For the first time since AlphaGo debuted four years ago, AI has appeared on the go circuit as a cheating tool. According tomedia, recently, the South Korean district court accepted a Go championship AI cheating case, the relevant personnel were sentenced to one year in prison. On January 14 this year, at a Game of Chess match in South Korea, a chess player’s unusual behavior caught the attention of the referee, and then the examination revealed that the player had hidden wireless headphones, micro-cameras and other illegal electronic products.
South Korean chess players have been convicted of cheating with AI.
It turns out that chess players are using these electronic products to communicate with their partners, trying to use AI to cheat. The miniature camera on its buckle captures chess board information in real time, and the companion responsible for receiving it uses AI technology to analyze the board situation and deliver feedback to the player’s wireless headset.
It is understood that the AI technology used by the chess playerist is the AI project Leela Zero, developed by Belgian programmer Gian-Carlo Pascutto (GCP), which is one of the few open source projects in the field of Go and can be downloaded by all on Github.
As the referee found it timely, the player and his partner’s cheating was not successful. However, the Korean Chess Academy considered that the two men’s behavior was bad, committed a “business obstruction crime”, and after the police, also commissioned lawyers to draw up an indictment, the two people were criminally charged.
Recently, the Eastern District Court of South Korea issued a final judgment: suspect A for the purpose of professional segment, and accomplice B after careful planning, the use of intelligent technology to violate the rules of the game, seriously undermined the fairness of the game, fair, very bad nature. It was found that the suspect A was sentenced to one year’s imprisonment, his accomplice B one year’s imprisonment, suspended for one year, and 120 hours of social services. “
So far, the disgraceful cheating incident is a due penalty. However, behind the incident, there are netizens lamented, now any AI can beat the professional players?
There are netizens joked that artificial intelligence has begun to enslave human beings!
When it comes to AI beating humans in Go, most people probably think of Google’s AlphaGo first.
In 2016, AlphaGo beat Li Shishi, the world’s top professional, 4-1.
In 2017, AlphaGo Master again won 3-0 against China’s Go A-League champion Ke Jie.
In fact, since then, AI’s technology research and development in the field of Go has never stopped, and after four years of development, it has been far more than alphaGo to beat the human world champion level of AI.
Leela Zero, the cheating tool, has a good record.
Friends familiar with Go may understand that Leela Zero is not an unknown AI chess player. It has won the 32nd, 38th and 42nd KGS Computer Go Competition, and won the silver medal in the 9th Go and the 19th Round bronze medal in the 2008 Computer Olympiad.
Leela Zero is GCP’s extension to Alpha Zero, Google’s most powerful open source project. Like Alpha Zero’s development path, it does not train entirely from scratch without any human knowledge. At the same time, it uses distributed computing, which is transmitted to the server for training through a computer-generated set-up of chess games, to support the Leela Zero project with the power of volunteers around the world.
It is understood that Leela Zero’s chess power is enhanced by the constant change of neural network weights (weight: a series of parameters that characterize the internal connection of neural networks). When the server trains the chess spectrum, a new network weight will be generated, the new weight will be against the previous strongest weight match, to test the chess force, if it wins more than 55% in 400 games will be updated to the current strongest weight.
Nearly 600 people now train for Leela Zero every day, and in the last six months, Leela Zero has played 7 million games of her own game and gone through 128 weight changes.
In addition, Leela-Zero has become the best chess player in mankind with her excellent chess skills. Its unconventional style of play, unexpected layout, often can bring the breakthrough of human inspiration. So many Go enthusiasts like to improve their chess skills by playing with it or re-training.
One Weibo user also shared his feelings about watching Leela-Zero’s war games, especially pleasing.
Chinese chess man Ke Jie also sent a message saying: Leela Zero training is very good, I also want to save money to buy a supercomputer to run AI, now learn programming is still too late?
The Leela Zero project, which has been open source at Github since 2017, has now harvested 4.1k stars. Github open source address, link contains detailed download installation process, follow the prompt after the completion of the operation, you can play with AI chess players online game.
AI crushes the human, light and relaxed.
