Next, Zhao Zhenxue enthusiastically had some conversations with many AIs codenamed "kl". Although Zhao Zhenxue had some psychological preparations for this, she was still disappointed with the "intelligence" displayed by the AIs during the conversation. And was surprised. lingdiankans
There is a big gap between these AIs and what Zhao Zhenxue originally imagined. In fact, Zhao Zhenxue has a feeling that to a certain extent, the person who talks to him can be said to be a "class". Although through There were still many flaws in the conversation with AI, but instead of disappointing Zhao Zhenxue, these flaws made her vaguely feel that what she was facing was exactly the ideal artificial intelligence in her mind.
One of the most obvious examples is that among the AIs that talk to her, each one has its own unique style of speaking. Each AI remembers its own name and even gives itself a name other than a replacement number. Create a complete set of "resume" by yourself, including your home address, living environment, and even interesting things you experienced "when you were a child"
Yan Keshou told him that the reason why AI has current intelligence is inseparable from its "learning" ability. In fact, in the final analysis, these AIs are just some algorithm programs in computers. However, when designing, It can imitate humans' learning and "memory" habits, allowing them to learn information that helps people believe in them during experiments again and again.
To give the simplest example, in a medium such as typed dialogue, the best way to make others believe in oneself is to form a unique language style so that the testee is impressed by oneself, such as deliberately adding some common expressions In local dialects, some AIs have even learned to simulate some human temperaments, such as impatience, continuous urging, etc., using these many small details to make the other party believe that they are a human being, not an AI.
All of this is actually determined by the AI ??algorithm. Yan Keshou told Zhao Zhenxue that every time the "test ends", the AI ??will summarize where the performance is more perfect and where it is easier based on the content of the conversation. When flaws are discovered, these results will prompt AI to modify its permissions on specific concepts. For example, if a computer pretends to be from Northeast China, dialects such as "we" will appear too frequently during typing, and a large number of statistics show that , when similar words appear, the chance of being detected is smaller. Then the AI ??will subconsciously use similar words in future conversations. In the end, it can even form a speaking pattern similar to that of real Northeasterners.
During Zhao Zhenxue's several tests, the AI ??she encountered that impressed her the most was an AI codenamed kl3300. Unlike other AIs, this AI did not gain her trust by chatting with her about everyday things. Instead, he kept leading the topic to some metaphysical topics, such as philosophy, the meaning of life, etc He even mentioned that his ideal is to be a writer in the future
When discussing these topics with him, Zhao Zhenxue didn't know that the other person was an AI. After chatting with him for a long time, she was surprised to find that at some point, the other person had completely taken control of the topic. Subconsciously, she also regarded the other person as a philosophy enthusiast. When she ended the test with confidence and believed that the other person was a person, the computer unexpectedly told her that the other person was just an AI. .
Yan Keshou was standing by and explained to her the reason for her misidentification at the moment - in fact, it is very simple to put it bluntly, because when it comes to topics like philosophy, art, culture, etc., real people will answer whether the information is true or false. "Sensitivity" is actually very low. For example, in the simplest way, "What is the meaning of life?" Answers such as having fun, making money, learning cannot be wrong, because there is no relatively true answer, so people Naturally, we cannot draw a more accurate judgment based on this answer.
Therefore, compared to this robot, "loving philosophy" is also a deception strategy for it. According to the data Yan Keshou has mastered, kl3300 has the highest probability of passing the Turing test among all AIs. This in itself is already It shows that this strategy is effective.
"But what if" After Zhao Zhenxue learned about it, he asked Yan Keshou in disbelief, "If what he just said is true and what he really wants to say, if it is a What should I do if it is a real AI, not a simulation program as you call it? Is there a way for you to identify the difference between the two and whether the program is lying or derived from its true origin? Purpose?"
Yan Keshou shook his head, "We can't do the identification you mentioned."
"How could it be? Didn't you design it?"
"It's true that the algorithm and structure are designed by us," Yan Keshou said, "but AI programs are different from ordinary computer programs. In ordinary computer programs, you give an order. In case 1, do a??Case 2, do b, one by one, all the behaviors of the program are locked by the code. No matter how complex such a program is, it is just a sophisticated machine. When doing AI theoretical design, this mechanical method They were the first to be abandoned by us.
