Frontline | Xu Li, CEO of SenseTime: Advocates that the industry society should uphold the "development" concept of artificial intelligence ethics governance

On July 8, the three-day World Artificial Intelligence Conference 2021 (WAIC) kicked off at the Shanghai World Expo Complex.

Last year, WAIC took online as the main position, and after returning to the offline exhibition this year, WAIC is still not inferior to the past in the guest lineup. Guests who delivered speeches on the opening ceremony day included Baidu founder Robin Li, Chairman and President dong Mingzhu of Gree Electric Appliances, Hu Houkun, rotating chairman of Huawei, Xu Li, co-founder and CEO of SenseTime, and other industry celebrities.

In addition, Ma Huateng, Chairman and CEO of Tencent, Shen Nanpeng, Global Managing Partner of Sequoia Capital, and Sun Zhengyi, Representative of Software Banking Group, participated in the opening ceremony agenda in the form of video/connection.

Xu Li, co-founder and CEO of SenseTime, delivered a keynote speech entitled "New Paradigm of Artificial Intelligence Innovation - The Conjecture of Machines", in which he said: Breakthroughs in artificial intelligence will be able to regularly expand the boundaries of human cognition. He pointed out that whether it is the induction of past experience, reasoning and deduction, or the current use of big data and simulation, the four scientific research innovation paradigms in the usual sense have established patterns, but many subversive scientific innovations and major breakthroughs in human history are anti-consensus and unpredictable, and mostly stem from "genius conjectures", such as the apple that fell on Newton's head and the light that Einstein fantasized about riding in his childhood.

Li Xu, co-founder and CEO of SenseTime

"The reason for this phenomenon lies in the limitations of human cognition of the unknown world. But in today's era of artificial intelligence, machines can also make conjectures, which is expected to help us discover the essence of scientific laws earlier, explore and discover the unknown faster, which is the new paradigm of innovation brought about by artificial intelligence. Xu Li said.

With the development of general artificial intelligence technology, AI algorithms have not completely relied on labeled big data, but the exploration of huge unknown possibilities must rely on large computing power. Xu Li said that SenseCore AI device built by SenseTime provides a strong boost to "machine guessing". Basic disciplines such as life sciences, materials science, and medicine will be the first to benefit from "machine conjecture" and enter a new paradigm of scientific innovation.

Taking the development of automatic driving as an example, from unmanned parking, to unmanned connection in closed parks, to automatic driving on open roads, it is a process of defining the boundaries and specifications of technology through the application of data, so as to gradually achieve the controllable development of technology and better promote the progress of the industry.

Innovation in artificial intelligence comes from the subversion of past perceptions, so the issue of ethical governance in industry applications is becoming more and more important. "People-oriented, technologically controllable and sustainable" is a widely agreed framework for the governance of technology ethics in the world, but at different stages of development, a balanced development of the three is required.

In this regard, Xu Li advocates that the industry society should uphold the "development" concept of artificial intelligence ethical governance, one is that when talking about ethical governance, we must consider the goal of inclusive development, that is, using artificial intelligence technology to promote social progress; the other is to consider the rapidity of industry change while landing advanced technology, and find out the development balance under different governance frameworks for different stages of development.

Xu Li finally said that the "machine conjecture" plays the role of the apple that fell on Newton's head, and Shanghai's inclusive, open and innovative temperament has given this apple the best soil, helping it to take root, grow apple trees, promote innovation in various industries, and develop inclusive and sustainable artificial intelligence.

This year's WAIC exhibition set up a "1+6" section, and the "AI@ Shanghai" theme exhibition area, innovative technology new momentum, intelligent economy new vitality, a new picture of a better life, collaborative governance new development, a new ecological future, a harmonious humanities new chapter 6 sections.

Among the exhibitors, this year's offline exhibitors are expected to exceed 300, of which the proportion of first-time exhibitors exceeds 40%, and enterprises outside Shanghai and foreign enterprises account for more than 50%.

<h3>The following is the full text of SenseTime co-founder and CEO Xu Li's speech at WAIC 2021:</h3>

Honorable Secretary Li Qiang, Minister Xiao, Mayor Gong Zheng, leaders, distinguished guests, good morning.

It is a great honor to stand on this podium and share a little of our thoughts on the paradigm of artificial intelligence innovation. SenseTime was very fortunate to witness the development of WAIC over the past 4 years with you, and I looked through my photo album and found that I took photos every year, and then every year I took pictures of the theme of the conference. From the new era of artificial intelligence empowerment in 2018, to the infinite possibilities of artificial intelligence we discuss in 2019, and then to the home and city in 2020 and 2021, artificial intelligence seems to be getting closer and closer to life.

In fact, artificial intelligence conferences are held all over the world, and there are many places in the country, but if we look at the popularity of search engines, we will find that all the peaks are during our WAIC conference, that is to say, WAIC has invisibly become the most important participant, witness and recorder in the history of artificial intelligence.

Then talk about innovation. In fact, I think artificial intelligence is an innovative park. We have been talking about the paradigm of innovation in the past, from inductive summary to deductive reasoning, when there is a computer, deductive reasoning can be simulated by a computer, or inductive summary can be used by a computer to do big data-driven algorithms, which is what we usually call innovation in the four ways. But the real disruptive innovations don't come from here, they come from the brain holes of genius, or the conjectures of geniuses, from "thought experiments."

