The reason why artificial intelligence needs to be vigilant is that the "IQ" of artificial intelligence can be so high that human IQ is completely incomprehensible!

Imagine standing in front of a group of ants with your mobile phone, taking a photo of the ants with your mobile phone, and the camera flashes. At this time, can the ants understand your behavior? Can you understand the logic behind your actions, or purpose?

In the face of real artificial intelligence, humans play the role of ants. When you are faced with a "thing" with an IQ that is 100 million higher than yours and so complex that it is completely impossible for you to understand, are you afraid?

The answer, of course, is fear. So the question arises, why can artificial intelligence be so smart? How can humans create something that is 100 million higher than human intelligence?

Because, strictly speaking, artificial intelligence is not created by humans at all, but is trained and guided by humans.

For example, when ordinary programs make judgments, the judgment logic is written by the programmer. For example, I want to write a program now that has a function: if you give it a picture of a cat or a dog, it can determine whether the content in the picture is a cat or a dog. So, how do I write this program? Generally speaking, I will write a few conditional sentences: If the animal in this picture has a long nose, then I will let the program output: this is a dog; if the nose is short, then I will let the program output: this is a cat. I can also write some other judgment statements to assist in this judgment, so that a program is written. At this time, as a programmer, I am completely familiar with how judgment is made inside this program, so this program is completely under my control.

So, how should an artificial intelligence program be written for the same function? First of all, I have to write a training framework. This training framework can judge whether the output results are correct or wrong. As for what logic is used in the judgment process, the program has to summarize it by itself; then, I keep showing it to the program. The program keeps outputting whether it is a cat or a dog based on the picture. The training framework keeps judging whether the output results of the program are correct or wrong. The program keeps adjusting its judgment logic based on whether the results are right or wrong, and so on. Loop; Finally, after the program looked at a billion pictures day and night, its judgment was no longer wrong, and the artificial intelligence program was completed. So at this time, what logic does this artificial intelligence program use to determine whether the picture belongs to a cat or a dog? I don't know! Although I created this program, I don't know its internal operating logic at all, because its judgment logic is summarized by it after looking at a billion pictures, and its judgment logic is probably too complex for me to understand it at all (because humans cannot see a billion pictures in their lifetime). For example, it may be based on the nuances of the eyes of cats and dogs.

This is why, the creators of the program AlphaGo cannot understand AlphaGo's chess logic. No human in the world can understand AlphaGo's chess logic, because AlphaGo's chess logic is based on its own game of more than a billion games of chess, and humans cannot play so much chess in their lives.

So, since artificial intelligence is so powerful and terrible, why are we still alive and well now?

Because we can't build real artificial intelligence now. With the current human chip manufacturing capabilities, it is currently impossible to manufacture chips that can support artificial intelligence that require great computing power. All the so-called artificial intelligence humans currently play with, including ChapGPT, is like the first computer 70 years ago, a giant with particularly limited functions and takes a long time to manufacture.

The bottleneck of artificial intelligence has never been software, but hardware, and the computing power of the chip. Why can't real autonomous driving be achieved after so many years around the world? In fact, it can be achieved with current technology in software, but the current chip computing power cannot reach it. For example, using current chips to develop artificial intelligence (including autonomous driving) is like asking you to take 1980 hardware and make a game like "Red Dead Redemption 2". No matter how much your code is written, the hardware bottleneck is restricted, and it will not work even if it is written.

So, if anyone tells you that artificial intelligence (including autonomous driving) has been developed now, there are only two possibilities: either he is a traveler who stole chips from that era 30 years later; or he just wants to cheat you out of your money.

However, if Moore's Law remains effective, in less than a few decades, humans will create chips with extremely powerful computing power. By then, it will be possible for a true sense of artificial intelligence to emerge in the entire field. By then, humans Facing an "intelligence" whose overall IQ is several billion higher than its own and is so complex that it is completely incomprehensible, humans are just an ant in front of it.