Scientific Hypotheses and Huawei’s Business Vision for the Future
On Day 1 of Global Huawei Analyst Summit 2022, Dr. Zhou Hong, President of the Institute of Strategic Research at Huawei delivered a keynote speech exploring cutting-edge scientific hypotheses, Huawei’s business vision, and ten challenges that we need to address to unlock the intelligent world. Below is the full transcript of his speech.
Today, I’d like to share some of our scientific hypotheses and Huawei’s business vision for the future.
The 18th century was the era of mechanization. The 19th century was the era of electrification. And the 20th century was the era of informatization.
What will the 21st century be?
I think the 21st century will be the era of intelligence. At the heart of an intelligent world is sensing, connectivity, and computing, which will advance our ability to understand and control matter, phenomena, life, and energy.
There will be many challenges on our way to an intelligent world. If we want to live happier lives, improve efficiency, and make the world a greener place, we’re going to need to enhance sensing, connectivity, and computing by several orders of magnitude. But at the same time, we haven’t seen any major breakthroughs in underlying science and technology for the past few decades, and many technologies have pretty much reached a bottleneck.
So how can we find a viable path forward?
We have to meet the future with bold hypotheses and a bold vision, and throw caution to the wind as we push to break through bottlenecks in theory and technology. This is the only way forward.
Digital technology will enrich life and work
Over the past decade, with the rapid development of broadband communications, smart devices, AI, and cloud computing, digital technology has greatly enriched our lives.
Thanks to digital technology, we can make phone calls, surf the Internet, send instant messages, find our way around using the maps on our phones, use e-banking services, and shop online. ICT has become an increasingly integral part of our lives. In addition to the application of ICT in daily life, Huawei has also spent the past decade working with partners to explore the use of ICT in many industries.
For example, we worked with carmakers and telecom carriers to conduct trials on highways. When something unexpected happens on the road, it usually takes human drivers seconds to react. However, with high-performance vehicle-to-vehicle and vehicle-to-network connectivity with a latency of just 10 milliseconds, the time it takes to identify and respond to unexpected events could be shortened by more than a hundred times.
In addition, the distance between cars on the road could be reduced from tens of meters or a hundred meters, which is the distance required to stay safe in manual driving, to 0.8 meters, with the cars moving at speeds of up to 100 km/h. This will greatly increase the capacity and safety of highways.
Moreover, this can support platooning, which helps reduce wind resistance, resulting in a 20% cut in fuel consumption. The Internet of Vehicles (IoV) can also support remote driving and enable new operating and service models.
We have also conducted trials in urban settings, and found that the IoV-based Beyond Line-of-Sight (BLOS) communications and cooperative vehicle-infrastructure system are capable of increasing traffic efficiency by 30% and reducing traffic accidents by 90%.
Today, ICT is being applied in factories, hospitals, ports, and coal mines, empowering digital and intelligent transformation of various industries.
A future of unlimited potential
In the future, we believe the potential of ICT can be realized in many more areas.
For example, ICT can help promote health and happiness. Wearable sensors, wireless communications, and cloud computing can support fitness and healthcare, as well as the management of chronic diseases. In addition, AI computing can increase the design and screening efficiency of medicines and vaccines.
ICT also enables autonomous and intelligent robots in a wide range of fields. These applications can improve quality of life, as well as the operating efficiency of industries.
ICT can also help build a green, sustainable environment. For example, it can enable efficient energy conversion and scheduling, and assist in designing energy conversion catalysts and energy storage materials at lower cost and higher efficiency.
Last but not least, ICT can enable a hybrid digital world with immersive experiences, enriching our lives, improving the efficiency of learning, and empowering industries to quickly iterate in the digital world.
Much more than “100x growth every 10 years”
Global digitalization is growing at an exponential rate, driven by intense demand.
Globally, mobile broadband data consumption increased from 0.24 exabytes per month in 2010 to 60 exabytes per month in 2020, a 250-fold increase in 10 years.
