea of the "third brain"? What are the main research directions around the "third brain"?
Chen Shiqing: In fact, more than a decade ago, my research direction turned to brain-like computing, hoping to rely on "elastic distribution" so that supercomputers can be as low energy consumption and high efficiency as human brains, while improving the degree of brain development. For me, such a turn is tantamount to a "self-revolution."
In addition to the brain (first brain) and cerebellum (second brain), each of us has an "extended" "third brain". The "third brain", also known as the bionic brain, is a comprehensive science that combines cognitive neuroscience, bioinformatics science, biological systems science, biological engineering science, life science, clinical medicine, information engineering science, and supercomputer science. Relying on a highly trusted, secure, efficient, distributed and collaborative Hypergrid Cloud platform, we have conducted R&D efforts around the world.
At present, the "third brain" mainly has four major research directions - researching the brain, protecting the brain, developing the brain, and extending the brain. Research in the field of developing the brain is relatively mature.
The first direction is to study the brain, to gain an in-depth understanding of the underlying biological mechanisms of the brain. Only after understanding its basic architecture and why it is intelligent can we do brain-like computing. Imagine a brain that commands about 100 billion neurons and consumes only 25 watts of energy, half of which is only for the brain's biological needs. Such a structure is full of mysteries.
The second direction is to protect the brain, after in-depth study of the brain, we know what its shortcomings and weaknesses are, so that we know what protective measures to take to prevent brain damage and find early and accurate detection and intervention methods for brain diseases. Our research focuses on how to carry out early detection, predict and intervene in advance from the information level, physical level, chemical level, etc., reduce the risk of disease, and move the "window period" of treatment forward.
The third direction is to develop the brain, and to study and develop efficient learning methods for brain-computer fusion. God has given us such a smart brain, we should focus on how to realize the potential of a person's brain, not to be suppressed by other factors.
We piloted it at Fremont High School in the United States, and eventually proved that "third brain" related technologies can significantly improve the speed and effectiveness of learning. I hope that in the future, we can use our theories and technologies to fully stimulate the learning potential of the brain, kindergarten to the doctoral age of 17 or so can be completed, and then through 3 years of social practice, a person to 20 years old can have doctoral level thinking and analysis ability. "Just as I'm 77 years old, my brain is only in my 30s, and my ability to think and innovate is now on par with when I was in my 30s."
The last direction is to extend the brain, extending the brain-computer fusion technology to the cloud, realizing lifelong companionship from small to old.
NBD: Today, research in the field of brain science has long been on the cusp of the tech circle, tesla CEO Elon Musk, Google founder Sergey Brin and so on have entered the game. So, what are the cutting-edge application scenarios of brain science?
Chen Shiqing: Taking the field of intelligent medical care as an example, we are working with world-renowned medical schools to study brain diseases and cancer prevention, with the help of brain science and artificial intelligence to achieve "uncertain prophets", diagnose diseases in advance, and realize non-invasive disease treatment at an early stage.
For example, according to the study of the brain optic nerve, before Alzheimer's patients have begun to lose memory, the newly developed precision instruments are used to detect the decline of brain cognitive function early, and non-drug physical intervention and rehabilitation are carried out to improve the quality of personal life. At present, Parkinson's syndrome, autism, depression, Alzheimer's disease, epilepsy, these 5 brain diseases are the key projects of intelligent medical treatment.
In addition, taking intelligent education as an example, we hope to find a law from brain science, which can determine very early whether a child is suitable for scientific, technical, engineering, artistic or medical learning education.
Brain science combined with artificial intelligence will transform the education system of the future. China's first demographic dividend has passed, in the past we relied on labor-intensive, and now the society has entered the aging stage, which needs to rely on the power of science and technology. Brain science can accelerate the release of young talents into society and help achieve the second demographic dividend.
NBD: Previously, when Elon Musk talked about brain-computer fusion, he believed that humans had only one choice - to become artificial intelligence (AI). What do you think of such a statement?
Chen Shiqing: Actually, Musk sometimes expressed it a little more extremely. The brain-computer fusion and brain-computer interface technology he talked about is a combination of brain science and AI technology, which helps people with disabilities live a normal life by implanting chips.
But I think another important direction for brain science is to unleash the brain potential of most people and promote thinking innovation. The human brain is the smartest, and we can always think about inventing a new tool to enhance some aspect of our capabilities, just like telescopes and microscopes for the human eye, cars, airplanes for the legs. Artificial intelligence is only part of the "extension" of people, and it is not enough to have artificial intelligence, it must be deeply combined with brain science.
The best human-computer interaction in the future is actually a heavy and complex work done by artificial intelligence, and then through brain-computer fusion technology, let our brains focus on thinking about innovation.
"China's supercomputing has reached the world's leading level, but there is a shortage of application-oriented talents"
NBD: Over the years, your scientific research experience has been inextricably linked to China, which has promoted the development of China's medical and health big data platform. Please share your experience in participating in a supercomputing project in China.
Chen Shiqing: As early as 2004, I decided to return to China, and initially designed my third-generation super blade computer in Shenzhen. At that time, it was not sold at the first time, but brought the new generation of supercomputing to China's poor and remote rural areas and ethnic minority areas.
