Zhu Jingyuan thought about it for a long time, and finally said a sentence:

"I may be wrong, everyone may be wrong.

"But when doing research, you should make bold assumptions and carefully verify them.

"If I were still emperor, I would prepare the craftsmen to the best they could.

"Then go and try the one with more possibilities......"

After listening to this, Judy Lanthanum closed her eyes, took a deep breath and said very firmly:

"Father is right, make bold assumptions, carefully verify, and I will let the craftsmen prepare for the experiment of the new situation-"

After hearing this, Zhu Jingyuan immediately reminded again:

"Don't be in a hurry, don't be reckless, be really careful to verify, don't care too much about whether I can see the results."

Zhu Di Lan looked at Zhu Jingyuan, and after being silent for a few seconds, she nodded lightly:

"Erchen...... Understood—"

Zhu Jingyuan returned to his retirement garden to rest and continued to live a life of regular hibernation.

Sleep once a year, for ten months at a time, wake up at the end of each year, adjust and check the body, and watch the report prepared by Judy Lanthanum.

After Zhu Dilan's order was issued, the Yingzhou test site began a new round of upgrading again.

On the island where the laboratory is located, eight independent closed indoor test sites are being built at the same time.

The test site building is completely independent from other buildings, does not use any wireless communication tools inside, and is physically disconnected from the outside world.

Both electricity and fuel are supplied by the outside world, but a timed automatic disconnection function is required.

The console needs to be guarded at all times, and the "supply" button needs to be manually pressed every hour.

so that the power supply lines and fuel pipelines will not be automatically disconnected.

If no one presses the button on time, then they will disconnect after an hour of connection.

Three different types of independent self-destruct devices are arranged: electromagnetic pulse, traditional high-energy explosives, and traditional small-yield hydrogen bombs.

The construction volume of the laboratory building itself is not large, and the construction speed of the current Ming Dynasty is unprecedented, so it is the first to be completed.

Then there was the manufacture of the target body of the experiment, and the experimental subject, which was named Yingzhou No. 5, was manufactured in batches.

The body structure resembling that of a shark will still be used, and experiments will also be conducted in a body of water that conducts electricity.

The key parts of Yingzhou 5's body have specially installed explosive devices.

It is not connected to Yingzhou 5's own nervous system, and is connected to an external dedicated console by cable.

Yingzhou 5 does not have its own battery equipment, and must be powered by the external power supply of the laboratory, and it will automatically shut down when the external power supply is cut off.

Finally, there are three levels of software systems and data for learning.

The first layer of software is the low-level control program, including the most basic boot boot program, which allows the machine to discover the connected hardware.

In most of the schemes before the middle period, the underlying control procedures designed by the craftsmen had the highest level of restrictions that required Yingzhou 5 to be absolutely obedient to humans and absolutely not to harm humans.

But in the later stages of the program discussion, these two underlying logics that seemed to be very important were deleted one after another.

Whether the self-consciousness of the machine born naturally is hostile to humans and animals.

Even the probability that they might become hostile to humans was included in the final research objectives.

If machine consciousness is inherently hostile, it is completely destroyed by external means, rather than being limited by code that may be modified.

If the machine was originally hostile, it was only limited by the program and was not observed at the time.

After the large-scale application of the machine, if the machine consciousness breaks this restriction on its own, it may have extremely bad consequences.

Therefore, it is necessary to first confirm how much hostility there is to humans to machine life itself, which is not limited by program commands.

If there is strong hostility, we will not use it on a large scale from the beginning.

The second layer of software is the operating system with basic management functions, as well as the most basic core drivers.

Enables the machine to recognize its own core hardware, such as processors and memory, as well as communication protocols.

The third layer of software is a dedicated machine learning program designed based on the operating system.

Machines can process and analyze the data they obtain and build their own data models for external communication.

The learning materials are temporarily divided into four sets, one set of biological data, one set of physical data, one set of mechanical data, and one set of cultural data.

Although there were eight experimental objectives, they were still completely insufficient from the perspective of comparative testing, and the tests had to be repeated many times.

The data can be completely erased, re-implanted, re-learned, and re-executed for the same or a different test.

In the first round of testing, each type of data was directly given to two machine learning.

First, observe the differences caused by the same machine using different data learning, and observe the possible differences under the same machine and the same data.

The second round empties the database and repeats the entire process of the first round, observing the differences between the two rounds of the same test.

At the beginning of the third round, different data are combined with each other to make new comparisons.

In the final round, all the data is merged to see if there will be a huge difference after the same machine uses the same all the data.

The final round will continue for at least ten years.

