Content source:2019 4 18 Day, Lu Qi became the first time after the founder of YC China, and chose to teach Tsinghua University to teach Tsinghua students the "Innovation and Entrepreneurship Wave in the Age of Artificial Intelligence", knowing that the exclusive production line "YC Lu Qi: AI technology The core essence and innovation "private class, with the consent of YC, authorized the note man to release the essence text version of the lecture for the first time.
Lu Qi:Mr. Lu Qi officially joined Y Combinator (hereinafter referred to as YC) in 2018 8, served as the founder and CEO of YC China, and served as the Dean of YC Global Research Institute, responsible for YC's business and strategic development in the Chinese market. YC China 2019 Autumn Venture Camp is now fully open for recruitment.
Speaker | Lu Qi
Cover Design & Chief Editor | Lili
Source | Notesman (ID: Notesman)
editor:
2019 4 Month 18 Day, Dr. Lu Qi’s private speech in Tsinghua’s non-public speech “The Nature of AI and Entrepreneurial Innovation” was exclusively available. This is the first time Lu Qi systematically exported his insights and practical experience in the AI industry in China. . In the lecture, Lu Qi led entrepreneurs to re-recognize AI at the height of history and gave directional suggestions for personal development in the AI era.
In his view, the core driving force of world change is knowledge. The progress of human history stems from the improvement of knowledge level. Machine systems such as AI help humans improve the efficiency of acquiring knowledge, and use knowledge to reconstruct any industry and realize business innovation. .
Below, enjoy~
The theme shared today is “The Age of Artificial Intelligence, The Wave of Innovation and Entrepreneurship”.
Artificial intelligence, no doubt, its depth and breadth of social impact will be unprecedented. So, in terms of innovation and entrepreneurship, how can we better seize the opportunity of a historic wave like artificial intelligence?
Today, what I want to share with you is some of my personal experiences in doing technology and products in the past few decades.
How to control the opportunities in the era of artificial intelligence? The overall idea has two points, first: standing higher, second: standing farther.
How to stand taller and further? One approach understands the core nature of artificial intelligence technology deeper, especially the nature of structure, and uses it to reason about future opportunities for innovation.
We look back at 60's digital industry history over the years and see its structural factors, using this approach to predict the structure of innovation opportunities in the artificial intelligence era.
So, first of all, I want to share the core structure and trends of artificial intelligence technology with you, and then use this as a basis to infer the innovation opportunities in the AI era.
At the same time, share my experience and tell you how to prepare, explore and participate to seize such an opportunity.
The essence of AI:Generic ability to produce and apply knowledge to accomplish tasks
1. What is the real core of AI technology?
Today's artificial intelligence technology based on deep learning, stripping the appearance of the display form to look at its core, is essentially a new form of calculation, and its underlying substrate is based on distributed overlapping vectors, in this way The vector space of is used as the feature expression space of any model.
In the traditional sense, all feature engineering used in machine learning, if mapped into the space of such an overlapping vector, can quickly learn the expression of the automatically expressed feature through the expression of the differentiable function.
This point is for students who have a background in computer science and have a technical background. I want to emphasize it a little bit. It is also an important experience that I personally have been doing technology for many years.
That is, if we can automatically learn feature representations, and these feature representations can be used to solve different tasks, it is knowledge itself.
We may see that knowledge is too humanized. We must use natural language. Do we have to use maps to express knowledge? Overlapping vectors are knowledgeable if they can solve multiple tasks very efficiently.
This is an important point. It doesn't matter if there is no technical background. You have to take away. The biggest point you must get is that we have found a new way of computing. It has different underlying layers, different substances, and its core is available from a large number of Get knowledge quickly in the data.
With this, we have an opportunity that we didn't have before.
We can use electronic, electrical and mechanical technologies to create a new generation of intelligent systems through the engineering methods available today.
But before we start a huge future, we need to talk about the intelligent system. What is its core feature and what kind of system is a typical intelligent system? Understanding this helps to predict its future and it will bring The change.
