Some people think that AI is too far away from us, it is too early to talk about it; some people think that the AI ​​era has arrived, and we should seize the dividends of the era. So, is AI in the spring or winter?
Instructor: Liu WeiCEO of BV Baidu Venture Capital

Let's first look at a few examples of AI application scenarios.

Convenience stores operate very efficiently. A small convenience store, its daily operation and maintenance are complicated processes. In the past 20 years, the efficiency of this industry has been continuously improved. Now it seems that it has been achieved to the extreme. However, in the first two years, many people started to use AI technology, that is, visual recognition to do convenience stores, the final effect is not good, why? Because the replacement cost is very high, a visual checkout system replaces a cashier. This system requires more expensive IT engineers to maintain it regularly. Once the system fails, the loss will be greater.In a word, on the basis of the original convenience store, the use of AI technology has not improved the efficiency of the industry.
The second example is a wheeled robot that everyone sees at the bank. It will take the initiative to say hello and be able to have some simple conversations with you. But looking at it now, such a robot is not useful because it cannot replace bank staff. Bank employees have received good training and have a lot of flexibility in communicating with customers, handling conflicts, and asking for help, which are not possible with robots. Jokingly, when it snows, the account manager can go out to clear the snow. If someone robs the bank, the staff can also play some security functions, which are not possible with wheeled robots. Such a product has only one intelligent gimmick, and it will soon be unused. Why is this happening?Again, the application of this AI technology has not improved the industrial efficiency of the banking industry.
The third example is a robot application scenario in large-scale warehouse logistics. Thousands of rotating robots quickly move and sort the goods, which greatly improves the efficiency of logistics. It is the application of this kind of technology that guarantees our high requirements for logistics. The application of AI technology here is successful and pursued by the industry.

To sum up, we can say:Whether AI is spring or winter depends on its ability to improve industrial efficiency.

The improvement of industrial efficiency is closely related to human happiness

What is efficiency? From a system perspective, efficiency is the work input to a system and how much work it is converted into. The ratio of the output work to the input work can be understood as efficiency.

In industry, there is also industrial efficiency. The input here is the cost of hiring employees, land rent, raw material costs, etc. If the same input, we get a higher output, we have a higher industrial efficiency.

In the ancient royal family, how did the efficiency of the royal dining room owned by the emperor compare to the efficiency of today's takeaway platform? At that time, the imperial dining room was to prepare meals at all costs, and to always be prepared. The dishes must be prepared whether they are eaten or not. The pot is always stewed. As long as the emperor has demand, the whole process from cooking to transferring the dishes will be mobilized. It only costs the emperor to serve them.

Today's take-out platform contains far more varieties of vegetables than the hundreds or thousands of dishes in the royal dining room. Due to the improvement of industrial efficiency, everyone today can enjoy the convenience of this high efficiency. . It was not the eunuch who delivered us the food, but the knights. They reduced the distribution cost to a few dollars by grabbing orders and trying to order, but they also achieved the provision of special services for a valued customer, which is an improvement in industrial efficiency Brought.

▍Efficiency improvement = cost performance improvement = geometric level improvement of demand

Why is efficiency improvement important? Because its improvement can lead to an increase in the price / performance ratio of the product, which leads to a geometric increase in demand. Ford's Model T is a classic case.

In the 20s, cars, as a luxury that only the rich could afford, were gradually moving towards civilianization. Ford reduced the price of the Model T from US $ 20 to US $ 4000, and further reduced it to US $ 800-200 through its assembly line production. This means that for ordinary Americans, one year's salary can buy a car. Even during the Great Depression of 300, total car sales were still rising.

The same example happens to the hotel industry, they follow the same logic. Today, five-star hotels are no longer in sight for the average middle class. More and more people can enjoy such services. How can this be done? The hotel management company has standardized hotel operations. It used to require 3 housekeepers to serve one room. Now through division and standardization, a cleaning staff can clean up 20 rooms, and another procurement staff is responsible for purchasing fruit for 100 rooms.

We have greatly reduced costs and increased efficiency through the specialization of processes. In a chain network system, the best practices can be solidified into standards and then promoted to sub-optimal hotels. This is the value of a hotel management company, or any type of management company, because it improves the industry's Average efficiency.

From another perspective, in the hotel industry, hotel management companies often do not own hotels, it just helps hotel owners to run the hotel. So why are these owners willing to let them manage and agree to share a share of the profits? Because these management companies took the money, but improved the hotel's efficiency to a level that could not be reached anyway.

哪些 What are the means to improve industrial efficiency?

Although it is said that the avenue is the simplest, the simplest is the naturalism. Picking up an apple in the suburbs and selling it is very simple and efficient, but what about 100 million apples?

Efficiency comes from complexity, and complex systems can achieve excess "savings".

Case 1:Scale up and dilute costs

The most typical example is the hotel chain, which reduces the cost by connecting the individual hotels to achieve the improvement of industrial efficiency.

Case 2:Division of assembly lines to improve professional efficiency

In this way, Ford's T-type car through the assembly line to improve the labor productivity of workers, thereby reducing production costs and improve industrial efficiency.

Case 3:Increase conversion efficiency through more complex processes

For example, the raw materials used to make clothes were cotton and wool. Now, you can use some non-natural materials to make clothes, such as special fibers. The same is true of the paper industry. Previously, only high-quality bark could be used to make paper. Nowadays, the materials used are very rich.

Case 4:Gather channels to improve distribution efficiency

The large supermarkets represented by Wal-Mart are such a format. Invite customers to sell more things to you. Why can Wal-Mart beat so many rural shops in the United States? It is precisely because Wal-Mart can buy everything in one stop, other small stores can't.

