From Pinocchio and Edward to Bumblebee and Jarvis, humans' vision and pursuit of machine intelligence have never left the screen. In real life, the most noticeable is "Robot Commander" Marc Raibert. He is a professor at the Massachusetts Institute of Technology (MIT) and founded Boston Dynamics in 1992 to develop various intelligent robots.
In 2009, Boston's BigDog (big dog) appeared on the Internet. Although it is huge, it walks steadily and steers flexibly, and its bionic state is amazing. Four years later, the humanoid robot Atlas released by it is even more famous. The original product can stretch its arms to balance through the narrow "single-plank bridge"; later, it also has inverted, flipped, aerial splits and other actions, and even parkour poses are also very good. beautiful.
Every product of Boston Dynamics can become the focus of Internet celebrities. Raiport and his company have also become an industry model that cannot be avoided when talking about AI. Therefore, at the Corporate Innovation Ecosphere Conference held in Dongguan on November 4th, Lei Boer also participated in the conference remotely. He was talking to Zhou Xi, the founder of Chinese AI (artificial intelligence) company Yuncong Technology . The latter is also the backbone of the industry that has emerged from the Chinese Academy of Sciences.
A major background of the dialogue between the two tech giants is that the current artificial intelligence is slipping from another wave of peaks: the outside world always expects AI to be omnipotent, but in reality it is like a "scrap giant".
Take Boston Dynamics as an example. Even after acquiring the company, Google, which is full of exploratory spirit, couldn't sit still watching the slow commercialization process, and finally sold and changed hands. However, domestic AI still seems to remain in the direction of voice and face recognition that are not "sexy" enough in public perception.
Indeed, after more than sixty years of development, AI technology has been greatly improved, and even the academic community is worried that AI is too powerful to bring threats and dominate mankind. However, the reality is that there are still not many areas of mature commercialization, and the impact on society is relatively shallow. Therefore, the society generally believes that AI is like a "waste wood giant", which looks very creative and destructive, but it has nothing to do except for a large amount of rice (burning) (a lot of money).
So, how can AI get rid of the embarrassment of waste wood? How to show muscles, move fists, and exert strength without ruling humanity?
The father of artificial intelligence can be traced back to Turing, who made a code breaker in World War II, and the starting point of the industry was the summer conference at Dartmouth College in 1956, the world's first AILab laboratory. In the ten years since then, the AI field has been booming: the US government invested millions of dollars, the checkers program was born and defeated human masters, text chat robots, expert systems, and robots that grab blocks according to instructions were born.
However, this wave of craze was poured cold and drenched by a report in 1973: So far, no discoveries in this field have had the significant impact promised. [1] Then the academic circle conducted a round of deep criticism and self-criticism, and AI research entered a calm stage.
In 1976, the expert system, which has been born for more than ten years, finally began to play a role in business, using the database to accumulate and participate in medical diagnosis and consultation. With the help of the expert system, the AI renaissance is rapidly unfolding. The Japanese government allocated more than US$8 billion to support research and development, and the UK spent more than 300 million pounds to build an AI project. However, after another ten years, people regret to find that the machine experts are not good enough, and AI research and development is in a trough again.
In the 1960s and 1980s, the explosion and decline of the two industries, the ecstasy and disappointment of mankind, in the final analysis are one sentence: machines are not as smart as people expected.
In hindsight, the result is obvious. In terms of hardware, the integration of transistors and chips is still at an early stage, and it is also very difficult to obtain information such as vision and touch. In terms of software, the amount of data to support the establishment of the model is still seriously insufficient, and the algorithm rule concept is not advanced enough. It can be said that it is not that the machine is not smart enough, but that the human beings are not fully prepared for their homework.
The two round trips have deeply questioned the industry, is AI in the same direction? Will the future be good? Some people even began to reflect that elephants don't play chess at all (Elephants Don't Play Chess), but elephants can make judgments and reactions based on environmental changes. The way humans set rules may be wrong.
Fortunately, scientists have not stopped exploring. The seeds sown in the cold winter began to take root and sprout under the warm wind.
Interdisciplinary research: probability theory, statistics, control theory, engineering, neurology, etc., more and more disciplines cross-border into AI research, the cross-system breaks the idea of setting rules in advance, and neural networks and deep learning technologies begin development of.
Hardware speedup: Driven by Moore's Law, chip integration has developed exponentially, and functional chips such as CPU, GPU, MPU, etc. emerge in endlessly. The birth of the cloud also allows the computing power to be expanded almost infinitely. The performance of optical lens, infrared and other sensors has also been improved, and the identification of the external environment and the capture of key information are also more efficient.
Software explosion wave: In the Internet and mobile Internet era, the amount of data generated has explosively increased, and the amount generated every year is equal to one million times the amount accumulated by humans in the past thousands of years. Real-world activities have been unprecedentedly recorded, providing sufficient models for building models. Material. In the Internet tide, the number of code farmers has also increased, and now there are about 30 million people worldwide.
