20 years ago, Jack Ma, Guan Mingsheng, Cai Chongxin, Wu Jiong, Jin Jianhang and Peng Lei held dozens of sheets of paper in the office and wrote the words about values ​​on the paper onto the glass plate one by one. After several hours of careful selection Choose, they selected 9 inspirational entries from them. They didn't expect that these entries suddenly became a satire for the Internet giant 20 years later.

A scene three years ago is impressive.

At that time, Xu Li, co-founder and CEO of SenseTime, sat opposite me. Speaking of excitement, he drew a line out of thin air on the table. "Look, this line is the development of technology." Then he immediately asked back: "How do we define an era?"

About 2 seconds later, he began to explain himself: "If you look at the definition of the word "time", it is said that it is a certain period divided based on politics, economy, culture, etc. But in fact, the Iron Age and the Steam Age , the electrical age, and even the information age, the naming of all the eras we are familiar with has nothing to do with politics, economy, and culture. It is only related to technology."

Then, he drew a line out of thin air. "This line is industrial development." He said: "Technology and industrial development always go hand in hand. But when does these two lines intersect?"

Xu Li proposed a concept: "industrial red line". In the 60 years since the birth of artificial intelligence, it is difficult for AI technology to cross this industrial red line, that is, it does not substantially improve production efficiency. His view is that industrial development is often leapfrog, and that jumping node is often the technology breaks through the "industrial red line", that is, the standard for large-scale industrial use.

Three years later, SenseTime continued to make rapid progress on the road of artificial intelligence. At the just-passed 2021 World Artificial Intelligence Conference, SenseTime launched an infrastructure SenseTime AI device that can rush to the "meta universe" and realize a new interactive experience of virtual and real; SenseTime's vehicle-mounted business has also been in the industry for more than 5 years. In terms of stack capabilities accumulation, Jueying, an independent new brand that launched smart car solutions.

Li Wei, founder of Songhe Capital, once said that when SenseTime was first founded, AI was not a field of concern, but in just a few years, the changes in artificial intelligence to various industries have been fully rolled out.

However, breakthroughs from zero to one have always been accompanied by huge challenges. In addition to the superficial problems that can be seen regarding technological breakthroughs and commercial implementation, the deepest concerns about artificial intelligence come from ethical suspicion. "I'm not worried about artificial intelligence thinking like humans, I'm worried about humans thinking like computers - abandon compassion and values ​​and ignore consequences."

Some thought-provoking problems are happening. Have we reached a moment to explore the ethics of artificial intelligence? If this issue is already discussed, what are the principles and foundations of these ethics? Xu Li has his own way of thinking.

On July 29, Xu Li participated in the Fortune China 40U40 Creative Conference. Image source: Fortune Chinese

"Fortune" (Chinese version): There was an exchange three years ago. At that time, you said that artificial intelligence will usher in explosive growth if it exceeds an industrial red line. Do you think artificial intelligence has surpassed that industrial red line now?

Xu Li: We actually benefited from the dividends of the development of artificial intelligence. The first wave is what you mentioned just now. Five years ago, technological breakthroughs proved for the first time that AI can surpass the "industrial red line" in some fields, which can usually be considered to surpass human accuracy. But at that time we were also facing another dilemma, that is, when technology gradually penetrated into the industry, we would find that the production cost of artificial intelligence itself was very high.

To give a simple example, if we assume "playing Go" as a production process. We can think that Alpha Dog () has surpassed the industrial red line, that is, the human level. But if the purpose of investing in AlphaGo is only to let him play chess as his main business, its investment may be higher than the total investment of nearly 300 professional chess players in China. So what the first wave of dividends usually brings is to take the lead in adopting AI technology in certain specific industries. Large-scale applications also depend on reducing the production costs of AI itself.

The second bonus of the development of artificial intelligence is that with the continuous investment of underlying AI infrastructure, the ability of neural network technology to train general models is gradually breaking through. At present, the mass production of automated, large-scale and intensive AI models is gradually being implemented.

When Sang Tang was first founded, he began to invest in and train general big models. Simply put, general-purpose large models have the ability to understand each other. After establishing general-purpose capabilities, in the subdivided scenarios of different industries, you only need to rely on small samples to iterate into high-quality model algorithms to truly break through the human-intensive ones. Invest to achieve the satisfaction of long-tail application needs. This is like analogy to human abilities. Human perception and cognitive abilities are very common. We only need small data to understand some new problems.