Case 5:Long-term conversion to improve retention efficiency

This is the most commonly heard way to improve efficiency in the Internet age: give you an apple, but all your tofu will be bought from us in the future.

Case 6:Complex marketing to improve matching efficiency

Through different pricing strategies and targeted precise marketing, it can match up transactions faster.

Case 7:Asset reuse to increase cost efficiency

The emergence of railways has far-reaching significance, but in the beginning, it was difficult for the United States to build interstate railways because the cost was too high for companies to pay. Later, a single-track railway was invented. Prior to that, the railways were two-track. Two tracks had to be repaired. After the single-track was adopted, the cost could be reduced by 30%. Single-track railways cannot be separated, so how to ensure efficiency if multiple single-tracks are to be built? At this time, it depends on scheduling.

How does the AI ​​industry land?

Before discussing how the AI ​​industry will land, let's take a look at what the AI ​​era has brought.

Judging from the law of the emergence of technology, it is not the most important to judge the right or wrong of a single point of technology. What is more important is to judge the overall trend of it. You can also get good benefits in this technology application.

1. AINot omnipotent

First, AI cannot directly perform "advanced thinking" and cannot solve general problems, so open tasks are not suitable for AI. Furthermore, AI cannot give a 100% reliable judgment. In this case, using AI technology to patch in the original industrial efficiency model is often not the best choice. Because the original system is oriented to high reliability, automation and high accuracy. Finally, in terms of AI replacing people, we believe that AI lacks flexibility and the cost of replacing people will be very high. Take the operator as an example. They are professionally trained and have strong flexibility. Even if they cannot solve the customer's problem, they can also resolve disputes through communication skills.

2. To play the true advantages of AI

The core of the AI ​​era is firstly the era of intelligent decision-making. The real advantage of AI is decision-making ability. For example, apple planting can analyze a model to determine the probability of apple maturity based on data such as sunlight, precipitation, color change, and gas release during the growth period of the apple. This is something that humans and automation cannot do. This capability may be less accurate, but it has great commercial value over time.

3. AIThe era is the era of intelligent human-machine hybrid systems

Since AI is a decision system, what kind of architecture does this decision system have? Do not think that AI is a narrowly defined algorithm. The calculation of AI data itself is an intermediate link. Take the call center as an example. Through the analysis of historical data, AI can understand the needs of each call very well. This kind of understanding is far beyond ordinary operators, but is the operator valuable? of course. He can understand many spoken expressions and is good at emotional communication. If AI and operator are combined, this is the man-machine hybrid that focuses on machine decision-making.

Five steps to landing the AI ​​industry

The first step is to find the absolute pain point.

Using AI to patch today's industry models and doing what others can do is not an absolute pain point. So what is the absolute pain point? This requires everyone to find it based on their industry experience, but I can tell you what it is not. It is not that all kinds of business factors have strong constraints. At this time, even if you are very efficient, you cannot defeat the inefficient people because the replacement cost is too high. Hotel management companies do a good job, hotels are willing to join, because hotel properties themselves are not scarce. If you optimize the efficiency of gas stations a lot, will not many gas stations come to join you? Probably not, because the circulation of elements is insufficient.

In 2019, China's oil and gas exploration field is liberalizing private capital. We can use AI technology to analyze where there may be better oil fields. But if this field has not been opened before 2019, even if you solve this problem, you will not be able to leverage the market structure.

The second step is to judge the potential of artificial intelligence

After finding the absolute pain point, we have to judge the potential of intelligence, because not all problems are suitable for AI (Artificial Intelligence) to solve, such as designing an open and creative problem such as a spaceship. Then there are the key issues of other alternatives. These problems often require very high reliability and are not suitable for AI to solve. For example, the early diagnosis of tumors, if all doctors today cannot do it, even if AI can play a little role, it is a kind of ability compensation, but how can there be other alternatives, the inaccuracy of AI will be multiplied .

The third step is to design a complete system

If the pain points found are painful enough, AI can be solved, and we can't expect simple application of AI technology to solve this problem. Therefore, in the face of such a situation, we have to dare to make the system "heavy", which is more complicated. The first is to dare to touch hardware. Although hardware brings complexity and higher cost, it can bring the efficiency of information collection and form a closed loop for decision implementation. The second is to dare to touch operations, not only based on human-machine operations, but also dare to be an operator. The third is not to rely on the existing data, but to think backwards: What data do we need to introduce to solve this pain point?

The fourth step is to find the path

As an early stage venture capital company, Baidu Venture Capital will discuss with every project we invest in. What is the future end? In the next five or ten years, will we be able to do something subversive and industrial revolution? Even if the founder talks clearly, going back to implementation, it is likely to start with a small thing to improve the product.

For example, agricultural planting, from the beginning, I want to develop a new planting method, so the verification cycle of the entire project will be very long. Can you use existing data to improve some efficiency and reduce some costs in the original method? Although this is not the end, it is a very important way to start.

The fifth step is to seek model change

Once we cut in, we will seek to change the model. But keep in mind that the dividend period after the project is often limited, how to find a way to change the model and seek uplifting in a limited time is an important proposition. In the case of salmon farming, salmon can only live in specific waters, so natural waters do not need to make many artificial compensations. With the development of AI technology, some places that are not suitable for farming salmon have become suitable. For example, in a piece of water, salmon can grow very well, but suddenly a cold current comes, and the fish are frozen to death. For such environmental changes, using an intelligent system can precisely predict and compensate.

If such a technical system proves feasible, we will be able to delineate the upstream elements of salmon farming at the right time and make this project larger. Eventually, the efficiency of the industry will be brought up through changes in the model. If this can be done by others, and we are more efficient, this is a premium ability; if we upgrade it to something that others cannot do, it is a premium ability.

Transferred from Chaos University,Original address