Driven by these factors, the AI industry once again swaggered: "Deep Blue" defeated the chess champion, AlphaGo completely abused human Go masters, driverless cars were also driving on the road, and AI even appeared on the US ban list. The explosive "singularity" of technology seems to be coming, so that bigwigs such as Bill Gates and Hawking have dissuaded them. Don't do it too fast, it is dangerous; AI will replace humans.
The actual situation is that this round of AI craze supported by machine learning, deep learning, and reinforcement learning technologies is not fast at all, and the evolution of pure deep learning algorithms has reached a desperate situation. OpenAI's GPT-3 model is expensive to train, but its IQ cannot understand the common sense of "whether the cheese in the refrigerator will melt".
After the 2008 financial crisis, in order to avoid another global crisis, the world's top financial talents began to revise the Basel Accord, and if it was handed over to GPT-3, it would have to experience thousands of financial crises before it could learn.
Zhou Xi believes, “At this time, we need another way. We call it expert knowledge. We must believe in the power of humans and combine artificial intelligence with humans. People can create creative ideas in a very complex environment with a small sample. Decided."
In this way, human-machine collaboration has become the inevitable direction of artificial intelligence evolution, and it is also an inevitable requirement for humans to use AI reasonably.
The movie "The Matrix" shows such a picture: human beings live in a virtual world, and the AI system called the "matrix" creates and controls this world. Humans seem to be animals raised by AI, without real freedom. In fact, humans are just lines of code. However, what is even more frightening is that 99% of humans are completely unaware of the virtual world.
At the end of the movie, the hacker Neo defeated the rationality of the machine with human love, but the movie still makes the back chill, how does the human body resist the super body of AI. Although super AI is still too early, AI has begun to replace some human jobs:
For example, computers began to replace the work of crew members, responsible for scheduling flight schedules and analyzing abnormal reports; algorithms also replaced editing, and automatically pushed content to users; robots replaced search and rescue team members to work in dangerous places, replaced doctors for efficient diagnosis and treatment, and reduced errors; Radar and algorithms replace the human brain to judge vehicle distance and avoid danger faster. There is no doubt that the intelligence of AI is improving human life.
However, AI may not be a permanent long-chi energy. In 2010, a failure in the trading algorithm caused a flash crash on the New York Stock Exchange, evaporating trillions of dollars within a few minutes; in 2018, a computer failure caused serious delays or cancellations of 15,000 flights in Europe. [2] In addition to monetary losses, the overuse of AI is also impacting human civilization:
Cambridge Analytica improperly used tens of millions of user information and personal privacy was breached; while applications such as face changing brought by DeepFake were popular on the Internet, social order and public safety were greatly impacted. Even in the military field, a large number of AWS (unmanned command, autonomously find and kill targets) weapons have been developed, greatly increasing the brutality of killing.
When discussing the threat of super-intelligent AI to humans, some people will always say that it will not be enough to turn off the power at that time, but the strange thing is that the goal of AI is to overcome thousands of difficulties in order to complete the tasks assigned by humans. "Electricity" is obviously also a difficulty it believes needs to be overcome. Therefore, if AI is only set to achieve the "maximization goal", then the ultimate AI is super AI, which is to take over humanity in an all-round way.
After all, the fledgling AI cannot possess human social experience and value judgments in a short time. Based on this, artificial intelligence scholar Stuart Russell (Stuart Russell) believes that humans should devote themselves to the research and development of "Provably Beneficial AI" instead of human and superhuman AI.
Russell proposed three principles for AI: maximize the realization of human goals; maintain awe of humans; set machine preferences based on human behavior. Condensed into one sentence, a good AI should be "human-oriented human-machine collaboration": human beings are in a dominant position. Based on human experience, judgment and preferences, the development of AI that is subject to human interests is still "us". Neither do half machine half human kind, nor do human beings under the machine.
To make AI an assistant, there are a total of three steps: the first step is to achieve technological breakthroughs on multiple human-machine collaborative sensing ends, and liberate manpower from complicated work; the second step is the whole process of perception, cognition and decision-making Improve efficiency and help decision-making; the third step is to help creators in the human-computer interaction experience and enrich the content of terminal products and services.
This "three-step" strategy is logically rigorous and promising, but it is not easy to achieve. It requires every step to be counted, and every step must be taken steadily. Many domestic companies are taking the first step, and work on the perception side has become a race for precision. The second step is the first to complete the goal of human-machine collaboration faster. And this has a prerequisite: the market share has the final say.
As long as the number of sensing terminals distributed in the terminal is larger, the more data will be fed to the cognitive and decision-making algorithms, and the result of training feedback will also approach the optimal. This is also the basis for statistics and probability theory to play a role.