2. The structure of the intelligent system
The best intelligent system is a biological system-human, any system, biological system, mechanical system, if it is intelligent, it must have these three components in structure, the intrinsic architectural trinity (trinity):
① Perception system
It has a perception system that must have an ability to perceive the environment. By observing, we call it an observation system. Through the observation of the environment, the data is used to express it - the data represents a digital medium to record the perception of the environment. It is itself a carrier of knowledge.
② Intelligence system
In the middle of the above picture (picture) is the intellectual system, the thinking system.It uses memory and induction to acquire knowledge. There are only these two behaviors. Memorization& Generalization there is nothing else. Memorization & Generalization there is nothing else. Through these two we can acquire any knowledge.
③ Activity system
Through the second system intelligence system, we can plan for the target, and finally use the third system: the activity system to control and interact with our environment to achieve the goal.
Any system with intelligent capabilities, biological or mechanical, there must be such a structure. The important thing is that with the first new calculation method, we can use the engineering technology that we can use today. To manufacture, artificially create such an intelligent system. This is the so-called AI system today.
This is a very important point. With these two definitions, it is natural that we can define the nature of AI as a universal ability to acquire knowledge and use knowledge to achieve goals.
Let me summarize, two points have been mentioned before:
First, the core of artificial intelligence technology
Second, looking at history, the structural factors allow us to predict what kind of scope in the artificial intelligence era will be digitized.
After having the above basic understanding, Dr. Lu extracted the law of AI as a general history of technological development.
Changes brought by AI technology
Change one: from the giraffe to the intelligent system that will learn
It should be emphasized that this is the first time in our human history that we have the real ability to build a non-living intelligent system.
In the past, 60, the so-called information industry, has established all the software industry systems. If you use a natural biological world as a metaphor, it is basically an animal like a giraffe.
Why is a giraffe? Within two hours of the birth of the giraffe, basically all the skills it needs to survive are available. It can run and eat leaves, but it can't learn anything in life. It has no ability to learn.
Most of our software systems today are artificial giraffes. They don't have the ability to learn. All of their abilities are born to it. It is the knowledge that people give it. We are all written in the software code.
We started to build a human-like system today, and of course we are still far from human ability, but we have already started.
People are very different. It may be almost useless to be born to 10, but it is a powerful learning machine, Amazinglearning machine. How do we learn?
Through observation, through thinking, through interaction with the environment. We grow up not because of our parents, nor because God is writing code into our brains.
The reason why we become smart is because we observe, think and interact with the environment, so artificial intelligence has this kind of autonomous learning ability, which is very important to human significance.
This is the first time in our history to have the ability to build such a system. Once we have improved the power of the algorithm in the next 5 year 10 year, we can build an intelligent system, and the accumulation of knowledge will become stronger and stronger.
Therefore, first, the significance of artificial intelligence technology to human beings is very important. This is the first time in human history that it has the ability to build such a system.
Once the algorithmic capabilities we can build with the next five years of algorithmic capabilities, the accumulation of knowledge will become stronger and stronger.
Second, artificial intelligence is also very important in history. So far, the progress of human history has been the improvement of human knowledge, the new ability of people, the inventing of new methods, and the promotion of social progress.
From now on, the machine systems established by people and people jointly invented knowledge. In the future, these systems will have far more knowledge of human beings than humans, so it is also a very important point in history.
This is the difference in the core structure brought about by artificial intelligence technology.
Change 2: Creating knowledge from people to creating knowledge with people and AI
The creation and use of knowledge is also very, very important in history. So far, the progress of human history has been the improvement of human knowledge, the new ability of human invention, and the new methods of invention to promote social progress.
From now on, the machine systems established by people and people jointly invented knowledge. In the future, the knowledge gained by these systems will far exceed the knowledge that people can acquire. This is a very important point in history.
If the change is the change in the core structure of the intelligent system brought about by artificial intelligence technology, then the second change is the change in the time dimension of the history of technological development.
