Professor Fran Li’s friends know him as an enthusiastic fan of board games. He competitively played Go, a strategic board game popular in East Asian countries, while in high school and college. At the time, he never thought that he would someday link his hobby with his research.
Now his Go hobby has grown into the territory of a research proposal funded by the National Science Foundation at $330,000 for three years, beginning August 2018. His project, “From AlphaGo to Power System Artificial Intelligence,” uses game-based artificial intelligence (AI) technology to investigate power-grid issues.
Go, also called Weiqi or Baduk, originated in ancient China some 2,500 years ago. It is played by two people alternately placing black and white stones on a board marked with a 19-by19 square grid. The goal is to surround more territory on the board than one’s opponent, so it is sometimes literally translated as “the surrounding game.”
While Go is often compared with chess, it has a higher measure of complexity—10^170 in state space, while chess is merely 10^47. Go’s popularity has increased throughout Europe and North America in recent years, especially in academia. Mathematicians and computer scientists found that Go is one of the best targets for testing AI algorithms.
AI has made many significant achievements in gameplay in the past a few decades. The chess software DeepBlue beat the legendary chess champion Garry Kasparov in 1997. But the best Go AI still could not beat an average amateur Go player, let alone any professional player, because of the difficulty of mimicking a human brain’s logical reasoning ability in this complex game.
That situation changed between 2015 and 2017 when Google DeepMind’s AI program called AlphaGo defeated several world professional Go champions. This was considered such an epic milestone in both the AI and Go communities that the AI triumph was featured on the cover of Nature.
Li was immediately attracted by the success of AlphaGo. Since then, he has examined the reasons that AlphaGo can achieve what past AI efforts could not, and investigated ways to use the strategies and algorithms in AlphaGo to solve some complex problems in the field of electric power systems.
After laying out a detailed comparison of Go, AlphaGo, and some power system problems, Li proposed ideas to address a number of emerging problems in the modernized power systems under the smart-grid era, such as strategic market bidding with renewables, security assessment under multi-scenario and multi-period paradigm, and other aspects.