Memory inheritor

Chapter 65 The Black Box

"Our second phase plan is to organize and refine the elements of the target audience's life scenarios, create a multimodal stimulation package, and then conduct closed-loop neural feedback stimulation to construct a cross-modal model specific to the target audience. Then we can interpret that 'abnormal neural signal'."

"However, this plan requires a thorough understanding of the target audience. Obtaining this information will take a considerable amount of time, and even if the deciphering is completed by then, it may be of little value."

Yu Xiufeng gave a complete introduction to the work they had completed and their follow-up plans, without shying away from mentioning the process of inviting Wei Hongsi.

In his view, this was perfectly normal. How could he know if someone was suitable for the project team if he didn't put them through a test?

Wei Hongsi was not surprised that they had already completed the first step of the work.

The idea of ​​extracting information from Mu Jie's mind must have been proposed by these two experts from the task force, so how could there have been no progress?

Therefore, the plan he wrote for the task force skipped the first step.

After Yu Xiufeng finished speaking, he first nodded to Wei Hongsi, and then looked at Zheng Han.

"Then please let Professor Wei introduce your proposal." Zheng Han was still a little unaccustomed to speaking.

"Okay." Wei Hongsi nodded.

Although people often compare the brain to the memory mechanism of a computer, there are fundamental differences in their information storage patterns.

Computers store information in a "structured" manner, where data is precisely placed in specific physical or logical locations (such as sectors or pages), just like coordinate points, which are fixed and independent.

The brain's memory is a "network" structure. Information enters the brain through the senses, triggering electrical activity in specific neurons. These activated neurons connect with each other through synapses, forming a dynamic and reconfigurable neural network.

More importantly, the information stored by computers is essentially static encoding, while the brain, when memorizing information, often binds it to related sensory experiences, emotions, and contexts.

For example, saving a poem to a computer is like putting a file in a folder. When you need it, you can retrieve the file, and it will be exactly the same as when you saved it. This storage mode only preserves the text of the poem itself.

The brain's memory is like a net covered with things. When reciting a poem, the brain not only memorizes the textual information, but also automatically "sticks" many related details, such as the mother's voice, the lighting in the room at the time, and the emotional atmosphere of that moment.

Simply put, computers store "information," while the brain records "experiences."

Therefore, memory is never isolated.

Because memories are a mixture of many personalized feelings and impressions, each memory carries the traces of individual consciousness.

From this perspective, the content instilled in Li Yunxiang contained some "knowledge memory," which is more advanced than Peng Jun's purely "task script."

It was those fragments of consciousness in her "knowledge memories" that clashed with Li Yunxiang's "id," causing her brain to accumulate experiences and feelings that did not belong to her, leading to the emergence of a second price.

Let's return to Mu Jie's question.

The basic principle used in Zheng Han's second-phase plan is the associative nature of memory.

The "abnormal neural signal" in Mu Jie's brain must also be related to a specific scene, person, sound, or emotion. As long as this information can be understood, a large amount of training data can be created; through the corresponding records of stimulation and feedback, a translation from "neuro-language" to text can be trained exclusively for him.

This process is similar to the original translation of a language.

For example, imagine two people who don't speak the same language but live in similar environments. Person A points to their nose, and Person B says "nasus." Person A then knows that "nose" = "nasus." Person A takes out a key, and Person B says "clavis." Person A then knows that "key" = "clavis." By continuing in this manner, from simple to complex, a translation of Person B's language can be created.

The problem is that Mu Jie's "abnormal neural signals" are most likely related to his overseas experience. Currently, the police know very little about the relevant situation, so they cannot establish the initial elements of the life scene.

Let's take A and B from the previous example again.

B often mutters a phrase, a trick for catching a certain fish at the seaside. A lives inland, has never seen the sea, let alone heard of fishing, and naturally cannot imagine B fishing at the seaside, nor can he find anything to compare it to. In this situation, it becomes very difficult for A to translate B's trick.

Zheng Han and his team encountered this problem with their solution.

Only by understanding Mu Jie's unique experience overseas can we construct effective elements for a real-life scenario. But if we already know his unique experience, what's the point of deciphering the meaning of that unusual signal?

Wei Hongsi knew Mu Jie's general situation, so he had considered this possibility during their phone call last night. The plan he designed was actually based on the same principle as Zheng Han's plan, but it was an advanced version.

Since no effective clues could be obtained through background investigation, Mu Jie's brain was treated as a black box. Various basic stimuli were input into it, and changes in its output signals were observed to deduce its internal coding rules in reverse.

The specific method involves establishing a basic semantic stimulus library that covers the broadest possible semantic range. Then, a "closed-loop neural probe" is performed. When the target brainwave appears, a set of preset stimuli is injected, and the brainwave is monitored with high precision to determine whether it is amplified, distorted, interrupted, or unaffected. Finally, corresponding data on stimuli and responses are established.

This process transforms stimulus strategies from precisely targeted information like the nose and keys into a systematic approach of probing and feature capture.

For example, start with the most basic numbers 1, 2, and 3 to obtain corresponding feedback. After verification and feature extraction, the corresponding translation information unus, duo, and tres is obtained. Then, using these as anchors, the system is continuously expanded and explored to gradually build the entire "neural translation package".

After listening to Wei Hongsi's explanation, Zheng Han and Yu Xiufeng both fell into deep thought.

The principle is not complicated, and they understood it immediately, but there are still a few difficulties in this solution.

Some of these problems can be solved, such as handling signal ambiguity and high-noise environments, and the real-time and synchronization requirements of closed-loop systems.

There are also some problems that they can't figure out right away, such as how to control the effectiveness of stimulation and how to accurately identify individual neural codes.