William Marshall
December 12, 2017

Submitted as coursework for PH240, Stanford University, Fall 2017


Fig. 1: An example of a game of Go. (Source: Wikimedia Commons)

For decades, humans have been intrigued by the limits of the practical powers of computers and their power relative to humans has been an area of great intrigue. Gameplay has often been thought of as the standard by which to match up humans with machines. Decades ago, the games of checkers and chess slipped fell to the dominance of computers. However, the game of Go (see figure 1) has been viewed as the gold standard for what it meant to be able to reason, and until recently was a game dominated by humans. However, that changed in 2015 when Google's AlphaGo program beat human Go champions.


After a complicated set of training by both observing play by top humans and simulating games against itself, the AlphaGo program was able to achieve performance well beyond anything achieved previously by humans or machines, winning 99.8% of the games it played and defeating the reigning world champion 5 games to 0, achieving an ELO rating of well over 3,000. [1] Indeed, all of the evaluation games that the program lost were against a weaker version of the same algorithm, as it went undefeated against all other opponents. [1] Even when opponents were advantaged by being given a handicap, the AlphaGo program still won between 77% and 99% of its games against previous state of the art algorithms. [1]


Although Go is only a board game, this result still has major implications for the future of AI and the relation between humans and computers. Go was seen as the last stronghold of a problem that supposedly showed that humans could reason while computers could only mindlessly carry out instructions. Advances in machine learning will continue to force us to question what logic and reasoning truly is and what it means to think and to be human.

© William Marshall. The author warrants that the work is the author's own and that Stanford University provided no input other than typesetting and referencing guidelines. The author grants permission to copy, distribute and display this work in unaltered form, with attribution to the author, for noncommercial purposes only. All other rights, including commercial rights, are reserved to the author.


[1] D. Silver, et al., "Mastering the Game of Go With Deep Neural Networks and Tree Search," Nature 529, 484(2016).