In March 2016, AlphaGo played a surprising move 37 in game two of its match against Lee Sedol, a top professional Go player, which appeared to be a gift to its opponent but ultimately helped AlphaGo win the game and the match 4-1, according to technologyreview.com. This move challenged perceptions of AI creativity and reasoning in complex games.
Unlike Deep Blue’s chess victory in 1997, which relied on brute-force calculation of millions of positions per second using human-coded rules, AlphaGo faced a far more complex game where evaluating all possibilities would take billions of years. Instead, AlphaGo had to quickly assess who was ahead and devise novel moves, demonstrating a form of reasoning that current large language models lack, technologyreview.com explained.
AlphaGo’s success highlighted a key difference between past AI systems and today’s large language models (LLMs). While LLMs excel at pattern recognition and generating plausible text, they do not possess the reasoning capabilities that allowed AlphaGo to make creative and strategic decisions in Go. This distinction underscores ongoing challenges in developing AI systems that can truly reason and produce reliable outcomes.
Lee Sedol himself acknowledged the significance of move 37, stating that it changed his view of AlphaGo from a mere probability calculator to a creative entity. The match remains a landmark event illustrating the gap between current AI models and the reasoning power demonstrated by specialized systems like AlphaGo, as reported by technologyreview.com.