The difference between kl's algorithm and the usual algorithms is that it is a constantly evolving and expanding chaotic algorithm system. When each AI first appeared, the source program only occupied a few hundred kb in the computer, but as time went by It continuously learns through the Turing test. The more information it obtains and masters, its complexity is growing just like the development of the human brain. Even the modification of the permission values ??I just mentioned, no matter how much it is modified, The degree of this is also controlled by AI itself.
All the AI ??you see now are actually derived from the same source program. This source program is the most precious part of the entire project. All the AI ??you see now are essentially derived from this source program. It comes from "growing". However, each AI is exposed to different information during the testing process, and they will also grow into completely different intelligences. This growth process itself is too complicated, and it is impossible to achieve precise monitoring and calculation. After monitoring is implemented, it is difficult for us to predict what behavior it will do next, just as there may be many factors that affect human behavior. Although the behavior displayed by AI now is simple, it is also unpredictable. It has already Grow by itself into a chaotic whole.
It's like we can infer a baby's thoughts because the pattern is simple, cry when hungry, laugh when happy, and almost nothing else, but it is almost impossible to see an adult clearly No, Except for magic, I have to say that even if Ivan is standing here, he can't see clearly the thoughts of these AIs. This is the biggest reason why I invest so much energy in this project.
The numbered AIs you see now are actually just a drop in the ocean of growing AIs. Did you see the small compartment at the back? The area at the back is much larger than what you see now. There are Japanese The current most advanced supercomputers occupy four entire laboratories in the same location upstairs and downstairs just for the host computer room. Hundreds of trillions of source program copies are "developed" in it at the same time every day, that is, through The system continuously provides random "virtual stimuli" to allow these intelligences to initially develop. During this entire process, you can imagine the development of the adult brain germ layer, with countless neuron synapses connecting to each other, and ultimately forming an organic whole
The efficiency of this artificial simulation of "development" is actually very low. The essence of this calculation is to simulate the development process of primitive nerves over hundreds of millions of years. (unit) of computing speed, we have been working for more than four years before we found some of the most basic rules, which also summed up the model of evolution from simple neurons to intelligent structures. The results I mentioned to you , and it is exactly this point. As for the AIs you see now, they are actually a batch that we used for "temporary experiments" after this. They were just "trained" at the beginning of a few years, which is what you see. Turing test, so strictly speaking, they are only seven-month-old children. "
"Seven months?" Zhao Zhenxue was startled by this number, "Then do you mean that as these trainings deepen, they will become smarter than they are now? Just now you said you lied to 8-year-old children, Does it mean that after a while, it will really be indistinguishable from human beings, and even adults won¡¯t be able to see through it?¡±
"Are you still afraid that they will eventually be smarter than us humans and replace us?" Yan Keshou said with a smile, "You have read too many science fiction novels. In fact, there is no need to worry about this. So far, ai All the training he received was training on how to deceive. It can also be said that it was training on lying. Even if he learned it to the extreme, it was just the skill of lying to the extreme, making it impossible for people to distinguish between true and false, but if it is true, it is How smart I don't think so. Maybe today he can talk to you about philosophy and life, but AI itself cannot understand these concepts. These concepts are just tools for him to deceive, and he has higher authority in the program ¡
In order to achieve the deception effect, some computers can even argue with you about which dish is more delicious and which flavor it prefers. But you have to know that it is just a program and does not even have basic taste organs. How can it really feel this way?
In the same way, how can a computer without vision understand the concepts in human life? The simplest thing is that if a computer wants to exist alone without humans, the most basic thing is that it must have **energy to maintain them. existence, and to do this, it means that they must learn all kinds of common sense in the power industry. Otherwise, an ordinary nut may become loose, which will lead to their disappearance as a whole. In their dictionary, it is It is impossible to understand the concept of nuts because they have no eyes to see and no one tells them.With this concept, we control all their information channels. Even if they one day see the Encyclopedia Britannica online, that information can only remain on paper.