Many domestic companies are also doing this. Yuncong Technology, established in 2015, relies on the full development of the "industry expert + engineer" model, and Yuncong has the most extensive AI coverage for banks and airports. This ensures that when the company extends to the cognition and decision-making side, there is enough high-quality and massive data to use, which fully guarantees the second step of success.
Human-machine collaboration has become a close contact between humans and AI, and demand has become the key to AI competition and an advantage in China's AI development.
As of 2016, in the field of global deep learning, the total number of papers and citations in China ranks first, and the number of invention patents in the AI field ranks second in China; and in terms of commercialization, Megvii, which was established in 2011, was established in 2012. Pictured, Shangtang, established in 2014, and Yuncong, established in 2015, have both risen rapidly in a short period of time. They are called China’s "AI Four Dragons". The capital rushing to invest money has been ranked from China to the United States. .
These companies basically started with computer vision technology CV, so the public generally believe that the so-called AI is nothing more than face recognition, chess, essay writing, or government integration outsourcing in the name of AI. Although the commercialization of AI is slow, there is no doubt that these impressions are still outdated.
A virtuous road to AI commercialization is taking shape.
Taking Yuncong as an example, AI has realized the closed-loop technology from perception (face, human body, object, voice) to cognition (semantics, knowledge graph, big data) to decision-making (risk control, recommendation, portrait). In addition, Yuncong's "3D structured light face recognition", commercial cross-mirror tracking (Re-ID), human body 3D reconstruction and other technologies are also in the forefront of technology and commercial stage.
Human-machine collaboration has long replaced computer vision and has become the new label of this small giant. Zhou Xi made a summary: every technological advancement is an increase in efficiency and an extension of people.
The past few scientific and technological revolutions were merely extensions of human limbs. The human-machine collaboration brings a new qualitative change-the extension of the human brain. Since human thinking has no boundaries, then in the direction of human-machine collaboration, artificial intelligence should no longer be constrained to concrete objects. It is able to expand infinite boundaries like thinking.
Take the application of AI in medical treatment as an example. If there are only traditional computer vision technology and speech recognition technology, AI can only look at CT at most, and help doctors enter cases by the way. However, for those patients who are struggling with death, the biggest difficulty they face is that there is only one academician Zhong Nanshan who can rush to the front line in his old age, and there is only one doctor Tao Yong who can devote himself to selfless dedication.
Human-machine collaboration can model and automate the knowledge and skills of Zhong Nanshan and Tao Yong by putting expert knowledge in the black box of artificial intelligence technology, solve more than 90% of the diagnosis and treatment information through AI, and assist doctors to spend 10% of their energy processing Other key issues, which in turn expanded the service capabilities of medical experts by more than 10 times.
In the customs supervision system, Cloud has begun to carry out the phased practice of human-machine collaboration, and has developed customs three-dimensional supervision decision-making command system, container full-process supervision and smart inspection systems, and digitizes, structures, and models the realistic scenes of customs supervision. To help customs experts make better decisions.
Beyond the cloud, SenseTime's algorithm platform and Megvii's Internet of Things have also replaced computer vision and become their new goals. China's AI industry is relying on the vast demand market and embarking on the world stage.
Regardless of whether it is a bank risk control expert or the chief doctor of a tertiary hospital, it is an extremely scarce resource in China, but the people's demand for high-quality services is real. Human-machine collaboration can not only benefit the people of high-quality products and services, but also avoid conflicts between human-machine positions.
For a populous country like us, the development of inclusive AI with human-machine collaboration and extending our wisdom is the trend of industrial development and an inevitable choice for the integration of engineer dividends and the needs of the masses.
Although firearms and cannons have been used in China for a long time, the artillery of the Qing Dynasty "only knows that the barrel is made of iron, and it is not scientifically measured, so it cannot be used with accurate head". [3] The inside and outside of the artillery are uneven. The largest depression, you can pour 4 bowls of water without overflowing. In the end, in the battle with the great powers, the Qing army was defeated and humiliated.
It is true that in today's world, it is no longer easy to break out large-scale wars, but it is undeniable that the competition and even confrontation between countries have never stopped. Science and technology, both in the Qing Dynasty and today, are the fundamental guarantee for self-reliance. In the field of AI, after several groups of people's efforts, China has been qualified to stand at the same stage as overseas companies. This is a precious achievement.
Ren Zhengfei once said that only by attaching importance to basic research for a long time can industry be strong, and artificial intelligence is the core variable that influences and shapes a country.
As China with a population of 1.4 billion, we need to improve efficiency to create more wealth, and we also need to focus on equality and common prosperity. Perhaps AI may seem like “waste wood”, but we cannot ignore the existence of “giants”. Tolerate innovation and tolerate those who explore, our basic education and our technologically powerful country will also be guaranteed.
Reference
[1] "LighthillReport", Science Research Council of Great Britain, 1973
[2] "AI Freshman", Stuart Russell (Stuart Russell), CITIC Press
[3] "Complete Collection of Chinese History", Xiao Feng, Liaohai Publishing House