"Fortune" (Chinese version): I saw you participated in an open event a while ago. You said that true disruptive innovation is not from the traditional innovation paradigm, but from the mind or thought experiment of genius' inspiration. , do you believe that the flash of genius's inspiration is accidental or inevitable?

Xu Li: First of all, many subversive scientific innovations and major breakthroughs in human history are completed through very accidental discoveries or accidental thought experiments by humans. They are mostly derived from "genius conjectures", and this kind of completion is unpredictable.

So what is the reason behind it?

Major innovations are anti-consensus, and as long as they are anti-consensus, it is impossible to plan, so it is impossible to use paradigms to regulate. It is impossible for humans to use a fixed methodology to plan something that the general public has no consensus and predict future occurrences. The reason for this phenomenon is that humans have limitations in their understanding of the unknown world.

But in today's era of artificial intelligence, machines can also make guesses and do not rely on human cognition. It is expected to help us discover the essence of scientific laws earlier and explore and discover the unknown more quickly. This is what artificial intelligence brings new paradigm for innovation.

I mentioned last time that if machines can guess Newton's law, will humans use this law? This is a problem we face. Of course, humans' conjectures to identify machines' effective guesses are themselves a very difficult problem.

I have read a novel by Liu (Liu Cixin), and a very strong agent wants to understand the fact that humans write poetry. Finally, I found that even if I could exhaust all the possibilities of poetry, I still couldn't know which poem was good, so it was also challenging to identify the pros and cons of machine conjectures in this situation without fully understanding.

"Fortune" (Chinese version): My understanding is the real difference between intelligence and wisdom.

Xu Li: For example, we have many machine speculations about future possibilities and technological breakthroughs. Some of these speculations can greatly promote technological progress in the future, but some cannot. So how to identify the validity and reliability of these conjectures will be a problem left to the future.

"Fortune" (Chinese version): So today's topic is to explore the next step in the future of artificial intelligence. In the visibility of the future, what do you think is the next step in artificial intelligence?

Xu Li: I just talked about the two major dividends of artificial intelligence in recent years. It is precisely under this understanding that SenseTime has invested nearly 10 billion yuan in total to build an AI infrastructure necessary for the large-scale production of artificial intelligence algorithms. We call it a large artificial intelligence device.

The reason for naming it a large device is to compare the physics particle collider, which is to imply that artificial intelligence helps to transform an innovative paradigm. Through the "particle collider of artificial intelligence", a huge solution space for massive data and even construction is Perform disassembly and collisions, and use certain randomness to break the boundaries of human cognition and application.

I think if some industries have many unknown fields, such as geoscience, life sciences, pharmaceuticals, and atmospheric science, machine guesses in these fields may give us some unexpected surprises.

"Fortune" (Chinese version): Is this random experiment the problem of accidentality you just mentioned?

Xu Li: To give an example, we have recently seen many breakthroughs in artificial intelligence in protein structure analysis. It may take a long time for humans to complete such a job, but machines can do it in a very short time.

The current hypothesis is that the sequence of amino acids is the only factor that determines the structure of a protein. But next we can do a lot of random impact experiments similar to particle colliders. For example, put more input factors into the experiment to see if there are any results that improve the prediction. When predictions become more accurate, we can use such results to re-understand the core elements that determine protein structure. This is to accelerate scientific research through random impacts. The results of research and development will accelerate the progress of biopharmaceuticals and other sciences.

Of course, sometimes the relevant breakthroughs may not be explained in this era. It may take a hundred years before the machine guesses Newton's law to appear to truly explain the reasons behind this.

"Fortune" (Chinese version): Are these also what SenseTime is doing? For example, the protein combination you just mentioned.

Xu Li: SenseTime AI large device provides a underlying capability platform. As mentioned earlier, as particle collider. In this process of having a certain degree of randomness, there is a lot of room for exploration. The demand for infrastructure, especially computing power, is actually an exponential growth. Data shows that in the past decade, the demand for computing power of the best artificial intelligence algorithms has increased by more than 1 million times.

This is contrary to the perception of many people. Everyone thinks that the more exquisite the algorithm, the fewer it should calculate, but in fact it is not. The better the algorithm, the more calculations are. It is actually verifying different possibilities and exploring the boundaries of applications.

"Fortune" (Chinese version): It is what you just mentioned against consensus.

Xu Li: Yes, anti-consensus.

"Fortune" (Chinese version): If we draw a little closer, the metaverse may be the hottest topic in the technology circle in the past year. What role has artificial intelligence played in the metaverse? How do you understand the metaverse?

Xu Li: I think the metaverse has always existed in various daily applications, including the games we originally said, which are actually some form of the metaverse.

