On November 6th, nearly 1,000 famous experts in the field of intelligence, experts and scholars in urbanology, city administrators, media and entrepreneurs gathered at the 2016 China (Hangzhou) Artificial Intelligence Industry Forum to launch a unique artificial intelligence keynote speech.
Learning that there are still many fans who can't attend the conference, I specially compiled the notes for the attendees' reports in shorthand, so that everyone can see it quickly~
This article is organized according to the notes
Artificial intelligence is running
Chairman of Chinese Association for Artificial Intelligence, Academician of Chinese Academy of Engineering
Li Deyi: Leaders, colleagues, I am very happy to discuss artificial intelligence and future situations in Hangzhou again. Today I want to focus on what we have done on artificial intelligence, and whether our work can spark in Jianggan. The history of the development of science and technology is the history of the expansion of human ability to interpret and change the world. It is the history of the development of human labor tools. Human beings have gone through farming, industrial, and information societies, and have entered a phrase when the utilization of "intelligence" becomes the mark of today's society and, especially, a phase when production tools are developing from power to intelligence. We used to rely on resources, capital, labor and other factors to support the rapid growth and scale expansion of China's economy, but this has become history. To accelerating the transition from factor-driven development to innovation-driven development, artificial intelligence is a solution. In farming societies and industrial societies, human production tools, especially the power tools based on matter and energy are largely developed; today, labor tools turn to intellectual tools based on data, information, knowledge, value and intelligence. Demographic and labor power dividends does not function any longer, the time for profession dividends comes. This is the artificial intelligence we are going to talk about.
Today, not only materials, energy, traditional manufacturing and power tools, but more importantly, intelligence and intellectual tools have become a new entry point for human beings to expand their interpretation of the world and their ability of changing. China is the world's most populous and a country with abundant intellectual resources. Intelligence has risen to the height of national strategy, and intelligent science and technology are more important than ever before in economic prosperity, national security, population health, ecological environment and quality of life. Intelligence is the need of the times to enhance the supply of innovation-driven development. Whether it is human intelligence or artificial intelligence, whether it is individual intelligence or collective intelligence, whether it is concentrated intelligence or network intelligence, it is the supply power that enhances the source of innovation-driven development, the original driving force of innovation, and the core productivity in productivity.
The robot will become an important companion for human beings to recognize nature and society, expand intelligence, and move toward smart life, and print a new form of human networking and thing interneting. China will become the largest market for robots. Robots are the jewel of the top of the manufacturing industry and an important symbol of national technological innovation and high-end manufacturing. Only original intelligent science and technology can make us an important maker and leader of robot products and their market rules. It is imperative to regard "China Artificial Intelligence 2030" as a national special project of "Science and Technology Innovation 2030"; it is imperative to add "Intelligent Science and Technology" as a national first-level discipline.
With this precious time, I will talk about two typical cases. Let's talk about Go's brain first. On September 16, 2011, Chinese Association for Artificial Intelligence Society organized the “Jiu Lu” man-machine war of Go. The Go robot won, and CCTV and other media reported it on the spot. On March 9, 2016, Lee Sedol and AlphaGo played Go. To surround the stones of the other player, there are many more steps in Go to deal with in a certain state than chess. From a statistical point of view, there are 250 possible moves for a particular situation. This is two important networks for the success of AlphaGo Go. AlphaGo broke through the traditional process and built two sets of deep convolutional neural networks that mimic human thinking. It fully learns from human chess players and learns from massive samples. The confrontation between the chess masters and the achievements of artificial intelligence development is a scientific experiment that allows the chess player to act as a tester. It is a confrontation between a person and a group of people, and is a confrontation between a biological person and a "human and its mixed creatures".
AlphaGo is just a Go brain. It is not a Go player, let alone a Go robot. It has no eyes and hands. It has no ability to perceive and behave. It only has the ability to calculate and infer. Lee Sedol complained that AlphaGo had no emotions and one could not analyze the psychological state of the opponent without emotion.
