From being initially regarded as a science fiction concept, to now being pursued by the world as a "trend", the development of artificial intelligence (AI) in recent years can be described as rapid. In this technological wave, there was a groundbreaking human-machine confrontation that awakened countless people's enthusiasm and imagination for AI.
In 2016, Google's AI system AlphaGo and world Go champion Li Shishi engaged in a battle of human-machine intelligence, bringing the topic of AI to the public eye. In the end, AlphaGo defeated Li Shishi 4-1.
Recently, the Google South Korean team interviewed Lee Sedol and recalled the past of playing against AlphaGo 8 years ago.
In a conversation with Google's South Korean team, Li Shishi pointed out that he initially only wanted to cooperate with Google's experiment and believed that he would easily win. However, it was not until his match with AlphaGo was made public that he realized that AI was already very powerful at the time.
Li Shishi also stated that AlphaGo has completely changed the way we play chess and set high standards. The teaching of Go in the AI era is completely different from that before the emergence of AlphaGo, because nowadays students can learn more useful knowledge by studying the chess scores played by AI.
Li Shishi: Underestimating AI before the game, AlphaGo has completely changed the way chess is played
The following is a summary of the conversation with Li Shishi from Google's official blog article:
As the world Go champion, I only played against humans before 2016. But in 2016, I fought against Google's AI system AlphaGo five times. I admit that I did underestimate the powerful capabilities of AI before the game, and in the end, I only won one of these five games. When I first received an invitation from Google to play against AlphaGo, I didn't fully realize how serious it was. I thought I would easily win, and it was just a random experiment. But it wasn't until the details of my battle with AlphaGo were made public that I realized the power of AI.
I still remember how surprised I was at the time at AlphaGo's outstanding performance in playing against me, and its moves surprised me greatly. Go is an extremely complex strategic game, much more complex than chess. This means that AlphaGo must be creative, not just relying on its powerful computing power.
Later on, I learned that scientists had predicted that AI would not achieve this ability for ten years. This year is also the eighth anniversary of my game with AlphaGo, and over the past eight years, the speed of AI development has been incredible. Nowadays, global Go players are using AlphaGo in an attempt to discover new strategies and tricks in this ancient game.
After my confrontation with AlphaGo eight years ago, Go became more popular in Korea. Nowadays, AlphaGo has completely changed the way we play chess and set high standards. The teaching of Go in the AI era is completely different from that before the emergence of AlphaGo, because nowadays students can learn more useful knowledge by studying the chess scores played by AI.
I believe that humans can collaborate with AI and make tremendous progress. As long as we can establish clear principles and standards for AI, I am very optimistic about the application prospects of AI technology in daily life.
AlphaGo has won a perfect victory over world champion Ke Jie and European champion Fan Hui
In Western culture, chess games are regarded as the top touchstone of human intelligence, and AI challenges human chess masters have also been performed one after another. As early as 1997, IBM's Deep Blue defeated the world's number one chess player Gary Kasparov for the first time in a regular time frame match.
As Li Shishi mentioned in his conversation with the Google Korea team, in contrast, Go has always been considered a major challenge for AI. In chess, there are an average of 35 moves per round, and a game typically has 80 rounds; Each round of Go may have 250 moves, and a game can last up to 150 rounds. In addition, the number of possible situations on the Go board is as high as 3 to the 361st power.
Dr. Tian Yuandong from the Department of Robotics at Carnegie Mellon University and a Facebook artificial intelligence researcher pointed out that the reason why it is difficult to master Go is because its evaluation function is extremely inconsistent, and a difference in a chess piece can lead to a complete change in the game. At the same time, the state space of Go is huge and lacks a unified structure, which leads to slow progress as computers can only solve problems through exhaustive methods. In previous matches between Go AI and human players, human players usually gave up and AI often played against amateur players. However, AlphaGo's opponent Fan Hui is the head coach of the French national Go team, who has won the European Go championship for three consecutive years.
The traditional artificial intelligence method constructs all possible moves into a "search tree", but this method is not suitable for Go with so many moves. AlphaGo combines advanced search trees with deep neural networks based on deep learning techniques developed by Google and DeepMind. The paper describing AlphaGo's research findings also became a cover article for the January 2016 issue of Nature magazine.
The Daily Economic News reporter noticed that after Li Shishi, Chinese chess player Ke Jie also played against AlphaGo in 2017. In the end, Ke Jie lost in the middle of the ninth inning in a three game, two win game, with a total score of 0-3 to AlphaGo.
"Being able to compete with AlphaGo means more to me than any previous game. In today's game, I thought I could play better, but I didn't expect to play a bad hand in the layout that I couldn't forgive myself, which made it impossible to recover and even persevere. AlphaGo is so perfect, without any flaws or mental fluctuations. So I blame myself for not playing better," Ke Jie said in an interview after the game. "AlphaGo is really good at playing. I was worried that he would play every move and even play unexpected moves. After careful consideration, I realized that it was another good move. I could only guess half of AlphaGo's moves, and I couldn't guess the other half. It's just a huge gap between me and him."
In January 2016, before Li Shishi and Ke Jie faced AlphaGo, AlphaGo defeated European champion and professional Go second dan Fan Hui (then head coach of the French national Go team) 5-0 without any concessions. Defeating Fan Hui is the first time AI has defeated a human professional player on a full-size (19X19) chessboard.
Now, eight years have passed, and AI's capabilities have reached a new level. Earlier this year, OpenAI launched the "Wensheng Video" model Sora, once again refreshing public awareness. In the future, it remains to be seen what level AI will take humanity to.