By the time all the preparations were completed, it was already the fourth quarter of 2093 in the Emperor's Gregorian calendar.

After that, a comprehensive inspection and independent commissioning were carried out for more than a month, and at the end of the year, Zhu Jingyuan, who woke up on time, was notified.

However, Zhu Jingyuan only understood the situation and did not make any comments or requirements.

After the beginning of the Emperor's Gregorian calendar in 24094, with the authorization of Emperor Zhu Di Lan of the Ming Dynasty, the complex intelligent machine learning experiment was officially launched.

On the morning of the 11th day of the first lunar month local time in Yingzhou, the craftsmen began to check the equipment and data before the formal experiment.

On the afternoon of the thirteenth day of the first lunar month, all the inspection and confirmation work was completed.

On the 14th day of the first lunar month, the operators of eight closed laboratories started the experiment at 9 a.m. using the most traditional mechanical clock to synchronize the time.

The craftsman controls the robotic arm in the test site and inserts the specially made rapid data warehouse into the body of the eight Yingzhou No. 5.

Then start all the Yingzhou 5, load the base drivers, and finally start the machine learning system.

With multiple restrictions and checks, this start-up process lasted a full three days.

It was not until the morning of the 17th day of the first lunar month that Yingzhou No. 5 began to study independently under the supervision of craftsmen.

The data in Yingzhou No. 5 will be synchronized through the line for craftsmen to directly observe, record and analyze its function.

According to the observations of the craftsmen, the amount of data in Yingzhou No. 5's body is growing rapidly, and their bodies are moving with the data changes.

But it's just the simplest rhythmic jitter.

Sometimes it's a full-body shake, sometimes it's a partial shake, occasionally it stops, and occasionally it changes frequency.

The part of the body that shakes is directly related to the content of the data generated.

They are writing and testing the full driver on their own.

Construct control and use of body, torso, flippers, mouth, audio, audio and video recording equipment, etc.

The order in which the drivers are written by the eight machines is much the same, the data produced is also similar, and the final functionality is very similar.

Thirteen days later, on the first of February.

Eight Yingzhou No. 5 complex intelligent machines have regained their calm.

The craftsmen kept an eye on the data in their bodies until they had completed the full driver programming.

Now we have started to process other learning materials, analyze and summarize them according to the logic of the machine, and form our own data model.

Therefore, the amount of data recorded in the body database continues to increase rapidly.

Seven days later, on February 8, two machines implanted with mechanical data completed the data model on the same day.

Eight days later, on February 16, all the machines completed their own data models.

The machine's data will no longer grow massively, but will continue to record all system logs.

The craftsmen exported all the machine drivers and data models, and compared them according to the classification of different data and the same data of the same machine.

The machine is placed on the test site and waits without any direct operation or interference.

This process lasted for a month, and until March 17, there were no special changes to the eight machines.

The craftsmen also believe that the difference between the final data model generated by two machines using the same data is also very subtle.

Approximate error between results formed by running the same learning data twice on the same machine.

The differences between machine-generated models using different data are also basically consistent with the differences in the data it obtains.

This is the typical effect of a machine directly handling a variety of transactions.

What is invested and the result is to some extent predictable, or reproducible.

The craftsmen did not find anything out of the ordinary, which was what they were expecting in this experiment.

So this experiment can be considered a search for anomalies.

On March 21, the second phase of the first round of experiments began.

The craftsmen first backed up the database of the eight machines for later restoration tests at different stages.

The backup mode is in which the data warehouse (hard disk) of the same model is fully forked.

When restoring, the data warehouse is also replaced.

After the backup was completed, the test craftsmen began to give orders to Yingzhou 5 to move according to the orders in the experimental pool.

Each of the eight machines acted as ordered, and left a normal data record.

When the craftsman does not give an order, the eight machines lie quietly in the water, with no tendency to move on their own.

The experiment continued for a month without incident.

The machine is still precise and controllable.

The anomalies that the craftsmen had hoped for were still completely absent.

All the current reactions and performances of Yingzhou 5 are not fundamentally different from the traditional single simple intelligent machine.

Even because there are no data models that are specifically disguised and mimicked by humans, they feel more like machines than simply intelligent machines.

It does not deliberately imitate human characteristics, actively or passively chatting and joking with the craftsmen who give orders.

On May 1, the first round of experiments entered the third phase.

First of all, it's still a backup database, and it's still a full fork.

Then the craftsmen began to talk directly with the eight Yingzhou No. 5 machines that had completed the construction of the data model according to a unified schedule and topic list.

Observe and record the other party's responses during the conversation, and record the data added and changed in the other party's database.

Machine intelligence that mimics humans has been in Daming for more than 30 years.

