Future Super Intelligent System

After four days of testing, Liu Fan showed the results to Zhao Jianguo and the six other people he brought in the meeting room. After playing the video for half an hour, Liu Fan got up and stood in front of everyone.

"The demonstration just now should be relatively clear. Now I should say that I am here to make a comment." Liu Fan is obviously much more comfortable in the face of such an occasion now. "There are two core factors in the issue of information leakage and transactions. The insider is the insider, and the second is the hacker. I can do very little about the insider. Let me first talk about how my big data processing system can effectively target hackers.

I heard that several of you here today are the heads of the technical department. You must know that the hackers who hacked the Pentagon like in the movie are one in a million hackers. More than 90% of the hackers use similar methods. , and easy to learn. To put it bluntly, it is to do the five things well, hide the IP, find the target's loopholes, implant the wooden dragon to tamper with the data, obtain the target, and quickly escape.

The most important thing they pay attention to when doing these five things is time. Under the authorization of Bureau Zhao, I studied the hacker detection system currently used by the police. There is no problem with the logic and execution, but the problem lies in the speed. This speed problem can also be divided into two parts, one is the processing speed, and the other is the immediacy problem.

So the most intuitive content I just showed is speed. Everyone should have seen the experimental comparison just now. The system I optimized can increase the data processing speed by more than three times. So how is the instant hacker alarm mechanism shown in the second part of the video realized?

This thing is still related to the particularity of my underlying algorithm. The investigation system we are using now uses trace information. It is said that more than 1,800 types of trace information have been found, right? However, under my algorithm calculation, more than 20 kinds of Trojan horse implantation logics have been found. With these implantation logics, the system can alarm when hackers are attacking. "

"Sorry to interrupt." Just then, a man raised his hand suddenly.

"What's your problem?" Facing policeman Liu Fan, he was still very polite. It didn't matter whether he was cowardly or not, the key was respect for this profession.

"I'm really sorry." The man apologized to everyone again, "The main problem has troubled me for a long time, and I just want to ask about it. Mr. Liu mentioned your algorithm, and I have done it before. Research, now many people think that your algorithm can bring humans to the era of real artificial intelligence.

My personal understanding of real artificial intelligence, or your algorithm, is to rely on the powerful processing capabilities of computers, so that computers can think in the same way as humans, and computers can break through the limitations of the human brain. Artificial intelligence realizes a kind of Similar to the existence of superhuman beings, they can deal with countless problems that humans cannot handle. Also includes what you mentioned earlier about finding order in disorder. "

Liu Fan thought about it carefully, nodded and said, "It can also be understood in this way."

"Then I have a doubt here. If computers can really anthropomorphize, then while computers can think like humans, does it mean that computers will make mistakes like humans? Throughout human history, humans often A mistake made is a blind eye. When human beings have a better understanding of the world, they will find that their previous understanding of the world was wrong.

So how do you make sure that the function or rule derived by your algorithm is correct? Although the use of face recognition is indeed very effective at present, what if the amount of data becomes larger and larger in the future and these rules are wrong? Will there be some confusion then? "

"It seems that this police officer is as inquisitive as I am." Liu Fan smiled and said, "I can answer you this way. Your argument just now is that after anthropomorphizing a computer, it may make mistakes like a human. I have also considered it. So there is a function introspection mechanism in my algorithm.

In simple terms, once the algorithm discovers the functional rules in the data,

He needs to match the function rules to all relevant data once, and the 100% correct function rules will be retained. When the data increases, each added data function rule will be matched once.

Take face recognition as an example. When more and more face information is input, this function rule has been matching. Once this function rule cannot match a certain person, even if the accuracy rate is as high as 99.99%, The system will immediately remove this function from the rule list.

That is to say, in the existing big data environment, the rules in the rule list can be guaranteed to be 100% correct, and as long as the criminals we need to find are all in this database, there will be no accidents. If an accident occurs, it is also under extreme circumstances. For example, a new face data has just entered the library, and the function does not match, and this person happens to be a criminal, and this wrong function happens to be triggered, and it happens to be photographed. down. "

"I understand." The man nodded, "If this is the case, there is no need to worry about any accidents in face recognition. With the development of technology over the years, it won't take long for us to cover almost everyone's face data.

But like this anti-hacking technology, the data is constantly expanding. Is there another possibility that the system shows that the so-called rules appear, but because the rules are wrong, there are false positives? "

"False positives are indeed a possibility." Liu Fan did not shy away, "This was raised during the face recognition project, although it is a very small probability event, but I have also prepared enough plans for this problem.

If there is only a single false positive with a small probability, it can be reported manually, and we will manually eliminate it. If there are more than two false positives within a certain period of time, the reverse deduction algorithm stripping program I set will activate, perform self-inspection and retain the traditional technology.

But even if we only keep the traditional technology, we still have advantages in speed and accuracy. This has also been shown before. "

"But what if there are a large number of false positives at the same time?"

"Technology is just an auxiliary tool." At this moment, Zhao Jianguo suddenly spoke, his deep voice carried an inexplicable deterrent force.

"Zhao Bureau, I just want to ask clearly about possible hidden dangers."

"I understand what you mean." Liu Fan hurriedly answered with a smile to relieve the suddenly depressed atmosphere, "However, I have actually taken corresponding measures to address your concerns, such as the weight threshold response mechanism.

Among all the logics, there is actually a proportion distribution of the frequency of use. For example, the proportion of the use of the A logic is 80%, and the proportion of the use of the B logic may be 10%. An alarm triggered by a logic will also have a proportion distribution.

As I mentioned before in the implementation of face recognition, dynamic features are still features, and my technology will still fail if it is too extreme. The same is true for this hacker tracking. If the Trojan horse implantation method used by hackers just avoids these logics, then the advantages of this system in this regard will be gone.

In addition, the crime itself is a small probability event, so we can make a time difference. Let's assume that the average police dispatch time is half an hour. If there are two false alarms within an hour, the system will automatically separate it, so the one you just mentioned This is an extreme phenomenon. First, you have dozens or even hundreds of alarms within an hour. At this time, the system will display abnormalities. According to the weight ratio, 10% of the alarms with a high proportion will be prioritized, and the remaining 90% will be automatically activated. Stripping programs, but not doing nothing, just moving to traditional tracking programs.

If there is no false alarm upload after a certain time, subsequent system alarms will continue to be triggered according to the settings, which avoids wasting resources and energy of the police force due to a large number of errors in the system within a short period of time.

And I have set dozens of plans like this in the system. To put it bluntly, your worry must be because these rules are derived by algorithms, so you think it may be uncontrollable. I have considered these issues. Looking back We can chat privately. "

When Liu Fan said this, the previous man nodded in satisfaction, "Please Mr. Liu, I feel more at ease when you say that."

While the bosses were not paying attention, Fang Hua suddenly poked Wan Cuicui.

Wan Cuicui was taken aback, and looked at Fang Hua slightly angrily, "What are you doing?"

Fang Hua pointed to Wan Cuicui's eyes, smiled and said, "There are stars in your eyes."

Wan Cuicui: "..."