This blog post delves into the evolution of artificial intelligence (AI) and its capabilities in creativity, particularly through the lens of games like chess and Go. It discusses historical milestones, the differences between brute-force algorithms and machine learning, and highlights significant achievements such as IBM's Deep Blue and Google's AlphaGo, ultimately questioning the nature of creativity in AI.
In recent times, artificial intelligence (AI) has been a hot topic, often associated with various threats, including the potential disappearance of numerous jobs. While it is true that AI can perform many tasks more efficiently than humans, such as processing large datasets and performing complex calculations, these machines primarily imitate human actions rather than create something new. Creativity, whether in art or science, has long been considered a uniquely human trait. However, as we will explore, machines are increasingly demonstrating creative capabilities, particularly in the realm of gaming.
Since the inception of AI in the 1950s, games have served as a benchmark for comparing human intelligence with machine intelligence. Chess, often regarded as one of the most intellectually demanding games, has been a focal point in this exploration. The fascination with machines playing chess dates back to the late 18th century when Wolfgang von Kempelen showcased a mechanical Turk that could play chess against humans, often defeating them. However, this was merely an illusion, as a human was hidden inside the machine, controlling the moves.
The real journey into AI chess began after World War II, with pioneers like Alan Turing developing algorithms capable of playing the game. By Christmas 1976, the first commercial chess program, "My Chess," was available, and as computing power advanced, so did the strength of these programs. A pivotal moment occurred in 1997 when IBM's Deep Blue defeated world champion Garry Kasparov in a six-game match, marking a symbolic victory for machines over human intellect.
Despite its historic win, Deep Blue was not truly intelligent or creative. It was a powerful machine designed to evaluate millions of possible moves per second, using brute-force calculations rather than strategic thinking. Deep Blue could analyze up to 200 million positions each second, but its decision-making relied on evaluation functions created by human experts. Thus, while it could out-calculate human players, it lacked genuine creativity.
Fast forward to the present, and we encounter another significant milestone in AI: the game of Go. At first glance, Go may seem similar to chess, but it presents unique challenges due to its larger board and the complexity of evaluating positions. In Go, players place stones on a 19x19 grid, leading to an average of 250 to 300 possible moves per turn. This complexity made it difficult for traditional brute-force algorithms to compete with skilled human players.
For years, AI programs struggled to match the capabilities of top Go players. However, in 2016, Google's AlphaGo achieved a groundbreaking victory against world champion Lee Sedol, utilizing a different approach based on machine learning. AlphaGo learned from a vast database of 160,000 games played by top players, employing deep neural networks to evaluate positions and select moves.
AlphaGo's victory was not merely a result of imitating human strategies; it demonstrated a level of creativity that surprised even seasoned players. One notable moment during the match was the 37th move, which was initially perceived as a mistake by commentators but ultimately proved to be a brilliant strategic play. This unexpected move showcased AlphaGo's ability to think outside conventional strategies, leading to its victory.
While AlphaGo's success was impressive, it relied on a supervised learning approach, meaning it learned from existing data. However, researchers later developed AlphaZero, an algorithm that learned to play games like chess and Go solely through reinforcement learning, playing against itself without any human input. AlphaZero's training involved millions of self-play games, allowing it to develop unique strategies and outperform previous AI models.
AlphaZero's approach highlights a significant shift in AI capabilities. Unlike its predecessors, AlphaZero demonstrated creativity by rediscovering classic openings and developing its own unique style of play. Its willingness to make aggressive sacrifices and prioritize positional advantages over material points showcased a level of strategic thinking that resembled human creativity.
Moreover, the exploration of AI creativity extends beyond games. Techniques such as reinforcement learning and deep learning are being applied in various fields, including art and music, suggesting that AI's creative potential is far-reaching.
The evolution of AI from Deep Blue to AlphaGo and AlphaZero illustrates a remarkable journey in the realm of creativity. While early AI systems relied on brute-force calculations and human-designed strategies, modern algorithms are beginning to exhibit creative thinking through learning and self-play. As AI continues to advance, the question remains: can machines truly be creative, or are they merely reflecting the creativity of their human creators? The exploration of this question will undoubtedly shape the future of AI and its role in our lives.