From a historical perspective, artificial intelligence technology is an important step in the long process of human digital technology, and it is also an inevitable step.
In the past, 60's history of information technology has been a history of human digitization, and the scale and scope of digitization has become larger and larger, and the digitization process has become faster and faster.
Why does digitalization bring such a big change? Because once we digitize, we extract information, the ability to acquire knowledge is greatly improved, and the speed of innovation will continue to increase in any field that is digitized.
Because the entire social economy is increasingly driven by knowledge, digitalization allows us to gain the speed of knowledge, and the ability to innovate is greatly enhanced.
The history of the entire IT industry is the acceleration of the digital process. Basically, there is a new generation of technology every 12 years, and there will be factors such as defining experience, which drive the speed acceleration and widening of the digitalization. In short, there are basically some structured rules that follow the rules.
The changes brought by artificial intelligence certainly do not stop at these two points. These two points are for the non-professional, non-technical students to understand the two key points of this change: one is that the core structure of the intelligent system has undergone fundamental changes, the machine can Learn independently as a person and invent knowledge independently;
One is that once the machine can invent knowledge on a large scale, all the technological, social, and economic fields will be rewritten, Redefine everything.
What are the important development directions of AI?
From the historical summary, and then look forward to the era of artificial intelligence, what are the important development directions?
First, the platform for digital technology.
The first driving factors are the front end and the back end. The front end is mainly the coverage and interaction benefits of human-computer interaction.
Early on, it was a mouse keyboard plus an image display. Apple has taken a big step in this regard: finger (touch).
Once you have a finger (touch), you can make this interaction reach more people, because the mouse and keyboard plus the display allows you to basically sit in the office, but once you have a finger, you can accompany each person's future interaction. More extensive, natural interaction, no need to learn, everyone can use.
Therefore, interaction is always a driving factor.
The back end is the size of the computing resource storage computing bandwidth, how much information can be processed, how much knowledge can be carried, including the coverage of the space.
For example, a distributed database can cover a single enterprise, and the global Internet covers the entire world.
It is precisely because of the global Internet that the world has become flat. The industry chain of the whole world has undergone major reforms in 20 for many years. It is precisely because of the coverage of data that the digital process must be promoted by a commercialized ecology.
Second, a defining experience.
Every digital platform business ecosystem has such features. First, it has a defining experience.
iPhone:
When the first iPhone is in your hands, you know that the new era is coming, because the experience it brings to you is totally different.
Windows 95:
When you first get Windows 95 (the first really good PC in history), you can also feel a new era.
The first time you look at the browser, you know that the future is completely linked together, knowing what a digital society looks like.
Therefore, a defined experience is very important.
At the same time, there must be a wide range of scenes, narrow scenes are not enough, and the office is wide enough. Search, e-commerce, social, you must have a wide scene, and at the same time, have an effective business model, as long as these elements are in place, the platform can be supported.
Third, the progress of digitization can be accelerated.
Let's summarize the past history. At the beginning of digitalization, IBM digitized the desktop, and then Microsoft, which digitized the information management of the entire enterprise. The world is already a trillions of industries.
Next up is Apple, which digitizes people's social behavior, e-commerce, and so on.
In the era of artificial intelligence, the nature of digitization has begun to change, and the scope can be expanded to almost all places in the physical world. Why? Because in the era of artificial intelligence, through the sensor, through the perception system, thinking system and activity system, the physical world and the digital world are completely integrated.
Because of this technology, each physical entity will be a front end, and we can interact. The digital process probably has the first smart objects we can predict:
1 Through the dialogue assistant, each object can interact, both front ends.
2 Autonomous system. Ability to automatically drive vehicles and robots.
3 Smart place. New retails like Amazon's shopping will increasingly bring smart space and time.
At the same time, the back end will be a smart cloud, including large-scale data and knowledge, just like a human brain.
Therefore, through history, we can basically predict that the front end of the artificial intelligence era has at least these contents, and its back end is based on knowledge.