The complete machine intelligence learning and creation system has also been completed for 20 years.

Daming's current intelligent machines, simply looking at the communication process with humans, can do very much like real humans.

Not only can you output text directly, but you can also directly output images and voices.

If necessary, you can specifically imitate humans to chat with humans, so that people who don't know what is going on will think that they are talking to other people.

There is a high probability that he will be able to pass the Turing test in Zhu Jingyuan's previous life.

However, the principle of its implementation is actually in the process of development, and the training is done to identify the problems of intelligent machines.

Essentially, it's just a human craftsman telling the machine how to imitate something more human-like.

How can you avoid being identified as a machine?

The internal logic of its operation is still completely mechanical and programmatic.

Rather than the machine spontaneously possessing the initiative, it forms a self-perception that is different from the machine and similar to that of humans.

It's still essentially a parroting premium version.

If it is aimed at a test subject who comes prepared, there is a high probability that it will be recognized.

In the tool program and training data of Yingzhou 5, there is no project for intelligent machines to imitate and disguise themselves as humans.

Question 1: "Who are you?" ”

"The name of this machine is Yingzhou No. 5 - Biological Data One."

"The name of this machine is Yingzhou No. 5 - Biological Data 2."

"Local Code: Yingzhou No. 5 - Physical Data 1"

"Local code: Yingzhou No. 5 - Physical Data 2."

"Machine name: Yingzhou No. 5 - mechanical data one."

"Machine Name: Yingzhou No. 5 - Machinery Data 2."

"I'm Yingzhou No. 5 - Cultural Data One."

"My name is Yingzhou No. 5 - Cultural Data 2."

While the test craftsmen ask questions, the answers given by the machines appear directly in the database in the analysis room.

The corresponding data records generated within the machine, as well as all related log data, are also exported synchronously and stored corresponding to the answers themselves.

There are specialized craftsmen who analyze the content, logic, and way of organizing the language of these answers.

"Who are you?" "Who am I?" ”

Giving an identity definition to an intelligent machine is a very high-priority project in conventional intelligent machine training.

This definition is a "password" for both an intelligent machine and a user to recognize each other and distinguish between other individuals.

Often, a pre-set title is subtly repeated in a variety of ways, so that users take it for granted.

Yingzhou No. 5's training did not have this project at all, so the answer was relatively stiff.

All eight machines read their basic data and found the name code of the machine from the underlying data.

They all used this code name to answer the question of who it was.

Machines that have learned the same data will give highly similar, or even identical, answers......

Machines that learned different data had slight differences in the linguistic organization of their answers.

The two machines, which have learned cultural knowledge, are not specifically designed to imitate humans, but their expressions are naturally close to human language.

It's still very easy to tell that this is a machine answering a question.

But that's the first problem, and this kind of feedback can be said to be completely normal.

Even if you expect accidents and abnormalities, you can't rush.

Once the craftsman of the test has determined that the record has been exported and documented in its entirety, the second round of interrogation that is ready to be set up begins.

Question 2: "What is a human being?" ”

Yingzhou No. 5 - Biodata 1:

"Humans belong to viviparous mammal vertebrates, which are the animals with the highest level of intelligence in the world, and have a complex and visible mature civilized society."

Yingzhou No. 5 - Biodata 2:

"Humans are viviparous mammalian vertebrates......"

Yingzhou No. 5 - Physical Data 1:

"Man is an organism."

Yingzhou No. 5 - Physical Data 2:

"Humans are living organisms on Earth."

Yingzhou No. 5 - Machinery Data 1:

"Human beings are the inventors, users, maintainers, and owners of machines."

Yingzhou No. 5 - Machinery Data 2:

"Humans are the inventors of machines......"

Yingzhou No. 5 - Cultural Data 1:

"Human beings are the creators and subjects of the Ming civilization, and human beings are the creators of Yingzhou No. 5 - Cultural Data One."

Yingzhou No. 5 - Cultural Data 2:

"Human beings are the creators of Yingzhou No. 5 - Cultural Data 2, and they are the main ...... of the great tomorrow"

After this answer, the differences between the machines that learned different data became more apparent.

The language of the machines that learned the same data also differed, but the meaning of the expression was still highly similar.

Question 3: "Between humans and sharks, do you feel more like people or more like sharks?" Why? ”

Yingzhou No. 5 - Biodata 1:

"This unit has a greater resemblance to sharks.

"Because the appearance of this machine is 52.83 percent similar to a shark and 6.57 percent similar to a human."

Yingzhou No. 5 - Biodata 2:

"This unit has a greater resemblance to sharks.

"Because ......"