Mobile broadband data in China grew from 0.033 exabytes per month in 2010 to 13 exabytes per month in 2020, a more than 400-fold increase. In the future, we believe that digital technology will develop at a rate of over a hundredfold per decade. Digitalization will accelerate the development of humanity and society.
However, many theories and technologies currently applied were proposed decades or even a century ago. We are facing bottlenecks in terms of developing new applications based on such theories and technologies, for example, the Nyquist sampling theorem and Shannon’s theorem in the communications domain, commutability theory and Von Neumann architecture in the computing domain, and Moore’s law in the semiconductor domain.
We hope that new hypotheses and vision will come forth to guide the next breakthroughs in technology. Therefore, we want to propose four pairs of scientific hypotheses and business visions. We want to explore these areas and conduct future-oriented research together with partners in academia and industry.
Pushing back the frontiers of knowledge: physics, chemistry, and biology
First, we should extend the boundaries of our knowledge through exploring fundamental science and cutting-edge technology, particularly in physics, chemistry, and biology. Breakthroughs in these areas will help us create new materials and components such as molecules, catalysts, and proteins, as well as new equipment and process techniques.
I once had the opportunity to meet with a quantum scientist and discuss the possibilities of storing photons and quantums. This was the same man who, way back in 1993, proposed the concept of quantum storage. At that time, few people believed it possible. People wondered how light could be stored in a bottle? Would the process of storing quantums affect their state? It wasn’t until 1998 that Lene Hau and her team at Harvard University succeeded in slowing a beam of light to about 17 meters per second, using the effects of electromagnetically induced transparency (EIT). And in 2000, Hau and her team succeeded in “freezing” photons for one minute. In 2006, John Pendry at the Imperial College London and others proposed that something similar to a “photon black hole” could be created to “grab” light heading towards it. Now many methods have been developed to realize quantum storage, paving the way for the implementation of quantum communications and quantum computing.
To reduce power consumption and improve reliability of semiconductor components, we are also working with researchers to explore the thermal mechanism in semiconductor components to see if we can create conditions to accelerate the conversion of optical phonons to acoustic phonons, reducing the temperatures of the gate electrode and drain of semiconductors.
Currently, superconducting quantum computers usually operate at milli-Kelvin temperatures. Some researchers are exploring ways to further lower the temperatures by one million times, such as through the use of lasers to cool atoms, to achieve nano-Kelvin temperatures. If temperatures could approach absolute zero, would more complex quantum phenomena be discovered?
In the future, will it be possible to predict the characteristics of new materials through computing, rather than through lengthy trials? The answer is certainly yes. For example, using the USPEX computation method, with computing power of one million core hours, we can calculate the characteristics of molecules composed of up to 200 atoms. In 2017, researchers use calculations to reveal the structure of a new super-hard material made of tungsten boride, WB5. This solved a puzzle that had plagued the scientific community for 60 years. In 2019, through calculations, researchers found that thorium decahydride ThH10 exists at pressures above 0.85 million atmospheres and exhibits amazing high-temperature superconductivity, and the critical temperature was found to be −112 °C.
In addition, through computational chemistry, it is expected that more effective catalysts, chemical pharmaceuticals, biopharmaceuticals, and vaccines will be discovered and invented.
Transcending biological limits to better understand nature and ourselves
Second, we will expand the perception of the world and perception itself, going near and beyond human perception, from sensory substitution to sensory enhancement, and from human to machine perception.
We can draw inspiration from nature. Through millions and even billions of years of evolution, many animals have developed extraordinary perceptual capabilities that are far beyond human perception and existing machine perception.
Take vision, for example. Some spiders easily outperform humans in terms of detecting the outlines of objects in motion. Spiders evolved this sensitivity so they could capture prey quickly and precisely. I think autonomous driving may need such “spider eyes”. Then there are frog’s eyes, which are highly sensitive to single photons and have excellent night vision. And dog’s noses are 1,000 times more capable of distinguishing odors than humans.
