War of Intelligent Machines

Chapter 17: The First Setback

Chapter 17: The First Setback

After Jiang Yue Sheng finished speaking, the group leader scanned the room with a smile.

It was clear that he was asking if anyone had any more opinions.

After a moment of silence, the group leader saw that no one seemed ready to speak, so he said, “Alright, everyone has spoken almost enough today, but I think it’s still too early to make a final decision. Let’s take three more days to think it over and aim to have a preliminary conclusion after that.”

The day after the meeting, a young man from the team named Shi Jun approached Jiang Yue Sheng.

“Jiang, I think your solution has potential, so I’d like to work with you on this. What do you think?”

Jiang Yue Sheng was naturally pleased to have a supporter. Besides, from his observations, Shi Jun was quite capable, so he responded, “Sounds good. If our plan gets approved, we’ll tackle it together.”

Since this meeting was crucial for determining the initial plan, the group leader had invited two senior consultants from the company to attend.

“Alright, everyone, let’s start by sharing our ideas one by one,” the group leader said, glancing around the room.

Silence.

“Let me go first,” a female programmer broke the silence.

“I think Jiang’s software approach has some issues. I’m not questioning the fact that solving the problem from a software perspective is cost-effective and fast, but improving recognition based on architectural aesthetics is difficult for us non-experts. Mainly, we lack experience in building aesthetic models.”

“I also think Jiang’s approach has problems!” Another person raised their hand and began speaking. “Not only do we lack experience in architecture, but I also doubt the accuracy of this solution. Aesthetics itself is somewhat vague—criteria are unclear. Moreover, streets and buildings are so diverse that it’s hard to extract common aesthetic features. I’m still in favor of using physical parameters like electric or magnetic fields for detection. Performance and cost are directly proportional, so to improve performance, we need to raise the cost.”

Everyone started debating eagerly, exchanging their views.

Jiang Yue Sheng listened carefully and made notes.

“I think Jiang’s plan is feasible,” Shi Jun raised his hand to speak.

“I’ve spent a few days studying AI models, and I found that architectural aesthetics actually follow certain patterns. We don’t need to understand too many aesthetic factors, like color, depth, or texture. We just need to focus on the proportions of buildings. It’s similar to facial recognition technology, which mainly identifies the ratio of facial features and doesn’t care much about the person’s weight or skin texture. If we build some basic models based on the overall and local proportions of buildings and test them, we can see if the results are good. If they are, we’ll proceed; if not, we’ll rethink our approach.”

“That’s a valid point!” A senior consultant who had been silent up until then spoke up. “I think we can try multiple approaches in parallel—develop and test both the architectural aesthetic model and the electromagnetic physical parameter model. We’ll go with the one that shows the best results!”

“I agree with Dr. Liu,” the group leader chimed in. “Let’s split into two teams. Jiang Yue Sheng will lead the team focusing on architectural aesthetic modeling, while Sun Hui will lead the team focusing on electromagnetic and physical parameter recognition. We’ll work separately, and whoever comes up with results first will have their solution tested!”

Since Jiang Yue Sheng was new and young, not many were eager to work with him. In the end, he formed a team with Shi Jun and two recently graduated engineers—four people in total—dedicated to the architectural aesthetic modeling project.

Although Jiang Yue Sheng had limited experience in software development, having only worked on small robot programs, he was tackling large-scale modeling for the first time. Luckily, Shi Jun had a formal background in software engineering and was able to compensate for Jiang Yue Sheng’s shortcomings in this area.

Two months later, the electromagnetic parameter recognition team finished writing their code and entered the implementation and testing phase, but the results were disappointing. The accuracy of camouflage detection was low.

Jiang Yue Sheng’s team was smaller and slower, and it took them three months to complete the first version of their software and begin testing.

They installed the new software on a group of robots and brought them to the same test environment as before.

The test began, and the robots from the Solution Group, now upgraded with the new system, quickly detected the camouflaged enemy robots within their line of sight and launched successful attacks. They proceeded to push deeper into the complex building area.

Many people were watching the experiment, standing on the outskirts of the test field or following behind the robots. Everyone was eager to see if Jiang Yue Sheng’s new approach would succeed.

As the robots advanced, eliminating detected enemies, they reached the center of the dense building cluster.

This environment was highly challenging for the robots’ processing power because, as the robots moved, the layout of stationary objects constantly changed in their field of vision. The more complex the environment and the faster the relative movement, the greater the computational demands on the robots.

At this point, Jiang Yue Sheng noticed an issue. He saw six enemy robots standing together in a formation against a damaged wall, forming a camouflage wall about three meters high and two meters wide. From a distance, it looked like a broken wall. He immediately recognized it as a robot disguise, but the robots from the Solution Group failed to detect it.

Others observing the test noticed the problem as well. Jiang Yue Sheng realized the software still had issues, so he signaled to the Scenario Group to pause the test.

He quickly connected to the robot using a specialized terminal and found that the robot had indeed categorized the camouflage wall as just a broken wall.

Since the camouflage was large enough, the algorithm had classified it as a physical wall. This meant the algorithm, which only relied on the proportions of buildings to distinguish between real and fake, still had flaws—or rather, it was not yet mature.

“Well, well! Progress!” The group leader walked up from behind and patted the embarrassed Shi Jun on the back. “This shows that the proportional recognition algorithm still has potential, but we need to keep refining it. Let’s keep exploring this direction for now. If it doesn’t work out, we’ll try something else.”

“Leader, I have an idea,” Shi Jun said with keen determination, looking at the group leader. “The six robots disguised behind the wall could be detected with an infrared sensor. Six heat sources gathered together, or even four, could be defined as a suspicious target, right?”

“Right!” The group leader responded nonchalantly, but in his mind, he was evaluating Shi Jun’s idea. He felt it lacked a systematic approach. So he added, “Let’s all go back and think this through. We’ll reconvene in a week.”