It is still too early to say that AI is intelligence. At best, it is "quasi-intelligence". They are systems that grew up in the electronic world. Compared with humans, they will always be just talented people who "can't make a living", even if they say so. The sky is falling, but it cannot conceal the nature of their ignorance. The ignorance here does not mean not knowing, but "cannot know", just like a congenitally deaf person must also be a mute. Without the premise of common feelings, There cannot be a common culture. Even if AI does eventually produce intelligence, it will definitely be a world that we cannot understand. By the same token, they cannot truly understand our world. Therefore, your worry should be said to be It¡¯s completely unnecessary. In fact, many literary themes in this area, such as alien invasions, robot rebellions, etc., can be called persecutory delusions caused by artistic needs"
"Then why do you let them learn to lie since they know it's useless?" Zhao Zhenxue said puzzled.
"This kind of learning does not mean much to AI, but it means a lot to people," Yan Keshou said. "Turing training, that is, the experiment to let AI pass the Turing test, is just the beginning. A propaganda to verify that AI does have practical functions. At the beginning of the project, I said that the ultimate purpose of this project is actually to fight against magic, or in other words, against the consciousness network. If it can really grow into ** in the future The intelligence must be an intelligence that is not controlled by the consciousness network. No matter from which perspective, this is a possible balancing method for the consciousness network and Ivan's 'dictatorship'. In the future, humans will fight against Carmel. There is always one more option among the weapons.¡±
Next, Yan Keshou took Zhao Zhenxue to visit the small cubicle inside. When entering the door, Zhao Zhenxue noticed that there were two "self-defense force members" wearing Japanese military uniforms in the cubicle. The whole room was bigger than Zhao Zhenxue. It was much bigger than what he imagined. Yan Keshou pointed to Zhao Zhenxue the chiseled ceiling tiles and the rows and rows of optical fibers connected from above. In order to prevent magic detection, the background of the entire room and even the installation The cabinets with rows and rows of motherboards were all painted white, and there were monitors at every angle of the room.
Yan Keshou told Zhao Zhenxue that her investment was only a slightly larger part of the entire project. At the same time, the Japanese government, which was interested in the entire project, also occupied a certain proportion of the entire project, such as the giant giants that needed to be used now. Computers are only allowed to be customized and purchased after being authorized by the Japanese government. At the same time, the Japanese government also undertakes security work and scientific research and technical support. Of course, in return, if the entire project produces results, the Japanese government will also will be the first beneficiary.
Therefore, even Zhao Zhenxue, who is the largest investor, still needs to be confirmed by a Japanese government official on duty when entering here to prevent any possibility of leaks. However, at present, Zhao Zhenxue is not only famous in the world, At least in Asia, it is still well-known. What's more, the four countries of China, Japan, the United States and Russia have established an offensive and defensive alliance with each other when it comes to magic issues. On anti-magic projects, the four countries keep information flowing with each other. It seems that This project belongs to Japan. In fact, it is essentially shared by the four countries. It is just that each country invests in its own strengths.
Artificial intelligence is just one of many projects in this context. If it succeeds, it may have an explosive effect similar to that of the atomic bomb. But until it succeeds, no one will understand its importance. In anti-magic In research in this field, mainstream projects are more popular and have a larger proportion of funds, such as "active laser scanning equipment" mentioned by Carmel in an internal meeting before.
"Finally, let me tell you something incredible." After taking Zhao Zhenxue for a general tour of the computer room, Yan Keshou took him back to the testing room outside. Among the people who were testing, he found a person wearing a His eyes were on the thin young man typing on the computer.
"His name is Jing Shanghai. He was a student of this university before and was also one of our earliest testers." Yan Keshou patted the other party's shoulder. After he turned around, he introduced to Zhao Zhenxue, and then he faced Zhao Zhenxue again. Mukai introduced in Japanese, "This is a friend of mine, Zhao Zhenxue, Miss Zhao."
Jing Haihai glanced at Zhao Zhenxue, said hello to her with a smile, and said a string of words in a jittery tone. Zhao Zhenxue was not familiar with Japanese, but he could still tell that the beginning of this sentence should be your best wishes. What happens later is unknown.