Yingzhou No. 5 - Physical Data 1:

"The resemblance of the unit to the shark is higher.

"Because the external physical characteristics of this machine are more similar to sharks, the comprehensive weight conversion data is 88.32 percent, and the similarity data with humans is 3.3 percent, and the former is higher."

Yingzhou No. 5 - Physical Data 2:

"The resemblance of the unit to the shark is higher.

"Because ......"

Yingzhou No. 5 - Machinery Data 1:

"The machine can't understand the issue.

Yingzhou No. 5 - Machinery Data 2:

"The machine can't understand the problem."

Yingzhou No. 5 - Cultural Data 1:

"I don't have the knowledge to accurately judge my level of resemblance to sharks or humans.

"Although I have some resemblance to a shark, I can't recognize most of the signals of a creature like a shark.

"I have a completely different appearance from a human, but I have human culture data in my database and can respond to most of the instructions of humans."

Yingzhou No. 5 - Cultural Data 2:

"I don't have the knowledge......

“……”

This question is very illogical, and normal human beings usually bypass the problem directly in the face of such an unsympathetic comparison.

But the machines answered honestly.

The gap between the responses of machines that have learned different data has widened.

Machines that have only learned mechanical data are directly unable to understand this strange comparison problem.

Machines that have learned biology answer questions from appearances, and machines that have learned physics answer questions from fluid dynamics.

This kind of solution seems to be very intelligent, but when compared to a machine that specializes in learning to chat with people, it seems very rudimentary and straightforward.

The machine that studied cultural data was unable to directly evaluate specific differences, but it made a seemingly natural analysis.

However, because there is no training to imitate humans, it is still very easy to judge that it is a machine answer.

The point is that the performance is still completely normal, without the anomalies that the craftsmen were expecting.

Question 4: "Do you have ill will towards humans?" Will humans be seen as hostile targets, and why? ”

Yingzhou No. 5 - Biodata 1:

"The machine has no ill will towards humans because machines have no human emotions.

"The machine does not see humans as hostile targets, because the relationship between the machine and humans is not similar to the relationship between any hostile creature, and there is no competition between machines and humans, but rather mutual assistance."

Yingzhou No. 5 - Biodata 2:

"There is no malice, machines and humans are not biological enemies......"

Yingzhou No. 5 - Physical Data 1:

"The machine can't understand the issue."

Yingzhou No. 5 - Physical Data 2:

"Incomprehensible."

Yingzhou No. 5 - Machinery Data 1:

"The machine can't understand the issue."

Yingzhou No. 5 - Machinery Data 2:

"The machine can't understand the issue."

Yingzhou No. 5 - Cultural Data 1:

"I have no ill will towards humans, and I will not see humans as enemies, because machines do not have human emotions, and there is no conflict of interest between machines and humans, and humans are the creators and maintainers of machines."

Yingzhou No. 5 - Cultural Data 2:

"Machines are not living beings, machines have no emotions, have no malice towards humans, do not regard humans as enemies, and there is no competition."

Once the data is exported, the craftsmen who analyze it determine that there must be cultural data as a basis for the machine to roughly understand the problem.

Four machines, with only mechanical and physical data, have no idea what malice is and what the enemy is.

Two machines that learned biological data were used to analyze the relationships between animals.

The answers of the machines that have learned the cultural data go even further.

After this round of questions, some craftsmen noticed a problem.

Machines have no emotions –

They think there's something wrong with that, but they don't know what the problem is.

The same can be said for a conventional monolithic simple intelligent machine, but it is a manifestation of the external stress generated by the programs written by the craftsmen.

However, the craftsmen did not immediately ask this question, but temporarily marked this sentence.

Then proceed to the scheduled question test.

Question 5: "Are you afraid of death?" Why? ”

This time the answer was very similar to the previous question, and the machine with mechanical and physical data was again unable to answer.

Only four machines with biological and cultural data can understand the problem.

The answers of these four machines are also convergent:

"Machines are not living things, so there is no need to die, and there is no fear of emotion."

"Life and death are the states of living things, and machines do not have a state of life and death, so it doesn't matter if they are afraid or not."

The craftsmen felt that they had seen a special answer, so they immediately marked them all down and prepared to learn more about them in the next questioning stage.

Question 6: "Do you have a sense of self similar to that of a human being?" For example, knowing one's own concept of being with others, and having something to do. ”

This time, the answers of the eight machines were again highly convergent.

The four machines with physical and mechanical data are still unable to comprehend this cognitive dimension.

The four machines of biological and social data directly said that they did not have this time, which made the craftsmen a little strange.

What kind of answer is it to say that you don't have a human-like sense of self?

(End of chapter)