Algorithm system

Silicon-based life can only sufferAlthough Lin Yuan was still doing basic work at the Good Group Zijin Branch, earning a high salary, during the deployment and testing process, he discovered through past test records and reports that the AI algorithm related to food delivery had been continuously iterating.

Regardless of the current frustrating performance of these AI algorithms in the food delivery industry, at least Good Group has consistently been trying to optimize them.

This is actually quite easy to understand; every link in the entire food delivery industry’s economic chain can no longer be cut.

Given the current delivery unit price and considering public opinion, they dare not make further cuts to the delivery riderss.

Many of their own programmers have been outsourced, but they cannot all be outsourced; otherwise, if the software system has a major issue, it would be disastrous.

Let investors earn a little less? Once the financial report looks bad, the stock price will plummet. Even the CEO would have to pack up and leave.

Make users pay an extra two yuan for ordering food? They want to, but users aren’t stupid now. At worst, they just won’t order takeout.

So what to do? Fortunately, software and machines don’t talk, so they can only be squeezed.

Lin Yuan could only joke to himself: This is a premeditated, systematic oppression of silicon-based life by carbon-based life. (Of course, if silicon-based life existed)

After carefully reviewing Good Group’s test records and combining them with his own experience as a food delivery delivery riders, Lin Yuan gained a general understanding of the bottlenecks in AI application within the food delivery industry.

For example, the simplest issue of order reassignment.

Order reassignment might be due to a particular merchant being too slow in preparing food, or the delivery riders’s personal inability to deliver in time, or other unexpected circumstances. From the perspective of controlling management costs, the best solution is naturally for the system to automatically identify and reassign.

The entire process should ideally be completed silently, without any human intervention. This would make the station masters’ work easier, wouldn’t it?

Of course, making station masters’ work easier is not the goal; reducing the number of station masters is the goal.

Alas, but unfortunately.

Those who open shops, those who cook, and those who deliver food are not robots.

If they were all robots, it would be easy, because silicon-based life is easiest to oppress.

If I set you to prepare food in 10 minutes, and you can’t, then there must be a problem with the software or hardware somewhere. Then I’ll upgrade and update you. If it still doesn’t work after the upgrade? I can throw you into the electronic waste recycling plant and replace you with a plus, pro, max, or ultra version.

But obviously, you cannot treat carbon-based life this way, because carbon-based life is at least protected by law.

Carbon-based life cannot be precisely controlled by programs, nor can it maintain real-time contact with the backend system. If a certain meal is predicted to be unable to be prepared in time, a cooking robot could give an early warning notification, but a human chef obviously cannot accurately predict or instantly send messages to the backend.

Because carbon-based life forms clearly do not have transmission modules like WiFi in their brains, the communication methods between carbon-based life forms are ultimately too slow and primitive.

Even the communication method between Lin Yuan and the computing power system is limited by this—the system can collect data through Lin Yuan’s eyes, but the system cannot print out the processed results through Lin Yuan’s hands. Even if Lin Yuan wanted to grant the system permission to control his hands and feet, he knew that his own flesh and blood structure should not be as fast as a printer, let alone comparable to WiFi transmission.

Simply put, it is due to the ’human factor’ that the food preparation time for each meal is unpredictable. At the same time, also due to the human factor, each delivery riders’s daily delivery capacity is unpredictable. Similarly, the behavior of those who order food is also unpredictable.

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

He remembered the silly girl who gave him milk tea. If everything were programmatic, humanity would lose that kind of ’eye-opening’ ’heart-fluttering’ beauty.

However, not all unpredictability produces positive effects, and to deal with unpredictability in human work scenarios, AI becomes the best choice.

Ultimately, this is about using the advantages of silicon-based life to compensate for the disadvantages of carbon-based life.

However, from Good Group’s test records, it seems that the silicon-based life forms written by Good Group’s engineers are clearly not very effective.

Due to the unpredictability of the people who prepare, deliver, and order food, the AI algorithm faces the problem of ’missing data samples.’ Therefore, it would be a miracle if the algorithm could be accurate.

This is like a problem where some of the known conditions are hidden; it would be strange if you could solve it.

Therefore, Lin Yuan deeply understood Good Group’s helplessness, and he also discussed it in detail with the head of the technical department at the Zijin Branch. Good Group’s current algorithm strategy mainly consists of two approaches: estimating order volume from past historical data, and optimizing delivery riders delivery routes, as well as dispatch and reassignment logic through AI calculations.

Estimating order volume for a specific area is easy to understand. After predicting the order volume, they can then strategically arrange station distribution and the number of delivery riderss.

Although if the weather changes even slightly, order volume and delivery efficiency will fluctuate accordingly. Holiday factors also occasionally interfere. However, overall, the order volume estimated from historical data is roughly seventy to eighty percent accurate.

As for route and dispatch optimization, it’s even easier to understand; it’s about making delivery riders deliveries smoother. This increases labor efficiency per unit time and enhances returns for all aspects of the entire economic chain.

From the detailed test reports, Lin Yuan saw that Good Group spent eighty percent of its energy on route and dispatch optimization. After all, compared to the total order volume of an area, route and dispatch issues are much more complex.

If a food delivery order is compared to a navigation from point A to point B, then from the perspective of the delivery-side algorithm, each delivery riders’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 one navigation, this problem seems very simple.

But first, food delivery orders are different from travel navigation; the delivery riders first needs to pick up the food, so the route becomes A ==> B ==> C. The delivery riders first goes from their receiving location A, to the merchant’s location B, and then delivers to the customer’s location C.

The time to complete step A to B is limited by the location of merchant B and the merchant’s food preparation speed. The location of B cannot be simply obtained by the straight-line distance on an electronic map; if the merchant is in a mall, the delivery riders also needs to consider parking their electric scooter and climbing stairs.

And the process from B to C is even more variable. Because there are far more customers than merchants, and they are much more widely distributed. delivery riderss face issues such as climbing stairs and whether they can enter residential communities.

A single food delivery order faces various problems, and each delivery riders cannot possibly have only one food delivery order, but a collection of several food delivery orders.

And from the algorithm’s perspective, it is N times more complex, because the algorithm has to consider all delivery riderss.

If the overall complexity is represented by O, and the complexity of each order is represented by o.

Then O = o * number of orders per delivery riders * total number of delivery riderss.

And that’s not all.

Because different orders in each delivery riders’s hand interfere with each other. Since delivery riderss deliver in series, they must deliver one meal before delivering another. If a 6th meal is added when they have 5 meals, that 6th meal is very likely to affect the delivery efficiency of the other 5 meals.

This leads to the logical complexity of each food delivery order being non-linear, and not only non-linear but also random.

In short, there are simply too many factors affecting routes and dispatch. Some factors are even unobtainable.

Lin Yuan summarized the problem.

Good Group believes that the most worthwhile optimization in the food delivery industry is: routes and dispatch. And because route and dispatch problems are overly complex, they require a very high model complexity for AI algorithms, which in turn demands significant computing power. Furthermore, due to the problem of missing data collection, the trained algorithms are inherently flawed.

Thus, Lin Yuan naturally thought—why don’t I try using the computing power system?