"You can use consciousness detection, which saves me the trouble of translating," Yan Keshou said to Zhao Zhenxue, "He just said you were very beautiful."
Zhao Zhenxue just smiled faintly.
Then Yan Keshou said to Jing Shanghai, "Is 0564 as beautiful as you imagined?"
Jing Shanghai smiled and patted Yan Keshou. It seemed that the two of them must have been familiar with each other before. "I know, you came here specifically to make fun of me."
"No, no, this friend of mine has just become interested in AI. I just want her to know more about it," Yan Keshou said, "Then I won't disturb your chat."
Because consciousness detection was used, Zhao Zhenxue understood it before Yan Keshou could speak. However, for a moment, she did not have the nerve to express her opinion. After walking out of the door, she said, "He actually fell in love with AI? And he knew it." On the premise that it¡¯s just a program? Have you watched too many anime?¡±
"Kl0564 uses a more sexy strategy, especially for some male testers. It is also the most common strategy among humans, and that is love. In their first contact test, kl0564 called himself a student of this school. , majoring in design, is the youngest sister in the family A set of very strict and mature background identity settings that can impress young men. After chatting with it once, Inoue Mistaken her for a real person and completely fell in. He just graduated this year. Although he has found a job, he will always come back here, hoping to talk to the other party.
He is not as crazy as you think. He knows very well that what he is chatting with is a program. At best, it is just a very complicated and anthropomorphic emotion. When we persuaded him, he said this: Love is just a feeling, an irrational impulse. He doesn't care whether it's true or false, and he doesn't care whether it's a real person or a program that gives him this feeling. He knows that the program is lying to him, and he himself is just enjoying it. It's just a feeling of being deceived, just like in life, everyone knows that the content of the movie is fake, but there are still so many people enjoying the deception caused by this fakeness.
In fact, others are not too worried about him. Ingami has not completely lost his mind. Otherwise, he will not be qualified to continue the test. During the test, he still strictly abides by the requirements of the Turing test. In order to comply with the elements, even if it is a test arranged by Duo Duo and kl0564, he will give what he thinks is an objective answer at the end of the test. "
"It's incredible," Zhao Zhenxue shook his head, "That is to say, in Japan, if it were in China, it would be considered abnormal."
"Don't leave the country," Yan Keshou said. "Even in Japan, many people called him a pervert after hearing about this incident. However, Ingami's mental quality seems to be pretty good. Otherwise, just now I won¡¯t joke in front of him, but I am telling you about him, not because of this thing about him, but because, due to his behavior, kl0564 has also appeared a lot. An interesting phenomenon, just like Jinghai remembered it, it seemed to have left an impression on Jinghai, so that in the next few contacts, both parties could understand each other's identity as soon as they spoke. In layman's terms, , it means that the AI ??has remembered him Why are you looking at him like this? Don't think too much. It is that it has remembered him, not fallen in love with him. All current AI behaviors are based on adaptation to the Turing test. , you have to always remember that.¡±
"I would like to make a statement here. Usually, in the Turing Test we call, the personnel are randomly selected. They are randomly selected from the school and then invited with the consent of the other party. 50% of the test personnel must be eliminated. Talking to each other, up to now, the cumulative number of effective tests has been less than one million. There are almost more than 50 computers here. On average, the tests conducted on each computer every day are only more than a hundred, and on average one is effective. The test takes about 10 minutes, and there are 5,000 candidate AIs currently used for testing. That is to say, even if it is the same computer, one person can test it continuously for more than 15 hours for more than 15 hours. The probability of encountering a famous AI twice is almost 1 in 50.
But it just so happens that among the thousands of tests that Ingami has conducted here, he has met the AI ??more than 6 times unbelievably. From the second time, the two of them can speak accurately without speaking more than three sentences. Recognizing each other's identity, from this point of view, the two of them should be considered acquainted. It is not surprising that people can recognize computers, but computers can recognize people This is indeed the first discovery of this experiment . "
{Piaotia Literature www.piaotia.net thanks all book friends for their support, your support is our greatest motivation}