Chapter 142 The Puppy’s Tail
Hearing this question, Li Jingzhe couldn’t help but be startled:
“Well, Grandpa Huang, I don’t know much about Go, so it’s hard for me to judge this. "
"No problem, just tell me your first impression!" Huang Baoyang replied.
Seeing the other party's insistence, Li Jingzhe prepared to answer.
Originally, according to his thinking, in today's era of rapid development of information technology, once computers find the secrets and surpass some traditional human skills, there will definitely be no possibility of being surpassed again. So he subconsciously wanted to answer that OMEGAGO must be stronger.
But when the words came to his lips, Li Jingzhe suddenly felt that something was wrong.
Since the other party asked such a question, the answer should never be taken for granted.
Thinking about it, in the five games of chess between OMEGAGO and Cui Chengshi, the opponent also made a so-called "god move" (of course, as a reserve force in the industry, Li Jingzhe also knew that this was just a software bug. ), lost a game, and with the further development of Go AI technology, human chess players can theoretically "learn from the foreigners and use their skills to defeat the foreigners"...
Hiss... no way?
Li Jingzhe couldn't help but frown slightly and asked tentatively:
"Is this guy named Lei Qi better than OMEGAGO?"
Huang Baoyang smiled even happier: " Jingzhe, it seems that you are not only smart, but also very clever.
"Actually... this issue is very controversial. But I have to say that this issue is debatable and not outrageous. ”
Not outrageous...
Although Huang Baoyang did not give an affirmative answer, his expression still couldn't help but move when he heard these four words.
Huang Baoyang continued:
“Actually, this issue was basically not controversial a year ago.
“‘How can humans beat computers in playing Go? Well' - this is the common conclusion of people, just like countless people in the past firmly believed that no matter how powerful a computer is, it will never be able to beat anyone in Go. ”
Hearing this, Li Jingzhe smelled it.
"You mean...in the past year, the level of human Go has improved a lot, right?"
"Yes." Huang Baoyang nodded: "The top Go players, they have improved very quickly. Soon.
“A year ago, most people believed that although humans can now gain a certain advantage over OMEGAGO in the layout stage by learning the most advanced AI formulas,
< br>“But once entering the middle game, as the situation becomes complicated, these many advantages in the layout stage will be quickly recovered and overtaken by OMEGAGO relying on its powerful computing advantages. Considering that the stability of humans is far weaker than that of machines, humans will In the final analysis, it is nothing more than AI - even the first generation AI. "But now... whether OMEGAGO can stably recover from its layout disadvantage in the mid-game is a big question mark. ”
Speaking of this, Huang Baoyang casually handed over his mobile phone.
I saw that the current game of chess holy war was also on his mobile phone, but it had various curves on it.
Li Jingzhe took the phone and asked: "Grandpa Huang, this is...?"
"The AI coincidence rate chart of the two chess players, now the VVVIP of Yiguo has You can take a look at this service. Ah, you should understand the terms first choice and top three, right?"
"You know."
This kind of common sense that favors AI is not something that Li Jingzhe, who is engaged in computer science, does not know.
But he was still surprised when he saw this data chart.
This game of chess is coming to an end. The first choice rate of white is as high as 70%, and the first three selection rates are astonishingly 94%!
This is the overall AI compliance rate! The top three choices can actually reach 94%?
Li Jingzhe immediately raised his head and looked at Huang Baoyang. He was about to speak, but Huang Baoyang spoke first.
"The specific data analysis in the lower right corner also includes the specific layout, mid-game and officials...ah, it is the analysis of the coincidence rate in the final stage."
Li Jingzhe hurriedly went to check, but he didn't know this specific data It's even more amazing at first glance.
Originally, he thought that the first-choice rate data was so high because the players "memorized the music" and relied on step-by-step selection during the layout stage to increase the total data.
But he never expected that although the first three selection rates of the white chess layout stage in this game were as high as 98%, the first selection rate was only 63%!
On the contrary, in the middle game, although the first three selection rate dropped to 93%, the first selection rate increased to 67%!
The final result is even more outrageous, with a 98% selection rate for the first three selections and a high 76% selection rate for the first selection! "Grandpa Huang, this... why is the first pick rate of human chess players in the middle game higher than the layout!" Li Jingzhe's voice was slightly trembling with surprise.
“Actually, this is not necessarily true. Most of the time, the layout should be slightly higher.”
Huang Baoyang explained with a smile:
“But, Because the layout actually has a lot of options, and the first three choices are very close. As a human chess player, many times the action will not be completely followed by the AI.
