Xueba's computing system

Chapter 21: Silicon-based life can only suffer
Although Lin Yuan was earning a high salary and still doing little work in the Zijin branch of Haotuan, during the deployment and testing process, he discovered through previous test record reports that the AI algorithms related to food delivery were constantly iterating.

Regardless of the poor performance of these AI algorithms in the food delivery industry, at least people are trying to optimize them.

This is actually easy to understand. Any link in the economic chain of the entire takeout industry can no longer be cut.

With the current delivery unit price, and considering public opinion, they dare not cut the knife to the riders. Many of their own programmers have been outsourced, but they cannot outsource all of them, otherwise it will be a big deal if the software system causes a big problem.

Let investors earn less? If the financial report is not good, the stock price will plummet. Even the CEO will have to pack up and leave.

Asking users to spend two more yuan to order food? It’s tempting, but users nowadays are not stupid. At worst, they can just not order takeout.

What to do? Fortunately, software and machines cannot talk, so the only way is to exploit them.

Lin Yuan could only joke in his heart: This is a premeditated systematic oppression of silicon-based life by carbon-based life. (Of course, if silicon-based life exists)

After Lin Yuan carefully read the test records of Haotuan, and combined it with his own experience in delivering food, he had a general understanding of the application bottlenecks of AI in the food delivery industry.

For example, the simplest order transfer problem.

The order may be transferred because one of the merchants is too slow to deliver the food, or because the rider is not capable enough to deliver the food in time, or because of some other unexpected circumstances. From the perspective of controlling management costs, the best solution is to automatically transfer the order after automatic identification by the system.

It is best if the whole process is completed quietly without any human intervention. This will make the work of webmasters easier.

Of course, making the work of webmasters easier is not the goal; reducing the number of webmasters is the goal.

Alas, but unfortunately.

The people who open the shop, cook the food, and deliver the food are not robots.

It would be easy if they were all robots, because silicon-based life is best oppressed.

I set it to serve you food in 10 minutes. If you can't serve, there must be something wrong with the software or hardware. Then I'll upgrade it for you. What if it still doesn't work after the upgrade? I can throw it to the electronic waste recycling plant and I'll replace it with a plus, pro, max, or ultra version.

But obviously you can't treat carbon-based life like this, because carbon-based life is at least protected by law.

Not only can carbon-based life not be precisely controlled by programs, but it also cannot maintain real-time communication with the backend system. If a certain meal prediction is too late to be issued, the cooking robot can give an early warning notification, but the cooking chef obviously cannot make an accurate prediction or send a message to the backend in real time.

Because carbon-based life obviously does not have a transmission module such as WiFi in its brain, the way of communication between carbon-based life is ultimately too slow and too primitive.

Even the communication between Lin Yuan and the computing system is limited by this - the system can collect data through Lin Yuan's eyes, but the system cannot print out the results through Lin Yuan's hands. Even if Lin Yuan wants to give the system the power to do what he wants, he knows that his own flesh and blood structure cannot match the speed of the printer, let alone WiFi transmission.

Simply put, it is because of the "human factor" that the delivery time of each meal is unpredictable. At the same time, it is also because of the human factor that the delivery capacity of each rider on the day is unpredictable. Similarly, the behavior of the guys who take orders is also unpredictable.

Even so, Lin Yuan feels that the beauty of carbon-based life lies precisely in this unpredictability.

He thought of the silly girl who gave him milk tea. If everything is programmed, then humans will lose the beauty of "bright eyes" and "heart-pounding".

But not all unpredictability will have a positive effect, and to deal with the unpredictability in human work scenarios, AI becomes the best choice.

In the final analysis, this is using the advantages of silicon-based life to make up for the disadvantages of carbon-based life. However, from the test records of Haotuan, Lin Yuan found that the silicon-based life written by Haotuan engineers was obviously not very powerful.

Due to the unpredictability of people who serve food, deliver food, and take orders, AI algorithm recognition faces the problem of "missing data sampling". Therefore, it is impossible for the algorithm to be accurate.

It's like a problem in which some of the known conditions are hidden, then it would be strange if you could solve the problem.

Therefore, Lin Yuan understood Haotuan's frustration, and he also had a detailed discussion with the head of the technical department of the Zijin branch. Haotuan's current algorithm strategy mainly consists of two moves: estimating the order volume from historical data, and optimizing the rider's delivery route and dispatch and transfer logic through AI calculations.

It is easy to understand how to estimate the order volume for a certain area. After predicting the order volume, you can arrange the station distribution and the number of riders in a targeted manner.

Although the order volume and delivery efficiency will fluctuate as long as the weather changes slightly, and holiday factors will occasionally interfere, but overall, the order volume estimated based on historical data is also about right.

As for route and dispatch optimization, it is easier to understand, that is, to make the delivery of food by riders smoother, improve labor efficiency per unit time, and increase the benefits of all aspects of the entire economic chain.

Lin Yuan saw from the detailed test reports that 80% of Haotuan's efforts were spent on optimizing the routing and dispatching of orders. After all, compared with the total number of orders in a region, routing and dispatching issues are much more complicated.

If a takeaway order is likened to a navigation from point A to point B, then from the perspective of the delivery algorithm, each rider’s task is a collection of navigation routes. The algorithm’s task is to plan this collection of navigation routes reasonably.

If it's just a navigation, this problem seems simple.

But first of all, takeaway orders are different from travel navigation. The rider needs to pick up the food first, so the route becomes A==》B==》C. The rider must first go from the order receiving location A to the merchant location B, and then deliver it to the customer location C.

The time to complete step A to B is limited by the location of merchant B and the speed at which the merchant can serve the food. The location of B cannot be simply obtained by the straight-line distance on an electronic map. If the merchant is in a shopping mall, the rider also needs to consider the parking problem of the e-donkey and the problem of climbing stairs.

The process from B to C is more varied. Because there are many more customers than merchants and they are more widely distributed, the riders face problems such as climbing stairs and whether they can enter the community.

A takeaway order faces various problems, and each rider cannot have only one takeaway order on hand, but a collection of several takeaway orders.

From an algorithmic perspective, it is N times more complicated because the algorithm has to take all riders into account.

If the overall complexity is expressed as big O, the complexity of each order is expressed as o.

Then O=o*number of orders of the rider*total number of riders.

This is not over yet.

Because there are interferences between the different orders on hand for each rider. Because the riders deliver meals in series, they must deliver one meal after another. If there is a sixth meal when there are 5 meals on hand, then this sixth meal is likely to affect the delivery efficiency of the other 6 meals.

This results in the logical complexity of each takeout order being non-linear, and not only is it non-linear, it’s also random.

In short, there are too many factors that affect the routing and dispatching of orders, and some of these factors are simply not available.

Lin Yuan summarized the problem.

Haotuan believes that the most important aspects of the food delivery industry to be optimized are routing and dispatching. However, routing and dispatching are too complex, which requires a high level of model complexity for AI algorithms, and thus a high level of computing power. And due to the lack of collected data, the trained algorithm is bound to have inherent defects.

So, Lin Yuan naturally thought of this - why don't I try the computing power system?
(End of this chapter)