After an astonishing close challenge, Google’s AI takes 1-0 lead in a historic five-game matchup between Google’s artificially smart Go-playing computer system and world’s best Go player Lee Sedol at Seoul’s Four Seasons hotel.
A group of Google specialists spent two years at an AI lab and constructed AlphaGo in DeepMind, London. The machine has proved that it can compete with the game’s best, and also shows off amazing capacity to learn the game on its own.
AI specialists outside of Google were wondering if a breakthrough could happen anytime in near future, and many believed that it would take a decade for a machine to defeat the top people.
The good news – Google created this machine recently and proved everyone wrong. Rémi Coulom, the French analyst behind the top intelligent Go player said it happened faster than he suspected.
Lee Sedol and AlphaGo’s operator, DeepMind researcher Aja Huang were ready for the game. The game took place in a small, closed room with a handful of authorities being present. The game was viewed by the press from two separate commentary rooms, one for Korean speakers and one for English.
Lee, a South Korean who exhibits a bowlish hair style and looks more youthful than his 33 years, spent a significant part of the match inclining forward, supporting his jaw in his hand. DeepMind developer Aja Huang sat opposite to Lee. He physically set the stones on the board in positions picked by AlphaGo. Lee played aggressively from the beginning, putting AlphaGo on edge.
Both AlphaGo and Lee were committing errors and the match was a very close call when Lee finally acknowledged AlphaGo had built an unrealistic lead. AI had scored a win in the most innovative and intricate games ever devised, after conquering Chess in 1997.
“Reinforcement learning” was infused into AlphaGo by the DeepMind team which helped the machine to adjust its own neural networks based on trial-and-error. AlphaGo also narrowed down the near-infinitely available space for the next best move to a reasonable number and was able to predict the result of each move.
Remaining four more matches will take place up to Tuesday. In the event that DeepMind’s AI can comprehend Go, then perhaps it can understand a whole lot more like human’s intuition and other complex tasks.