"On the contrary, in the middle game, because the battle gradually unfolds, chess players must try their best to grasp the most important points at the moment, so sometimes there are some chess games where the first pick rate in the middle game is higher than the layout."
"But...the first three selection rates in the middle game are 93%, and the first selection rate is 67%?" Li Jingzhe was still surprised: "Is this really possible for people?"
Although he can't play Go, but after all, there is a chess fan in his family who has influenced him. He still knows that Go is different from chess and chess. The middle game is complicated and there are many options.
As a result, under such circumstances, humans actually made 93% of the top three choices and 67% of the first choice? !
You know, with the development of AI level in recent years, if the current Go AI wants to play the first generation OMEGAGO, it will definitely be a merciless massacre!
And if one person can achieve such a high matching rate of modern AI even in the middle game. Then based on the first version of OMEGAGO that year, it might be really difficult to recover the disadvantage in the layout in the mid-game stage just by relying on computing power!
"How is it? Our professional chess players are working hard, right?"
Just when Li Jingzhe had not recovered from the surprise, Huang Baoyang said proudly.
“Ah, of course, this game should be regarded as a game in which Xiaolei played quite well, but this is no accident.
“Just like the top chess player on the stage As I just said, in the last game of the last Nongshim Cup, South Korea's Ryu Se-hyun performed extremely well, almost exactly the same as today's situation.
"It was after that game that some people on the Go forum began to doubt whether the top human players could compete with the first version of OMEGAGO."
"But...how do people do this? Yeah!" Li Jingzhe still couldn't believe this.
As a talented person who is interested in front-line computer science research, Li Jingzhe naturally knows that modern Go AI represented by OMEGAGO all use deep learning to ensure effective calculations by reasonably reducing redundant calculations in chess. It has powerful computing power that is absolutely unmatched by humans.
But now, humans can actually compete with modern deep learning AI? !
You must know that the complete version of "Snowfish", the most powerful national elephant computer software before the emergence of modern AI, has almost never been defeated by humans!
But "Yueyu" learned chess for 4 hours using the OMEGAZERO software derived from OMEGAGO, and was brutally beaten by the opponent with 28 wins, 72 draws, and no defeat.
OMEGAZERO's first record is 25 wins and 25 draws. This winning rate is simply astonishing in a country where the final solution is generally a draw!
So, it can basically be said that in chess games, humans are crushed by dimensionality reduction in the face of deep learning AI.
But as a result, seven years after Go players were mercilessly crushed by deep learning AI, they actually started to come into contact with the deep learning AI that first crushed them? !
After fully understanding this fact, Li Jingzhe was in a daze for a moment.
The reason why OMEGA software used Go as its deep learning AI debut was largely to catch people's attention - what could be better than conquering "the most difficult chess game in the world" and "the last dignity of human wisdom" What about the eye-catching ones?
However, when the OMEGAGO software easily shattered the pride of countless Go players for thousands of years, it did not stay here any longer.
After all, we are a business team. How much money can we make from a niche board game like Go?
Since then, the top commercial AI has basically paid little attention to this niche project that is only popular in the three East Asian countries. The most cutting-edge AI researchers have also been too lazy to pay attention to the Go circle that they have turned into a pot of porridge. status.
Only some AI companies in China, South Korea, and Japan are iterating AI development according to the needs of the Go circle. However, this development is basically limited to iteration and optimization, and there has never been a qualitative change in the core.
But who would have thought that after Go completely faded away from cutting-edge AI research, a new and jaw-dropping situation appeared!
Faced with Li Jingzhe’s question, Huang Baoyang raised his eyebrows:
“Jingzhe, I, an old antique, really don’t know this!
“But… …Maybe this is why we still love Go as much as ever? ”
Just as Huang Baoyang finished these words, there was suddenly a new noise on the stage.
At the 286th hand of the game, Black Lin Ruixin surrendered!
【Plot related】
Well, the third chess holy war still has some finishing touches after the game, and I will start writing the plot related to the stage match tomorrow.
【Realistic Situation】
Regarding the plot of this chapter, I feel that it needs to be explained. This is indeed based on reality.
Whether Shen Tiandi, the current No. 1 player in Go, can beat the original AlphaGo, this topic is still controversial, but the mainstream view is that Shen has the advantage without being sick, or is good at winning. No one thinks that Shen Tiandi can't do it just because the first-generation dog is a deep learning AI - even if the calculation amount is very different between the two.
Simply put, in Go, humans are indeed touching the tail of puppies.
(End of this chapter)