The match between Lee Sedol 9 and AlphaGo demonstrated that AI can go beyond mere computation and challenge human intuition. AlphaGo was able to use deep learning and CNN to learn the patterns of the game of Go and make intuitive judgments like a human, suggesting the potential for AI to be applied in a wider range of fields. As deep learning technology becomes more advanced, AI will play a larger role in our daily lives, including prediction, judgment, and mistake prevention, revolutionizing our lives.
One of the biggest stories of 2016 was the epic battle between Lee Sedol 9 and Google DeepMind’s AlphaGo. Lee Sedol 9 is considered the strongest Go player in the world, and Google DeepMind is a subsidiary of Google, one of the most innovative IT companies in the world. The matchup was more than just a game of Go, it was an intriguing and traditional “computer vs. human” confrontation that generated a lot of buzz. The match was seen as an opportunity to test the limits of existing artificial intelligence and raised questions about the technology of both sides and the limits of human capabilities.
The match lasted five games, and in the end, AlphaGo defeated Lee Sedol 9 by an overwhelming score of 4:1. After the game, people believed that it was virtually impossible for a human to beat a computer that could quickly calculate the number of moves in every case. Even though Lee Sedol IX only won one game, many praised him as a monumental challenger. If AlphaGo had simply relied on counting all the cases, its victory would have been the result of a computer with more computational power than before. However, AlphaGo’s victory did not come from hardware improvements, but rather from a breakthrough in its internal algorithms, which demonstrated that its understanding of the game of Go was far beyond that of existing AI. AlphaGo’s algorithms have since been applied to many fields and have the potential to have a profound impact on our lives.
Traditional algorithms: The Minimax algorithm and its limitations
To understand what makes AlphaGo so special, we first need to understand the Minimax algorithm used by traditional board game AIs. AIs have long been able to beat humans in games such as backgammon and chess, and it’s been decades since a computer beat a world champion in chess. The Minimax algorithm used at the time was based on the concept of “choose the best possible move given all possible moves. As the name suggests, the Minimax algorithm tries to find the best possible outcome for itself against the worst possible outcome for its opponent. In the case of chess, the number of possible moves is relatively small, given the limited 8×8 area of the board, so if you have enough computing power, you can look as far ahead as possible.
Go, however, is a different story. The board is 19×19, and there are few restrictions on where the stones can be placed, resulting in hundreds of millions and trillions of possibilities from the first number. If we tried to compute all of them with the Minimax algorithm, we would have to analyze about 22 trillion moves, just for the six moves ahead, and it would take 700,000 years to complete all of them, assuming we compute one move per second. Even chess was not able to analyze all the cases using simple math, so a number of shortcuts were used, an approach that is no longer valid for complex games like Go.
AlphaGo’s innovation: Intuitive judgment through deep learning
The reason AlphaGo was able to win against humans in Go is because it introduced a new approach to overcome the Minimax algorithm. Because it’s so difficult to choose the best move in Go, AlphaGo used judgment, similar to human intuition, to dramatically reduce the computational effort. The key to this new approach came from a technique called deep learning. Deep learning is an artificial intelligence methodology that mimics the neural networks of the human brain, allowing it to solve complex problems through learning and intuition.
Deep learning works by processing data fed into an artificial neural network through multiple layers of neurons. In this process, rather than simply calculating the number of moves in every case, AlphaGo uses learned patterns to predict which moves have a high probability of winning in a particular position and places them accordingly. Gradient descent, an important learning technique in deep learning, allows AI to make more accurate judgments as it reduces its errors through repeated training. These deep learning techniques have the advantage of being parallelizable because all of the computations can be performed on a matrix basis, which allows them to utilize GPUs to process computations at lightning speed.
Introducing CNN: Recognizing the checkerboard as an image
One of the key deep learning techniques used by AlphaGo is the convolutional neural network (CNN). CNNs are algorithms that originally showed great promise in image recognition and classification, and they excel at recognizing patterns and extracting features from images. CNNs work by analyzing each pixel in an image to classify the shape and features of the objects in it. AlphaGo was trained by replacing a checkerboard with pixel data from a CNN to image the placement and shape of the checkerboard stones. Based on the CNN’s features, AlphaGo was able to analyze each move on the checkerboard as if it were a color pattern in an image and calculate the probability of winning the next move. This enabled AlphaGo to make highly intuitive moves through learning and pattern recognition, rather than simple computation, and to make decisions on a different level than traditional AI.
AlphaGo has a self-learning capability that differentiates it from existing AIs, allowing it to develop human-like intuitive judgment through repeated training with its own notation. This means that deep learning technology has given AI the ability to learn on its own, allowing it to go beyond simply solving a given problem and perform tasks that require complex thought processes. This self-learning capability shows that AI is increasingly mimicking human thinking and is close to being able to generate new strategies on its own.
The future of deep learning: AI in our lives
The impact that deep learning techniques will have on our daily lives is beyond imagination. Even now, AI is helping us in a variety of areas, including photo categorization, language translation, and speech recognition, by mimicking human thinking and judgment. In the future, as deep learning becomes more widespread and AI is able to learn a person’s behavior and patterns, it will become more than just an assistant; it will be able to predict a person’s behavior, reduce mistakes, and make decisions for them. For example, when AI can learn a person’s health status and daily habits to proactively detect risk factors, or predict traffic conditions in real time to guide the best route, our lives will become more convenient and safer.
Furthermore, as AI becomes more and more involved in the lives of real humans, we will be able to experience new forms of life that we have never even imagined before. In this sense, the 2016 AlphaGo match between Lee Sedol and Lee Sedol was more than just a game of Go; it was the beginning of the blurring of the boundaries between humans and AI, and a foreshadowing of our future with AI.
