Wei Dongsheng: "The 42.2% win rate is the absolute win rate, which is limited to the absolute win rate of the limited cognition of Rotten Ke Go. Assuming that the opponent's strength is on par with it, the opponent's winning rate can only increase to 42.2% after the 78th move. That is to say, the fatal mistake of the 78th hand has limited impact on the overall situation, and even if the opponent changes to Rotten Ke Go, it is difficult to turn defeat into victory. Perhaps Rotten Ke Go judged like this: the previous dozens of sub-layouts, it has quietly reaped the potential advantages of at least 9 goals without any human chess player being able to understand it. ”

"What is Lee Sedol's true winning rate?"

"Ranke Go created several main data models for Lee Sedol: the Lee Sedol data model in 2010, the Lee Sedol data model from 2007 to 2010, the Lee Sedol data model at its peak, the enhanced Lee Sedol data model, the dynamic Lee Sedol data model, and so on."

"The 2010 Lee Sedol model, Ranco Go includes Lee Sedol's chess games from May 2009 to May 2010, and compares it with Lee Sedol's general database to analyze Lee Sedol's recent state in detail. Perhaps affected by last year's suspension turmoil, Rotten Ke Weiqi judged that Lee Sedol's chess strength has declined significantly recently, and he is quite out of shape. Among the four major models, Lee Sedol's model in 2010 had the lowest winning percentage, with only 1.3 percent after the 78th wrong move of Ranke. ”

"The period from 2007 to 2010 was the peak of Lee Sedol's domination of the world of Go, and Lee Sedol's data model during this period was relatively strong. After making a 78th wrong move, its win rate increased to 4.7%. ”

"At his peak, Lee Sedol's data model only collected high-quality chess games recognized by the Go world, and virtually shaped a Lee Sedol who could play high-quality chess games every time. After 78 wrong moves, its win rate increased to 11.9%. ”

"The enhanced version of Lee Sedol's data model, big data analysis of Lee Sedol's chess power evolutionary trajectory, change those imperfect Go concepts that are easy to be slain, and predict that he can make a breakthrough in some aspects. After making the 78th wrong move, its win rate increased to 18.6%. ”

"The dynamic version of Lee Sedol's data model, Rotten Ke Go frequently compares and analyzes the current chess game with Lee Sedol's previous chess games, and predicts Lee Sedol's next move in reality. If Lee Sedol wants to win with a past formula that is effective in experience but has loopholes, Rotten Go will quietly set up an ambush while ensuring the winning rate; If Lee Sedol has a self-breakthrough during the game, Rotten Go will immediately record those benign changes, and call on other computing power resources to calculate the impact of benign changes on Lee Sedol's data model. ”

"The peak data model helps Rotco Go understand Lee Sedol in the past, the enhanced data model helps Rotco Go understand Lee Sedol in the future, and the dynamic data model helps Rotco Go understand Lee Sedol in the present. In the seven-game game in the eyes of Rotten Ke Go, in the first and second games, Lee Sedol was not in the state at all, especially in the second game, and the state sank to the bottom; In the third, fourth, and fifth games, Lee Sedol gradually recovered and returned to his peak; In the sixth game, Lee Sedol was in a sluggish state; In the seventh game, Lee Sedol was in excellent condition, and he had some taste of self-breakthrough. ”

"Wait, wait."

"Rotten Ke Go knows Lee Sedol better than we thought."

"Some people may ask, is it so aimed at Lee Sedol, is it still a Go program? Do I need to readjust the data model against other players? If you change a player, you need to readjust it again, and Rotten Go is too mechanically stiff. ”

"Answer, of course not."

"If it is rigid to that extent, how can Spring and Autumn Search have the face to call itself artificial intelligence?"

"It's an enhanced version of the Lee Sedol data model, and it's a dynamic version of the Lee Sedol model, which sounds complicated. In fact, in actual operation, Ranco Go is very simple and fast, and all data models do not need to be manually entered and established. ”

"Ranco Go has a well-established data model building method, which can generate a series of data models in real time by simply scanning Lee Sedol's past chess games. If you authorize Ranco Go to connect to the Go database on the Internet, it can also retrieve Lee Sedol's information with keywords such as "Lee Sedol + Go", and automatically extract past chess scores. That is to say, even if the opponent is replaced by players such as Lee Chang-ho, Gu Li, Kong Jie, etc., Ranke Go can quickly generate a series of data models such as enhanced Lee Chang Ho and enhanced Gu Li. ”

"We say that Ranco Go is artificial intelligence, not only because it can win against Lee Sedol and conquer the myth of Go, but also because it can automatically collect data, process data, automatically establish data models, and efficiently and purposefully analyze the opponent's Go concept and Go thinking."

"When we human chess players are looking for loopholes in Rotten Go, Rotten Go also uses cutting-edge technologies such as big data analysis to find loopholes in top players such as Lee Sedol."