However, the core reason why the metaverse is very popular recently is actually the same as the reason why artificial intelligence has been paid attention to, because it is gradually approaching the real world.

If the virtual world is very unreal, just like you play games every day, and the content you play is completely unconnected with the real world, you won't think it's a universe. Only when it is more like the real world, when everything in the real world can be reflected and connected to the metaverse and can interact accordingly, you will feel that it gives me a second possibility, thus bringing the second, The third or even more universe.

In this process, the role of artificial intelligence is to connect the real world and the virtual world, allowing users to live in different universes. In the real world, everyone has their own identity, and in the virtual world, you can also use another role to complete real-time interaction with your surroundings.

"Fortune" (Chinese version): Can the conversion between the real and virtual world be accomplished through computing power?

Xu Li: Yes. Computing power can help project the real world into the virtual world. Usually the first step is to collect relevant information, and there is nothing really semantic, and only the so-called spatial digitization is completed. However, through the breakthroughs in AI and computing power, every thing is gradually being given meanings that humans can understand, not only spatial digitization, but also factor structure and process interaction, and then correspond to the virtual world. Let people not only access, but also use, modify real-world related content and even interact.

So with regard to the real metaverse, my understanding is that people can interact with the real world in it, and they can also project various experiences in the virtual world into the real world. Otherwise, its meaning is just a game, or just an APP.

"Fortune" (Chinese version): So if you take another step, I would like to discuss a hypothesis and premise with you first. Do you think we should discuss the ethical issues of artificial intelligence? Do you agree with this assumption first on this topic?

Xu Li: I think it is worth discussing. Innovation in artificial intelligence comes from the disruption of past perceptions, so ethical governance issues in industry applications are becoming increasingly important. At the same time, one of the paths to scientific development is deduction, that is, giving a basis point and constantly pushing it to the boundary. Since there are some core principles in this ethical governance today, we can completely evolve through principles to explore what the boundaries that our ethics and machines can reach.

"Fortune" (Chinese version): You just mentioned the principle. How would you formulate this principle if artificial intelligence was given to formulate it?

Xu Li: SenseTime has done a relatively comprehensive study on the framework of artificial intelligence ethical governance around the world.

We found that most ethical governance frameworks can be classified into three categories: one is people-oriented, whether it is human fairness, identity, dignity, confirmation sovereignty, data security issues, etc., they all belong to The people-oriented category; the second category is called technological controllability, including transparent computing, interpretability, and boundaries of technical security, etc.; the third point is sustainable development, seeking green and long-term development.

I think the future ethical governance of artificial intelligence is actually to achieve balance among these three pillars.

Of course, what we have always emphasized is to uphold the ethical concept of artificial intelligence of "development". The easiest thing to meet all governance framework requirements is to break through nothing, but in this way, everyone will forget what our initial goal was. Our original intention is to use artificial intelligence to promote the development of society, which is the goal. Taking mathematical optimization as an example, our optimization function is that artificial intelligence promotes social progress. There are three constraints that need to be met below. But many times, if we only talk about constraints and ignore the goal of optimizing the function, then this optimization is unsuccessful.

"Fortune" (Chinese version): If so, do you think artificial intelligence will increase the inequality or equality between people? What is the definition of this beautiful world? How can artificial intelligence achieve this so-called beautiful world?

Xu Li: Talking about the fairness of artificial intelligence is basically about talking about the deviation of data samples that may lead to deviations of artificial intelligence algorithms, which can be improved with the evolution of technology. And I think fairness on a larger level comes from the possibility that a few people can benefit from the convenience brought by artificial intelligence, which will bring inequality.

My understanding is that as a general technology, the core factor from quantitative change to qualitative change, artificial intelligence can truly reduce the cost of its production factors. If a certain technology is particularly high-end and requires very high-cost investment to be completed, it will inevitably lead to inequality if a few people, leading enterprises, and leading scenarios can use such technology, which will inevitably lead to inequality.

Looking back at history, whether it is the Iron Age, the Steam Age, the Electrical Age, or even the Information Age, all eras are named after technology. The technologies that naming this era have a very core common point, that is, they all significantly reduced the prices of production factors at that time, thereby significantly improving production efficiency.

If artificial intelligence technology wants to bring universality, universality and fairness, it must be that it will reduce the cost of production factors of our era by hundreds of times, such as the real sense that educational resources and medical resources that we could not obtain originally, and more Better, cheaper and more equitable to the whole society.

If you are interested in this topic, please click on the Fortune Plus applet below to express your views.

END