AlphaGo defeated the 9-segment Go master Lee Sedol 4:9, and the world was shocked. Can AlphaGo and Lee Sedol have a Fu Pan(replay the chess board and analyze the moves so as to do better the next time)? Before and after the game, did the AlphaGo program change? Let the same version or different versions of AlphaGo play, what is the ending? Go is the traditional culture of China. If we are always like this, it seems that one of our masters has lost. Is our level of Go low?[KC1] These questions deserve our deep thought. We believe that robots should have the ability to learn. If we let driverless car, or the wheeled robots and human racers compete in a game, where both racers and robots have full perception, cognition and behavior abilities, then the “men-machine war” will be more exciting.
The car, which was once called the world-changing machine, is being changed by artificial intelligence and the world. In order to improve the quality of human movement, it is necessary to fundamentally change the driving style of vehicles and free people from low-level, cumbersome and long-lasting driving activities. There are very difficult problems in vehicle dynamics. We hope to build sensing components, such as radar, on the basis of ensuring vehicle dynamics and artificial intelligence. At the same time, we need to implement assisted driving, and then develop driving brains to complete autonomous driving.
The formalization of driving cognition emphasizes how to replace the driver's perception and cognitive ability with artificial intelligence technology. We divide driving into grades. This is offline assisted driving. In local time and local areas, such as highways, you can drive automatically, but these two basic driving methods are mainly manual driving. We hope to achieve self-driving, autonomous driving, unmanned driving, people-oriented, man-machine cooperation, similar to knights and horses.
There are two recent words, autonomous driving and assisted driving. Many people think that the car should drive itself autonomously, and if the program is not suitable, it is adjusted by the person. In essence, it is the programmer that teaches the robot to drive through the program, and the car becomes a software-defined machine to realize automatic driving. I have raised doubts about this. Do we need such technology? The vehicle is full of uncertainty during driving. If the car is software-defined machine then as we cannot be sure how long any auto-driving window that meets the current driving conditions will last, it is necessary to switch to manual driving anytime, anywhere. Therefore, "automatic driving" is difficult to automate.
I have a multimedia animation here, it shows that when a automatic car moves together with two manual vehicles, what will happen. The animation tell us that after the recognition is interfered, the default auto-driving mode collapses and almost all turn into manual driving, so we call it an automatic driving trap. In this process, our innovation focuses on how drivers interact with the road environment change and the surrounding flow, how vehicle's movement behavior can allow the surrounding vehicles to perceive their own existence, and how to turn interactions with the nearby vehicles through lights and gestures into control of the own vehicle. Who is going to do this? Where does the driver's interaction with the environment and surrounding vehicles during driving go? I want to question this model. Autonomous driving is equal to the automatic driving mode. I have doubts about this equal sign. We have studied mathematics, learned logic. How many autopilot modes can fully cover a wide variety of driving scenarios? We raised doubts.
The difficulty in the development of smart cars is not only the nature of car dynamics and various sensor requirements, but more importantly, it is to develop a "machine-driven brain" that is as flexible as the driver, to simulate the real-time prediction and control of the real person in the circuit to deal with the uncertainty of the vehicle driving. It is unbearable to let the car manufacturer to simulate the autonomous driving in the circuit. It is unbearable to let the car manufacturer do research and development, so I am very happy to see the Project One of Secretary Wang. I think it will be better if Jianggan District is able to produce(the cars). With the help of the existing cognition, I divided it into three major blocks: GPS+IMU, radar, and image. Each sensing system is functioning to complete positioning, road rights detection, and navigation. It is the driver who can really teach the robot to drive. It shows that the car has been assimilated into a part of the human body and becomes a true extension of the seamless connection with the body. Therefore, the machine should be able to “sneak” self-learning while the driver is driving. We expanded the debug bus to a self-learning bus, created a self-learning section, completed statistical learning and evolutionary learning, and let wheeled robots drive like humans and develop smart cars with individuality. Once the current cognition is mixed with the memories of past experiences, driving the brain can make a suitable judgment for an uncertain future. The driving situation and the cognitive arrow chart mapping library generator forms a driving memory stick. The next time when the driver does not drive, the driving brain will drive according to the current driving situation.