So today I am thinking about this matter, that is, with artificial intelligence and supercomputers, can we ask the question "Does the machine think" asked by Alan Turing in 1950, then our proposition is simpler, can we ask "will the machine guess"? I think the answer is yes, because in some areas, machines have given us a very good model for humans, making it push us forward in human science in reverse.

What are the necessary conditions for machine conjecture?

First of all, with the development of computer computing power in the past 20 years, we can see that the demand for computing power by the best artificial intelligence algorithms has increased by nearly 1 million times in the past 10 years, which is a huge breakthrough. It is reasonable to say that the more advanced the algorithm, the less computing power will be required, but on the contrary, the larger the vast unknown space we explore, so we need greater computing power, which has nothing to do with the data. Because there are many algorithms that already rely on small data, or even do not need human data, but this exploration gives us a possibility to update iterative cognition.

That's why I say that the computing power device of artificial intelligence is a basic element that drives conjecture. Why is it called a large device? I want to like the particle collider, which is to create new particles in a random possibility to explore the unknown world.

Second, since we are all guessing, many things are uncertain, so when we really define the boundaries of industrial applications, or when we define it unreliable, we need to put them into the industry. In fact, many applications in our human history are to say that there are conjecture applications first, and finally we can give everyone a suitable explanation.

For example, the airplane, the Wright brothers certainly did not know the principle when they invented the airplane, and even to this day we can not use fluid mechanics to explain the power of the aircraft to take off, but this does not prevent the aircraft manufacturing company from creating a safe and controllable aircraft. In fact, a large part of the many artificial intelligence applications that we are familiar with are based on such conjectures.

Let's look at a few simple examples, a very small scene in automatic driving is called automatic parking, and old drivers can often summarize the rules of parking. However, these laws can not be converted to the computer in time, the computer is a set of laws evolved by itself, and even our scientists have no way to really explain the logic of computer reversing, but this does not prevent our automatic parking from landing on the scene in many places.

Promoting the driverless cars we are talking about so far, I see a lot of driverless cars outside the field, and our own driverless AR minibuses also began to take a ride yesterday, and to show our augmented reality. The first thing passengers ask when they get on the bus is, how many kilometers can you walk in this car without being hosted? I feel very relieved. It is precisely because we are cautious to gradually open up the application scenario, so our understanding of the unmanned driving scenario has been popularized to the general public, and in the future, we will embrace more of such a scene iteration, we can truly explore the possible boundaries of technology.

Another more complex application is called decision-making in an open environment, such as our game, when the reversing thing may still be guided by someone, then the spatial complexity of the game is much higher than that of reversing, or even Go. In this process, in fact, it is difficult for humans to give a standard answer, which is like the application of intelligent transportation that we talked about later.

Usually, when we solve the traffic light control, it has been difficult for humans to give a perfect answer. We are trying to solve this problem with artificial intelligence speculation in a very small range, which can save the average local waiting time by half, close to 20 minutes, so whether it can be generalized to a larger range is still waiting for us to do more tests. But the conjecture of the machine in this direction has given us a surprise.

Another type of application, I believe everyone may take for granted, is a big data-driven application, including the construction of the smart city two networks that we did in Shanghai, including our smart community. However, when we really enter these applications, we find that it is not so big data, and we have to solve the daily long-tail low-frequency, just-needed applications.

For example, when we solve the falling objects, it is impossible to have too much data on such falling objects; when we solve the problem of fire, we solve the problem of community care of the elderly falling, it is often zero data and small data, then the machine can only make extensions and generalizations through general technology to make possible conjectures about these scenarios.

Since machines in many cases, we can not have a complete control and complete iteration at present, so SenseTime also adheres to the key to promoting the development of better AI ethics, and has launched our ethical development concept based on controllable, green, sustainable and people-oriented artificial intelligence.

What I want to emphasize here is that we must look at the ethical governance of artificial intelligence from the perspective of development. Development has two meanings, first, we must take development as the goal, all ethical governance is to make artificial intelligence better serve our society, better promote the development of society. If you don't talk about goals, it will be very easy to meet any governance conditions, but a single optimization will often make everyone get into trouble.

Second, the development of artificial intelligence is very fast, so we must use the vision of development to balance a variety of different governance frameworks, at different stages to choose different governance policies, just like Secretary Li Qiang just said, we want to promote agile governance, from the completion of such a task.

A negative example is that when the First Car was built in Britain in 1865, the United Kingdom introduced the Red Flag Law, and a person waved a red flag 50 yards in front of the car, so the speed of the car could not exceed that of people. Of course, we have achieved complete safety and control, but the UK has also lost the dividend time of car development.

So today, I think the conjecture of artificial intelligence, the conjecture of the machine, may be the apple that fell on Newton's head, and Shanghai's inclusiveness and openness is its best soil, enabling it to find more innovative results. We may use the innovation of artificial intelligence to promote the inclusiveness of artificial intelligence, so that artificial intelligence can affect many industries.

Thank you.

The presentation on Inclusive Ai has come to an end, but I believe that the discussion on Pratt & Whitney AI has just begun, and then we have an important guest, Mr. Shen Nanpeng, Global Managing Partner of Sequoia Capital and Managing Partner of Sequoia Capital China Fund.