How Baidu’s Bingo AI Cracked the Go Challenge with Novel Algorithms

Historical Context

In 1997, after IBM’s Deep Blue defeated chess champion Garry Kasparov, Time magazine warned that computers would need another hundred years to master the game of Go. Less than two decades later, Baidu announced that its Bingo AI system had reached amateur‑to‑professional strength on both 9×9 and 19×19 boards.

Performance Milestones

During the “Challenge Bingo” human‑machine Go tournament on July 25, Bingo defeated two 5‑dan players with a 2‑1 score in each match, finishing the series with 38 wins and 4 losses, a clear dominance that showcased the system’s rapid progress.

Algorithmic Innovations

Traditional chess engines rely on Alpha‑Beta pruning, but Go’s search space is astronomically larger—approximately 10^120 times that of chess—making pure Alpha‑Beta ineffective. Monte‑Carlo Tree Search (MCTS) with the UCT algorithm became the standard, yet it suffers from massive sampling requirements and slow runtime.

Baidu’s team, led by senior engineer Yang Cheng, extended UCT with a new learning algorithm that generates several times more statistical samples within the same time budget, accelerating online learning. They also introduced a weakened form of Alpha‑Beta search that is tightly integrated with both online reinforcement learning and offline supervised learning, embedding it into the Monte‑Carlo framework to explore more promising branches.

Learning Strategy

Bingo employs a fully supervised learning pipeline augmented by massive online self‑play. Each move is evaluated for quality, and the system automatically aggregates millions of game positions to refine its evaluation function. Professional game records serve as additional reference data, allowing the AI to internalize high‑level strategies without direct human coaching.

Technology Transfer

The search and learning techniques pioneered in Bingo have been repurposed across Baidu’s product ecosystem, including web search ranking, personalized recommendation, image recognition, and mobile app recommendation. This cross‑domain application illustrates how advances in a seemingly isolated research problem can drive tangible improvements in consumer‑facing services.

Industry Significance

Artificial intelligence is hailed as a third‑generation technology and a top‑ranked general‑purpose technology by leading institutions. Baidu’s breakthroughs in Go AI not only demonstrate technical leadership but also signal a broader shift where sophisticated AI methods become integral to everyday internet services.