China’s largest game-maker Tencent is ramping up its research efforts in advanced gaming AI, a key step in its long-term pursuit of artificial general intelligence. The company recently took a huge leap when its homegrown AI system Wukong AI (绝悟) trounced a human pro team in China’s most popular mobile game Honor of Kings (王者荣耀), aka Arena of Valor. “Wukong,” aka Monkey King, is the name of a popular fictional character in China. Dragon Ball’s “Son Goku” is based on Wukong.
Honor of Kings is a multiplayer online battle arena mobile video game played between two teams of five players who must take down their opponent’s central base to win. The game is not so different from other popular MOBA games like Dota 2 and League of Legends, but is much simpler to master. The game environment has 92 “Heroes,” each with a unique set of abilities. In 2017 Honor of Kings had over 80 million daily active players and 200 million monthly active players.
Tencent’s development of Wukong AI follows the huge success of its predecessor Fine Arts (绝艺) — a Go-playing AI system that defeated all its opponents to claim victory at the AI-exclusive Computer Go UEC Cup in 2017.
Exploring AI’s capabilities in real time strategy (RTS) games has become a popular trend for researchers because such games are more computationally complex then board games; have rich human-computer interactions and involve multi-agents; and generate tons of data. San Francisco-based AI company OpenAI and Alphabet’s London-based AI lab DeepMind have developed specialized bots to play Dota 2 and StarCraft respectively.
Here are Synced’s fast takeaways from this most recent bots vs humans showdown:
- The match took place at the King Pro League (KPL) Honor of Kings World Champion Cup, held last weekend in Kuala Lumpur, Malaysia. The KPL is the world’s biggest Honor of Kings esports competition, launched in 2017 by Tencent.
- The human pro team consisted of four pro players from China (eStar Pro’s 兮兮, EMC’s Sun, Nova’s Seek, KingZone’s Night), and one pro player from Malaysia (M8HEXA’s Mike).
- The game lineup was set as follows:
- Humans: Cao Cao, Nakelulu, Wu Zetian, Di Renjie, Zhang Fei;
- Wukong AI: Damo, Athena, Wang Zhaojun, Yuji, Niumo.
- Although the pro team drew first blood by winning an early team fight, the AI bots soon found their form and secured a victory in 16 minutes.
Tencent AI Lab introduced its gaming AI system last December in the paper Hierarchical Macro Strategy Model for MOBA Game AI. In experiments the AI bots defeated a team of human players ranked among the world’s top one percent. Tencent researchers told Synced that compared to Go, MOBA and RTS games generally present four unique challenges:
- Computation complexity: the motion space and state space could lead the computation complexity up to 10²⁰,000. Go only reaches complexity of 10²⁵⁰.
- Multi-Agents: RTS games usually involve multiple agents. The task of coordination and cooperation between agents can be key to victory.
- Information incompleteness: Go is a perfect information game; whereas many RTS games include some sort of of “mist” feature which hides parts of the environment map to increase difficulty.
- Sparse and delayed rewards: A game of Go usually involves less than 361 steps, but a typical MOBA game such as Arena of Valor involves about 20,000 frames.
A key innovation of this paper is a novel learning-based Hierarchical Macro Strategy model for mastering MOBA games. Specifically, the research team proposed a two-layer macro strategy architecture: an Attention layer that aims to predict the best region on game maps to move Heroes, and a Phase layer that aims to recognize the current game phase (opening, “laning,” or late game). The output of the attention layers was used to guide the model’s micro level execution. In addition, the research team designed a cross-agent communication system to help their bots learn to cooperate.
Tencent AI Lab told Synced that Wukong AI will be deployed in Tencent’s core gaming sector. “With the advantages of algorithms and data, AI can provide data, collaboration strategy, and real-time analysis and recommendations for professional players. Meanwhile, we will further open-source research resources for game applications.”
Tencent is also hoping its heavy investments in advanced gaming AI can eventually return far-reaching benefits across a wide range of fields, such as healthcare, manufacturing, self-driving cars, agriculture, and smart city technologies.
Journalist: Tony Peng | Editor: Michael Sarazen