We have set up four clinical pilots in these areas, including Kangbao County in Zhangjiakou, Hebei, Xilingol League in Inner Mongolia, and Kazakh Autonomous County in Aksai, Gansu. My team and I spent 7 years in-depth local research, collected and excavated a large amount of medical data, especially local chronic disease big data, successfully established an information grid model for chronic diseases, realized the interconnection of basic medical care at the county, township and village levels, formed the 1.0 version of today's China's chronic disease medical system "medical association", and also promoted the development and improvement of the "new rural cooperation" medical payment system in the early rural areas.
In 2007, then-Minister of Health Gao Qiang inspected Zhangjiakou Kangbao County and was very satisfied after seeing our medical data and grid model, "Great, this is China's future health platform." Later, under the impetus of Minister Gao Qiang, 200 counties received active funding from the national medical reform fund, and the medical big data grid model was promoted in more places, and later gradually became the national chronic disease management platform."
From 2006 to 2012, my entire focus shifted to Beijing, independent financing team, continued to develop the 4th, 5th, 6th and 7th generation blade supercomputers, and developed hospital medical software, establishing China's earliest health cloud computing and big data service platform.
Originally, our idea of version 1.0 was to combine counties, townships and villages to create a "medical association". In the future, we hope that all of China will be a net, with a health code. People are always mobile, but under a net, no matter where you go, your chronic diseases and your family history can all be called up in a timely and accurate manner.
NBD: In recent years, China has gradually occupied the commanding heights of the world in the number of TOP500 high-performance supercomputers, and the supercomputing represented by "Sunway Taihu Light" and "Tianhe-2" has repeatedly led the world in computing power. Based on your observation, what is the level of Supercomputers in China in the world?
Chen Shiqing: In fact, from the perspective of system design and manufacturing, China is already at the forefront. Many of the world's top ten supercomputers come from China. For example, not long ago, at the Global Supercomputing Conference, the Chinese supercomputing application team led by Zhijiang Laboratory won the 2021 "Gordon Bell" Award, the highest international award in the field of supercomputing applications, with the quantum computing simulator developed by the new generation of Shenwei supercomputer , "Superscale Quantum Stochastic Circuit Real-time Simulation" (SWQSIM). This is one of the concentrated microcosm of China's supercomputing development level.
NBD: So, what are the shortcomings in china's current supercomputing development?
Chen Shiqing: First of all, There is a shortage of supercomputing application talents in China. The talents cultivated in China are concentrated in large institutions such as the Chinese Academy of Sciences, mainly researching centralized supercomputing, the field is relatively narrow, and only state institutions can afford it.
The future distributed intelligent supercomputing will be a small supercomputing scattered to communities, streets, and industrial parks, and there will be more subdivided and complex application scenarios. In view of this, there is still a large gap in China's artificial intelligence application talents, and at least 5 million related development talents are needed at present. Previously, we established the School of Artificial Intelligence at Chengdu Jincheng College, hoping to cultivate more high-end application development talents, and planned to replicate the successful training model to 100 colleges and universities.
In addition, when the market talks about supercomputing, there are more specific industries and products, and few people talk about computing power. In fact, the development of Intelligent Supercomputing in China is a large gap in computing power. In the future, all artificial intelligence scenarios need algorithms, and algorithms need computing power support, which needs to be vigorously developed.
At present, some cities realize that the core of the development of the digital economy represented by artificial intelligence lies in computing power, and have begun to make relevant layouts, and Huawei, Inspur and other enterprises are also participating in the construction of urban computing capacity. China should take industrial parks and communities as the core, lay out new infrastructure and computing power such as 5G base stations, and create a good supporting environment for the innovative development of small and medium-sized enterprises.
Finally, only by combining brain science, artificial intelligence and intelligent supercomputing can we complete the application scenarios from the end to the edge and establish a complete industrial chain.
NBD: Final question, what did you go through from joining Clay, which monopolized the world's supercomputer market, to parting ways with founder Seymour Clay? As a top scientist and veteran entrepreneur, how do you balance technology exploration and business?
Chen Shiqing: Because the supercomputing system he studied was surpassed by me (laughs). At that time, my second generation of supercomputing was developed, and his second generation had not yet come out. Slowly, the entire company's revenue "depended on my products."
In fact, it costs a lot of money to develop each generation of supercomputing to support a research and development team of hundreds of people. At that time, the company could not support both R& D teams that were "burning money". On the one hand, Clay is the founder after all, and the company must give priority to supporting his ideas and research; on the other hand, I bring a lot of commercial benefits to the company, but I still can't raise enough follow-up research and development funds. Over time, the two of us "separated", and then I voluntarily left the company with a research team of 45 engineers.
All along, I have been exploring more frontier areas. My supercalculation certainly generated significant gains, but I put it into my next research. Research – new product launch – commercial monetization – research, and so on.
My outlook on life is that I will benefit more people with my scientific research results, whether Chinese americans or Americans, the whole world needs help, and it is good to be able to do this! I want people in the tech industry to realize their ideas and values. The purpose of doing scientific research is not to make money, but to make the technology dividend benefit all people and help people at the bottom of society solve difficulties. The way of science and technology, with the rain!
Reporter | Xie Tao
Edited | Gao Han Duan Lian Du Bo
Visual | Liu Qingyan
Typography | Gao Han Wang Shujie
| This article is copyrighted by "Every Toutiao" |
Reproduction, excerpting, reproduction and mirroring without permission are prohibited
If you need to reprint, please apply to the background of this official account and obtain authorization
The Delta strain is still in the global pandemic, and the Olmikron strain is adding fuel to the fire,