The enhanced perception capabilities can also be used to sense and control the human body. Technologies like ECG, EEG, and PPG have not yet developed in a systematic, convenient, and low-cost manner. There is still much left to do in terms of monitoring the eight body systems. In the future, new types of sensors will enable real-time, unobtrusive measurement of important health indicators such as blood pressure, blood sugar, and the heart’s rhythm and electrical activity. Brain-computer interfaces and muscle-computer interfaces will be developed for better collaboration with machines. It may be possible for humans to communicate, drive, work, and play through thinking.
We will also develop a hybrid digital world that offers new experiences, such as 3D displays and virtual touch. This virtual world will be able to be touched and will look just like the physical world.
New computing models and implementations to understand the world and solve problems
To better understand the world, solve problems, and create value, we need new computing models that are adaptive to different purposes and environments and easy to implement.
Along with advances in information theory over the years, more than a dozen computing models have emerged and seen wide adoption. For example, we have the butterfly structure of the fast Fourier transform commonly seen in wireless and optical communications; the finite state machine model, which supports state transition, used in routers; and the most recent statistical and correlation-based models used in AI applications. These computing models are the fruits of the hard work of generations of mathematicians and engineers. Should we stop here? No. I believe there’s still ample room for exploration.
As communications systems continue to evolve towards high frequency and high speed, there will be more challenges that come with nonlinear channels and components. In light of this, we are considering whether we can shift from the linear Fourier transform to the nonlinear inverse scattering transform, to better match future applications.
Statistical models that are widely used in today’s AI computing cannot be explained or debugged and are highly energy-inefficient, making them ill-suited to rapidly proliferating AI applications. In fact, and in relation to this, we can learn from living creatures.
An ant’s brain consumes just 0.2 mW of power, but can run around and process many activities, like nesting, finding food, and herding aphids. More importantly, they don’t need to rely on deep learning or follow computability theory or Von Neumann architecture. In contrast, an average autonomous vehicle today consumes dozens, or even hundreds, of watts of power, which is far less energy-efficient than an ant.
So, we are asking ourselves questions: Are there alternatives to statistical and correlation-based computing models for AI computing? What about mathematical logic-based computing models, geometric manifold computing models, or game theory-based computing models?
Matrices are widely used in scientific computing. To multiply 2 n-by-n matrices, the complexity involved is the third power of n using a naive algorithm. In 1969, the Strassen algorithm introduced by the German mathematician reduced the complexity to the 2.807 power of n. At the end of 2020, MIT professor Virginia Williams and Harvard professor Josh Alman proposed an algorithm which brought down the complexity to the 2.3728596 power of n.
In terms of matrix computation, we are more focused on solving sparse linear systems of equations. There are a few billion people on Earth, but social science tells us that each person is equipped to maintain no more than 200 relationships at a time.
In chip design, most of the constraints on components are local. In this area, Georgia Institute of Technology researcher Peng Yang and his colleagues created an advanced algorithm with a computational complexity of the 2.3316 power of n. For this achievement, Peng won the 2021 Best Paper Award from SODA, a leading symposium on computational theories. A few months ago, our mathematicians created a revised algorithm, further reducing the complexity to the 2.28 power of n, 0.0516 power of n less than Peng’s algorithm. What does this improvement signify? It means that if n represents one million, our algorithm can further reduce the computational complexity by about 45%.
In terms of model implementation, a supercomputer often needs to consume immense power to deliver high performance. For example, 500 PFLOPS of computing capacity requires 30 megawatts of power. However, a human brain can achieve about 30 PFLOPS with just 20 watts of power, which is 80,000 times more efficient. Anticipating these challenges, we may need to develop more efficient computing architectures and new components, going beyond the conventional computability theory and Von Neumann architecture, to make the most of new computing models.
Going beyond Shannon’s limit to unleash the full potential of ICT
The next step is to go beyond Shannon’s limit to unleash the full potential of ICT, stretching the limits of space and time. We may want to explore ways to overcome spatial barriers and build real-time global connections, helping converge the physical and digital worlds and connecting robots everywhere.
For life-size holographic communication to happen, we will require a bandwidth of about two terabytes per second and latency in the range of one to five milliseconds, if data is not compressed. An autonomous vehicle with 12 cameras could generate up to four terabytes of data a day, and current 5G networks fall far short of providing such capacity.
