In late 2016, an anonymous player called Master and Magister logged onto elite online Go servers and beat the world’s best human players 60 times in a row without a single loss. It was an upgraded version of AlphaGo. This deep dive tells the biography of that machine, starting with why Go resisted computers for decades after chess fell in 1997: a 19 by 19 board with more possible configurations than atoms in the observable universe, where brute force fails and masters rely on intuition and aesthetic sense.

The hosts explain DeepMind’s dual network architecture, a policy network that prunes possibilities and a value network that predicts outcomes, combined with Monte Carlo tree search, trained first on 30 million human moves and then through self-play with injected randomness that shed human bias. They cover the secret 5-0 win over European champion Fan Hui in October 2015, the March 2016 match in Seoul where AlphaGo beat Lee Sedol 4-1 despite his out-of-distribution divine move 78 in game four, the 3-0 sweep of the world number one in Wuzhen in 2017, China’s Sputnik moment, AlphaGo Zero learning from scratch in days, AlphaZero mastering chess and shogi, and AlphaFold applying the same lineage to protein folding.

• Why Go’s branching possibilities made it the white whale of artificial intelligence
• Policy network, value network, and Monte Carlo tree search explained through a chef analogy
• Why AlphaGo played conservatively for one-point wins and why that terrified professionals
• How Lee Sedol’s move 78 exposed a blind spot in the machine’s training data
• From board game to biology: AlphaGo Zero, AlphaZero, and AlphaFold