As Ke Jie said, there are now a lot of AI players at home and abroad, and there are no few who can beat world-class professionalplayers. Such as the domestic classic chess player Tencent is absolutely unruly.
In most recent 2019, he won the World Smart Go Open for the fourth time. It is tencent AI Lab’s self-developed Go AI. Since the first release in 2016, more than 100 professional chess players have been beaten in major tournaments.
It’s worth noting, however, that the AI Go winning general lost to his brother Phoenix Go in the 2018 World AI Go Competition.
PhoenixGo, commonly known as “Golden Hair” (named after the avatar of the game), was developed by WeChat’s translation team to develop the artificial intelligence Go program. In this world-class AI Go competition, PhoenixGo won the title, defeating the first-class AI Go players from China, Japan, Korea, Europe and the United States and other countries and regions. Such as Leela Zero, TSGo, Dol Baram (Stone Cyclone), Golois, HEROZ Kishi, Baduki, etc. (Github open source address)
Speaking of the domestic more excellent AI chess players, there is a must mention, it is “Star Line Go.” In 2018, “Berry Gene Cup” World Artificial Intelligence Go Competition man-machine war, Star Group Go black 145 hands to win Ke Jie, and then, in the “Let the First Thirty Chess” challenge, the round against professional players When Yue, Jiang Weijie, Cui Zhehan, Yuan Yu, Zhou Junxun and other world champions, finally with 40:1 record to win the title, the winning rate reached 97.6%.
What’s more, Star-Spangdry Go is a pure lysing AI. Most of the other AI players are from Google Papers, Starform is originally developed its own research and development program, and its strategy is significantly different from Google’s.
Google’s algorithm is more inclined to win the ratio, in order to maintain the overall situation often choose temporary concessions, and the Strategy of the Star Front is a strong attack, put humans on the ground friction, can advance never give in.
It is understood that the predecessor of the star Go is Tsinghua University’s “God operator”, after deep visitor technology completed the follow-up technical upgrade. The company’s chairman, Dr. 涬, has worked for many years in the fields of big data, deep learning and artificial intelligence, and is an important promoter of artificial intelligence applications in the Go industry.
The best of all the go AI spent abroad, in addition to Google’s Alpha series, is probably Facebook’s ELF OpenGo. The AI Go program will also open source at Github in 2018.
The team said the open source is intended to inspire enthusiasts to think about the new application and research direction of the technology.
In terms of performance, ELF OpenGo played against Lela Zero in a match that ended in a 200-0 victory. It also set a 14-0 record against world-class professionals. According to officials, OpenGo uses a single GPU for 50 seconds of search time per step (80,000 searches per step), and human chess players don’t limit the time they spend thinking.
Switch from Go to RTS.
It’s been a very easy thing to see here that AI has crushed the top players in the world of chess. However, the purpose of AI research and development is not here, Go because of the complexity of the rules of the game and strategy by researchers as the best test field for AI training, on the one hand, AI through the game against humans, AI can quickly improve its deep learning and thinking ability. On the other hand, its comprehensive big data and combat layout, can also bring more enlightening thinking to human beings. It can be said that the two in the game with each other and grow together.
In addition, the complexity of Go is no longer sufficient for AI. In recent years, AI’s training scene has gradually shifted from Go go and Texas Holdis to more complex RTS instant strategy games. As a classic RTS game, StarCraft has become the research and development target of many teams because of its characteristics such as incomplete information, complex combat space and millisecond decision-making.
At present, the artificial intelligence company led by Google has also made a breakthrough in this area of exploration. Last October, Google published its latest paper, Nature, which noted that its research and development of AlphaStar surpassed 99.8 percent of active players on the list, reached the top level of the StarCraft 2 Human Battle Ladder, and gave video of the battle.
In addition, a domestic artificial intelligence company Kaiyuan World, its research and development “Interstellar AI” in the recent challenge with a 2-0 victory over the world’s top professional players. The research and development team said that since its inception around StarCraft around the development of star AI, the purpose of the game industry, to provide players with a better gaming experience, and the second is through The Training of AI intelligence, knock on the door to universal artificial intelligence (AGI).