"It is impossible for Rotten Go to be perfect without loopholes, but how can we humans be perfect enough to have no loopholes?"

"As long as Rotten Go can beat Lee Sedol better than Rotten Go, it will win!"

The Rotten Ke Go searched in the Spring and Autumn Period and the Google alphago in the memory of thirty years, although they are both Go programs, are completely different in nature.

Although he couldn't compete on the same stage, Wei Dongsheng believed that the Bad Go he programmed himself could easily beat Ke Jie's version of AlphaGo in May 2017, because the two Go programs were not at the same level at all. Wei Dongsheng overlooks the concept of deep learning with the computer intelligent life knowledge system, and his understanding of deep learning is far beyond that of the Google team. Even if he does not use the black technology of the computer intelligent life knowledge system, his practical ability is ten times stronger than that of the Google team.

And in addition to the gap in chess power, the artificial intelligence statement of Rotten Ke Go is more worthy of its name.

When you're looking for its vulnerabilities, it's also looking for your vulnerabilities.

Hearing Wei Dongsheng's somewhat cautionary words, the audience couldn't help but be shocked.

Some people are slow to wake up to the cruel truth: we think too much about intelligence, and we set the standard for artificial intelligence too high.

Perhaps, the threshold for wisdom is not as high as imagined.

Perhaps, the word wisdom is not so philosophical.

Is it an act of intelligence for a computer program to look for vulnerabilities in human chess players on its own?

Deep learning and other related technologies are used to improve their ability to judge the value of their own value, big data analysis and other related technologies are used to observe and find the weaknesses of the target, and the new computer programs represented by the Rotco Go program have been able to carry out offensive and defensive operations with humans to a certain extent. It's so weak, but it brings a possibility for the future.

As Wei Dongsheng showed, Rotten Go has established countless data models to analyze Lee Sedol's weaknesses, and can even predict the direction of Lee Sedol's self-breakthrough, how is Lee Sedol going to win it?

Some people may say that Lee Sedol's silver lining is to constantly seek innovation.

Lee Sedol breaks through the limit and reaches the unknown territory of Rotten Ke Go, and it is powerless.

However, this naïve idea is clearly somewhat optimistic.

Rotten Ke Go can frantically train and play 200 million games of chess in just half a year, and can analyze more than 700,000 games of chess for Lee Sedol in just half a month. The efficiency of Rotten Ke Go is so high, the speed at which it adapts to the new game is bound to be ten, hundred, thousand, and ten thousand times faster than Lee Sedol's continuous self-breakthrough.

Think about it.

Lee Sedol worked hard for a year and finally ushered in a self-breakthrough. Ke Langke Go quickly followed him, efficiently deduced hundreds of thousands of chess games and analyzed the value of Lee Sedol's self-breakthrough, and integrated Lee Sedol's self-breakthrough achievements into its own value network system in just one day.

The horizon then expands to the entire Go world.

Rotten Go closely follows all top players such as Lee Sedol, Lee Chang-ho, Gu Li, Kong Jie, Park Yeon-hwan, Ke Jie, etc., and each top player's interpretation of the unknown field of Go can be included in its value network system in time. Lee Sedol, Park Yeon-hwan, Ke Jie and other chess players want to fight against Ranco Go, as if it were a struggle between individuals and collectives, not only to win the deep learning results of Ranco Go's self-play, but also to win the contribution of the entire Go community to Go in the century.

It learns on its own and is better at learning from humans.

While humans are decentralized, AI is one.

This is the advantage of artificial intelligence, and it is also the horror of artificial intelligence.

If artificial intelligence can learn all human knowledge, even if it "lacks" the ability to learn and innovate on its own, it can become a human collective like the fantasy Gaia consciousness and Alaya consciousness. Unless the human race invents a new weapon that can destroy the AI in a very short time, and destroys the AI in waves before the AI understands, learns, and absorbs the new weapon, there is little hope for humanity to defeat the AI with an infinite numerical advantage.

Such a hypothetical assumption may still be far away.

However, the crisis has already begun to grow.

If you change the scene and logic, can it find the "weakness" of a specific program by big data analysis program source code?

The code looks profound, but it's actually very simple.

Programmers should be well aware of this, after all, working with code is much easier than dealing with people.

Especially those non-innovative codes, which are simply as manual labor as farmers farming, the word code farmer is really both vivid and appropriate. Theoretically speaking, if Ranco Go can discover the unknown Go concept of human chess players through deep learning, it will inevitably be able to analyze the source code of a specific program by increasing data analysis in deep learning, quickly find bugs and vulnerabilities in a small way, and optimize the execution efficiency of the program in a large way.

Just like the myth of Go, the myth that humans can write programs will eventually be shattered.

Of course, Rotco Go will certainly not be able to update its source code by itself.

But what about a little more refinement?

What about the emergence of new theories like deep learning and big data analysis?