In 2016, Google DeepMind's artificial intelligence (AI) "AlphaGo" faced 9-dan Lee Sedol and posted four wins and one loss. It was the "match of the century" in which AI beat humanity for the first time. Demis Hassabis, Google DeepMind chief executive officer (CEO), said the results of that match "became the signal flare heralding the dawn of today's AI era," and declared that Gemini and AlphaGo would usher in the era of artificial general intelligence (AGI).
On the 11th, to mark the 10th anniversary of the AlphaGo–Lee Sedol match, CEO Hassabis stated accordingly in an op-ed titled "From games to biology and beyond: A decade of progress since AlphaGo," posted on the Google blog. After founding DeepMind in 2010, CEO Hassabis has led AI innovation by developing various AI models such as AlphaGo and AlphaFold.
CEO Hassabis said, "Ten years ago, AlphaGo, Google DeepMind's AI system, became the first program to defeat a world champion in the highly complex game of Go," adding, "It was a monumental milestone that arrived a decade earlier than experts had expected at the time."
He cited move 37 in game two as the symbolic moment of the "AlphaGo vs. Lee Sedol" showdown watched by more than 200 million people. CEO Hassabis said, "Through a single move that has now become legend, 'move 37,' AlphaGo proved AI's vast potential, and Google showed that it had built the technological foundation to solve real-world scientific challenges."
Move 37 by AlphaGo was so unconventional and outside convention that experts initially thought it was a mistake, but it was ultimately judged to be the decisive blow that determined the match. He said, "It was an example that showed the remarkable insight and capabilities of an AI system that does more than mimic human experts—it discovers entirely new strategies."
CEO Hassabis explained that Go has long been a testbed for AI research because of its overwhelming complexity. He said, "The number of possible positions on the Go board is 10170 (10 to the 170th power), far more than the number of atoms in the observable universe," adding, "To handle such an enormous search space, AlphaGo used deep neural networks that combined reinforcement learning and advanced search, AI methodologies pioneered by DeepMind."
Afterward, AlphaGo learned promising moves through hundreds of thousands of self-play games and improved on its own by reinforcing strategies with higher probabilities of winning.
Google DeepMind has leveraged AlphaGo, which mastered Go, as a springboard to expand into other fields. He said, "At the moment of victory in Seoul, we were convinced AI was ready to be applied to 'scientific breakthroughs,'" adding, "As a first step, we set out to solve the 'protein folding' problem—predicting the three-dimensional structure of proteins, information essential to understanding disease and developing new drugs—a grand scientific challenge that had remained unsolved for 50 years."
Google DeepMind solved this long-standing scientific challenge with the "AlphaFold 2" system introduced in 2020. CEO Hassabis said, "Today, more than 3 million researchers worldwide are using this AlphaFold databases to accelerate a wide range of important studies, from malaria vaccines to plastic-degrading enzymes."
The next goal for CEO Hassabis and Google is AGI. He explained, "For AI to have truly 'general intelligence,' it must understand the physical world, so we designed the Gemini AI model to be multimodal from the outset," adding, "Gemini processes and reasons across not only language but also audio, video, images, and code, and builds a 'world model.'" A "world model" that understands and predicts the real world and physical laws is considered a gateway to AGI.
CEO Hassabis said, "True creativity is a core capability that such AGI systems must demonstrate," adding, "Move 37 was a moment that hinted at AI's potential to break existing molds, but truly original invention demands a higher order of ability."
Finally, he said, "Google DeepMind believes AGI will be the most significant technology in human history and the ultimate tool to advance science, medicine, and productivity."
UNIST distinguished professor Lee Sedol said, "The greatest lesson AlphaGo gave us was to show in advance that the AI era is not a vague future but will soon become a reality," adding, "We hope AI will become a powerful helper that solves the grand challenges humanity has long been unable to crack."