With this foundation in place, let us look at what kind of innovation opportunities can it bring in the AI era?
What kind of innovation opportunities can the AI era bring?
Before answering this question, let's look back at history. In the history of mankind, our economic behavior and living standards have hardly changed in the past 10,000 years, and it has only begun to change more than three hundred years ago.
A closer look at why the core cause of the acceleration of human social activities and living standards is the invention of universal capabilities.
Historically, we invented some common abilities early on, such as theification and domestication of plants and animals, inventing texts, etc., but only three hundred years ago, we began to invent more and more energy. Strong general capabilities, including steam engines, trains, computers, airplanes, etc.
But artificial intelligence, which is the most common of these inventions, because knowledge can be used in any application, knowledge is the most versatile, and the energy of knowledge is the greatest. Bacon once said that knowledge is power.
Knowledge is essentially the potential energy. With knowledge, you can do automation, you can make predictions with knowledge, and you can generate new experiences with knowledge.
So artificial intelligence brings us a universal ability that has never been seen before. According to this history, what will happen next is a huge change. The core driver of this change is knowledge.
If artificial intelligence is used to drive business innovation, it will certainly follow such a rule:
First of all, in order to create value, artificial intelligence must have application scenarios. After this scenario, we can find any scene, preferably a wide end-to-end scene, so that artificial intelligence will increase the efficiency to a greater extent.
First of all, the first question to ask, "What kind of knowledge do we want, we want to know", to increase the value of this scene.
Once you get from this starting point, the question you will answer right away is, "What kind of data do we have?" Data is not created out of nothing. Data is always generated from observations of the environment. If you need data, you must have sensors.
This is also true in human history. If there is no telescope in history, no microscope, we can't understand the physical world at all.
So today's artificial intelligence innovation, you can see that the core reason why startups do a lot of sensors is that you must start here.
To have knowledge, you must observe that you must have data. With the data, we can extract knowledge and extract knowledge.
Including software + hardware + algorithms, with this knowledge, we can decide, if we know this, we want to solve this problem and improve this value.
We apply this knowledge to achieve our goals. After achieving our goals, I create social and business value, and there will be more data at that time, creating such a closed loop.
Here I emphasize that data must be worth living data, and one-off data is not used at all.
So there must be a business ecosystem, and the data will be truly valuable, because the application is constantly changing and the technology is constantly changing.
Application drift and platform drift, the data must be alive, you will know if you have been doing products for many years. So there must be such a closed loop, this model is a must for any artificial intelligence to create value.
With such a structured approach, we will look at the overall situation and make two changes to the changes and innovation opportunities brought about by the artificial intelligence era:
First, we will have a new information industry.
First of all, as mentioned above, the future calculations must be ahead. The front is the observation system. There are many new sensors, especially optical sensors, not just sensors.
The sensor must have the ability to calculate, called on-sensor silicon software models, which I think is one of the current innovation frontiers.
At the same time, the entire chip must be rebuilt from the bottom to the top. Today's X86 ARM system architecture does not work at all. It is assumed that Von Neumann, assuming that your data dimensions are relatively low, your ControlForce can predict, your Cache The Hit Ratio must be high enough.
In the era of artificial intelligence, there are actually no such features at all. They are high-dimensional data and must be processed in large scale.
So the entire silicon wafer industry will certainly be rewritten, the instruction set can play a role, but the main silicon wafers areASIC (application specific integrated circuit) orFPGA This team. Because large-scale training and reasoning are applications of silicon wafers that primarily generate commercial value. So the entire chip industry team will be rewritten.
At the same time, software, from the underlying fabric to the middle middleware, to the operating system, to the application development, to the tools will be re-established, so if you are in the IT industry, there are many opportunities for innovation, and large enterprises are facing challenges.