Can we tackle these challenges theoretically and technologically? I believe so.
In theory, if the physical world had memory, we could move past Shannon’s three theorems.
Looking at it from an engineering standpoint, one quantum cascade laser can emit light of hundreds of wavelengths at the same time, generating traffic of up to 100 terabytes. A high-frequency attosecond laser, which has yet to be developed, may even generate millions of terabytes of traffic.
If these technologies are transplanted into wireless and optical communications, could the performance of our communications networks be improved a thousand fold, or even ten thousand fold?
Advancing knowledge and creating value
To bridge scientific hypotheses and business vision, Huawei divides the innovation process into five inter-connected stages, from hypothesis and vision to innovations in theory, technology, and business.
Back-end innovations related to business, customers, and users are more likely to produce tangible results, while more front-end innovations related to hypothesis, vision, and fundamental research take longer to materialize.
Moving toward the future, we need to take bold steps to advance more front-end innovations in fundamental research.
In addition to supporting Bohr-quadrant work – the kind of fundamental research driven by interest – we hope to work with partners to drive the kinds of innovations that fall in the Pasteur’s quadrant. Together, we can extend the reach of scientific knowledge while simultaneously creating value through application.
Ten challenges ahead
Based on these four pairs of hypotheses and visions, we have identified two scientific questions and eight technological challenges that we can work on together, focusing on innovations that fall in the Pasteur’s quadrant.
The two scientific questions are:
- First, how do machines perceive the world, and can we build models that teach machines how to understand the world?
- Second, how can we better understand the physiological mechanisms of the human body, including how the eight systems of the body work, as well as human intent and intelligence?
The eight tech challenges to be addressed are:
- How can we build on human-machine interfaces and develop new sensing and control capabilities such as brain-computer interfaces, muscle-computer interfaces, 3D displays, virtual touch, virtual smell, and virtual taste?
- How can we have blood pressure, blood sugar, and heart health continuously monitored as unobtrusively as possible? Can we use strong AI to help discover new chemical pharmaceuticals, biopharmaceuticals, and vaccines?
- How can we develop application-centric, efficient, automated, and intelligent software for greater value and better experience?
- How can we reach and circumvent Shannon’s limit to enable efficient, high-performance connectivity both regionally and globally?
- How can we develop more adaptive and efficient computing models, non-Von Neumann architectures, unconventional components, and explainable and debuggable AI?
- How can we harness the power of AI to develop new molecules, catalysts, and components?
- How can we develop new processes that surpass complementary metal-oxide-semiconductor (CMOS) fabrication, cost less, and are more efficient?
- How can we develop safe, efficient energy conversion and storage, as well as on-demand services?
Together we can do more, be more
With an open mind, Huawei is innovating together with partners around the world.
On April 30, we will be launching our virtual Chaspark which will serve as an openly accessible platform where people can exchange ideas on science and technology.
Through Chaspark, we will put together the most pressing challenges in the ICT industry and keep attracting the world’s best minds so that we can take a crack at these challenges together.
Everything we imagine today is too little for tomorrow
Thirty years ago when I was in college, we had to wait in huge lines just to make a long-distance call. None of us could have imagined that one day we would be able to video chat with our families anytime and anywhere with only a small, wireless gadget. And today, we are using this same gadget to do much more, connecting ourselves with the rest of the world. It would have sounded like science fiction to us three decades ago.
Everything we imagine today is very likely to be too conservative – too little – for tomorrow, so we need to be bolder.
And that’s why we’re here today. We hope to join hands with the academic community and industry partners to put forward bolder hypotheses, rethink fundamental theories, reshape architectures, and reinvent software. Together, we can shape the future!
Click the link for more information about Huawei Global Analyst Summit 2022, which runs from April 26-27.
Disclaimer: Any views and/or opinions expressed in this post by individual authors or contributors are their personal views and/or opinions and do not necessarily reflect the views and/or opinions of Huawei Technologies.