If we have cloud computing with every smart car connected, then the big data is formed, we can complete these things with recognition and navigation. Looking back at the competitions we have participated in in the past ten years, the various perceptions and cognitive means of smart cars are interdependent, and we have found new paths in many attempts. Our products and products in the future, driving brains, can pass the driving test like a human. It is more like a driving skill, which cannot be replaced by pace notes and maps. it can promote personalized driving and benchmark driving.
On November 24, 2012, a milestone test was completed in Beijing. Four years ago, we conducted 18 trials in Beijing. (Play a video) Many people ask if China's artificial intelligence is backward? Ten years ago, we used a vehicle which used for military traffic and equipped with radar. We completed 18 trials from Beijing to Tianjin. In 2015, we did another thing. We did a new benchmark racing driver's subject test(for the smart car), which means it should keep the speed as well as avoid obstacles. On August 29 last year, Yutong smart vehicle IBUS was running on Zhengkai Avenue. On May 19, 2016, the President of Ukraine took our smart electric patrol car.
Finally, a summary, the wisdom of knowledge and data dual-drive comes. We used to think that it is knowledge that drives function, and now we discovered that intelligence can also learn from big data. Interactive cognition tells us how to hear, feel and see. Language is important, images are more important, and interactions cannot be ignored. Vision dominates our emotional perception, and we are developing machines with emotions. Too much learning is arbitrary, cannot guarantee the convergence of the algorithm, lack of mechanism. In this case, we propose a method, we believe that knowledge-driven artificial intelligence emphasizes the program and data. Under the dual-drive artificial intelligence, the program is important, but the data is also important.
Today, robots, through large samples and deep learning, can finally be smart enough to rewrite the parameters in the program, or even their own local code. The core of artificial intelligence is not only algorithms, but also learning. It means that we should make the full play of the mechanism fragmental data cognition, overcome the difficulties of macro cognition formalization, subdivide the constraint interval, and formalize of a large number of micro-recognition to reduce the difficulty of formalization, narrow down the scope of online reasoning, generate software for data definition, and let the program rotate around the data.
Let's talk about the vision: On our planet, Go and vehicles are inventions of humans, but today we are welcoming robots as "new humans". Go robots and wheeled robots are developing into human partners. They have wisdom, personality, the ability of behaving, and even emotions. The impact of robots on humans will far exceed the changes that computers and the Internet have made to the world over the past few decades. By 2030, the number of industrial robots owned by every million industrial workers in China will reach about 300; rural urbanization will lead to a sharp decline in Chinese farmers, and unmanned tractors, agricultural drones, backpack robots and harvesting robot will become a new generation of "farmers". Cattle retires and iron cattle plough the fields; farmers enter the city and experts farm. Nearly half of the minimally invasive surgery robots will be produced by China in large and medium hospitals. The use of all kinds of service robots will be populated in the whole society: Every family in the country will have a robot, and each senior citizen, disabled and child will own one robot in average. The history of human development is the history of human beings using tools, manufacturing tools, and inventing machines. Machines make humans stronger.
Today, humans are inventing more and more robots. Smartphones can be your loyal assistants. Wheeled robots will drive better than ordinary people. Many of the previous jobs will be replaced by intelligent robots, but at the same time more new occupations will emerge naturally, and human beings will live more dignified, elegant, and wise. Human beings are always good at better tuning and helping robots. They are good at using the advantages of robots and making up for the shortcomings of robots, or using old robots to eliminate old robots. In turn, humans can also use robots to enhance their wisdom and ability. Robots will definitely make humans smarter. All kinds of man-machine collaborative robots make it possible for us to welcome a new era when humans and robots dance together. And with elegant dance music, there is no doubt that human beings are always lead dancers! Thank you all!