Second, a series of new pillar-type new industries will be born, and industries that have not been used before. At least will include:
① Conversational intelligent objects based on personal assistants will be a world that can awaken everything in the future;
② Autonomous systems, autonomous driving, robots, new mobility, this is the movement of the real physical world;
③ Smart places, any physical space, originated from retail, because it has a high value, but the rules of the game will be the same. Once intelligent, commercial and social values can be increased on a large scale, so these are new in the future Pillar industry.
At the same time, any existing industry, entertainment industry, manufacturing industry, finance, medical care, education, retail, all industries can use artificial intelligence technology to enhance and transform.
Because it brings knowledge, you can use knowledge to transform, you can rebuild, you can improve.
Any human industry, lawyers, doctors, teachers, analysts, you will work differently. Repeated work, mechanized work will gradually be gradually replaced.
We can spend more time. Do what we are better at, creative and imaginative (work).
Everything will be changed. This is a complete, all-inclusive change. It is not an exaggeration because it is an opportunity brought about by the core of artificial intelligence technology.
How to accelerate the entire innovation process?
In such a large context, how do we come to a structured development to accelerate the process of innovation?
Here are a few structured factors to share with you:
1. Capital.
① Financial capital.
There is a large-scale commercial value in any era. When it comes to production, it has financial capital and other production capital. The so-called functional separation, he will go further.
In a sense, market greed will go very fast, so we will see a larger capital investment. Because there are too many opportunities in the next 30 40 year 50 year, there will be a large amount of capital invested in this industry.
② Talent capital.
The university was born in the era of large-scale industrialization. Professional training, because the industrial age requires skills, tailors, designers, chefs, large-scale training of various skills.
But in the era of artificial intelligence, people's skills are suitable for the times to innovate and invent new methods. Because mechanized repetitive actions are gradually replaced. But talent capital will also have a new opportunity to move forward.
③ Data capital.
It is the most important and special. Because data is a major production capital in the era of artificial intelligence, the so-called Primary Means of Production.
for example:
a. What is the core production capital of the agricultural era? land.
Because agriculture is essentially photosynthesis, crops can only be grown if enough land has sufficient temperature and sufficient sunlight.
Human beings are smarter, more people, and useless. They give you the most intelligent people in the world with tens of millions of people. But with so much land, you can only produce so much.
Data plays a similar role in the era of artificial intelligence.
b. Some students may develop technology. You have to do something related to natural language and use voice to make conversations.
If you don't have the 2 10,000-hour annotation data, you can give it to all the engineers in the world. You can't do it.
This is not a human capital issue. Because data is the carrier of knowledge, without such knowledge, there is no way to engage in dialogue, so data becomes a major productive capital.
If we want to promote innovation in the era of artificial intelligence, we must pay attention to these three capitals, financial capital, human capital and data capital.
2. environment.
In particular, the government's policy environment. The protection of data security and privacy is the core key to obtaining capital value from data.
At the same time, infrastructure construction, as I mentioned earlier, a new pillar industry like driverless, must rebuild all road networks, road infrastructure, and with the driverless, the radius of the city will increase, this is A rule of history.
The more developed the traffic, the larger the city, the larger the size of the city in the future, and the need for new infrastructure.
3. A market-oriented path.
If we want to accelerate the innovation of artificial intelligence and let it create more social value, I personally think that market orientation is the most critical core path.
The market is very magical. In a sense, the market is probably one of the most important inventions of mankind.
What is the nature of the market?
① Converter.
The market allows each of us to participate in the market for our own will, personal private motivation.
But the result of our participation in the market is to create services and experiences and products that are valuable to others. It is such a converter, everyone is going to participate in the market for their own sake, but the result is to bring benefits to others.
② Optimizer.
Because the market reacts at high speed and even reacts in real time, any good idea, a viable technology, and a high-value scenario, the market must have a signal.
If there is no value, it should be filtered quickly. So the market is an efficient optimizer, the bigger the market, the better the optimization benefits.
If we want to quickly make artificial intelligence generate commercial value and social value, passing the market is the most effective way. If you don't need a market, how do you know that this is right? Why is this useful? No one in the market can force you to participate. Its characteristic is that it is a very good filter and an accelerator. Therefore, we must pay attention to capital, environment and market orientation, and do appropriate work to promote innovation in the era of artificial intelligence.
