Tencent has unveiled its Go-playing AI program, “PhoenixGo,” on Github, making its source code and training model available on the open source platform (in Chinese). The program, developed by WeChat’s translation unit, is the first open source Go AI project in China. A team of WeChat engineers focusing on the development of machine translation decided to use the spare processing power of WeChat’s servers to train the PhoenixGo program, which implemented AlphaGo Zero, the latest version of DeepMind’s champion-defeating AlphaGo, and increased its training efficiency. PhoenixGo’s source code and training model can run on a single GPU chip and perform at the same level as a professional Go play.er.
PhoenixGo began competing on FoxGo—Tencent’s online Go playing platforms for professional players and aficionados—under the alias “BensonDarrz” in January. In April, the program recorded a 200-game winning streak. PhoenixGo started gaining more attention when it claimed the championship of the “World AI Go Tournament 2018” in late April in Fuzhou.
To date, Tencent has released 52 open source projects, 22 released by WeChat. Most recently, Tencent led an $820 million investment in Chinese humanoid robot company UBTECH, which set a new financing record for the largest investment raised in a single round by an AI company.
Last year, China, the world’s second-largest economy, unveiled its plans of becoming the world leader in AI by 2030. Tech giants, Baidu, Alibaba, and Tencent, among other AI companies, are stepping up their investment plans, spearheading into the development of cutting-edge AI technologies.