AI era entrepreneurship
We all know that entrepreneurship is the cradle of innovation. The whole industry, the whole society, any living, high-value industry and enterprise, its birth is a business process.
Therefore, in a very special era of artificial intelligence, a healthy and prosperous early entrepreneurial ecology is vital to any country, any region, and the world.
Large companies will also innovate. There is no doubt that the era of artificial intelligence has brought many opportunities to many large enterprises, but large enterprises have the challenges of large enterprises, but the ultimate challenge is often culture and mechanism.
The new enterprise starts from scratch, without any burden and speed, so the overall early entrepreneurial ecology is very important in the era of artificial intelligence.
How can we create a good environment to build such a prosperous ecosystem? There are several elements:
1. Talent.
The entire economy needs more and more high-quality, highly enthusiastic new-type people to join the entrepreneurial industry. In China, Tsinghua is a very top school and the entrepreneurial atmosphere is very strong.
Therefore, we need to create an environment in which more and more high-quality entrepreneurial talents are involved.
2. Data resources.
As I mentioned earlier, if there is not enough data, there are more people and no use, which is determined by his artificial intelligence technology and the rules that create commercial value.
3. Some core AI technology capabilities.
These include software chip algorithms, which are all important. At the same time, investing in the AI era, especially with AI to completely transform, to increase the scale of traditional industry capital needs a lot, capital time needs long-term investment.
The ecological investment in the early days is a decade of return, not enough, so there is a lot of innovation and exploration in this area.
The size of capital and the length of capital must be properly adjusted to fully exploit the characteristics of creating value in the era of artificial intelligence.
Includes funds like the Softbank Vision Fund. I personally think that this is just the beginning. In the future, there will be more and more large-scale funds going to Sun Zhengyi for such a large scale and investing for a long time. He must create value in this way.
So capital also needs to be reformed. Of course, the more market scenarios, the larger the scale and the wider the iteration. This is a very good time in China, stronger than any other region in the world.
At the same time, talent and other highly intensive resources will play an increasing role. Everyone knows that in Silicon Valley, it is such a typical example.
Because of the need for more intensive data integration, coverage, iteration, and talent are all together. These cities in Beijing, Shanghai and Shenzhen are ideal.
At the same time, we also need more top-ranking universities, and we must cultivate hacker culture in the long run. The core of hacker culture is to do it by hand and solve problems through creative methods.
Don't just look at the theory, just take notes, just think about the problem, think about it and do it. It doesn't matter if you fail.
Because you can learn more formal knowledge from failure, it is very important that hacker culture should be further integrated in Chinese universities.
At the same time, the atmosphere of entrepreneurship, especially in first-class universities, is very strong in the United States like Stanford University and MIT.
Tsinghua is very good in China, and we must continue to advance in this way. Only then can we build a very prosperous early entrepreneurial ecology.
What are the entrepreneurial challenges in the AI era?
Because in the software age, in the Internet age, a startup company team is basically on the same starting line as a big company team. Why? Because the software is open source. I personally wrote a lot of various codes.
Linux, FreeBSD, Mexico, MongoDB, Node.JS. You basically don't want to write a lot of code, you just need to bring it, and your code is for your product, there will be a small product in three or four months.
This is not the case in the era of artificial intelligence. Because of the artificial intelligence era, we will write code, but these codes are mainly from the data to learn the model, extract knowledge.
Data You have to do a natural language interaction, which may require you to spend tens of thousands of hours tagging data. The first is money. The second is time.
If your business is based on computer vision, to identify all products, fruits, fresh, the data, labeling you need, the cost is very high.
Therefore, we need the whole industry together, through everyone's efforts, lower the threshold of innovation, so that any entrepreneurial team can quickly try, can quickly make products, and then explore the market.
Thank you!
This article is transferred from the public number note man (ID: Notesman)
Comments