Innovations of AlphaGo

One of the great promises of AI is its potential to help us unearth new knowledge in complex domains. We’ve already seen exciting glimpses of this, when our algorithms found ways to dramatically improve energy use in data centres - as well as of course with our program AlphaGo.

Since its historic success in Seoul last March, AlphaGo has heralded a new era for the ancient game of Go. Thanks to AlphaGo's creative and intriguing revelations, players of all levels have been inspired to test out new moves and strategies of their own, often re-evaluating centuries of inherited knowledge in the process.

Ahead of ‘The Future of Go Summit in Wuzhen’, we summarise some recent examples of AlphaGo’s strategic and tactical innovations, and the new insights they have revealed.

AlphaGo Style

AlphaGo's greatest strength is not any one move or sequence, but rather the unique perspective that it brings to every game. While Go style is difficult to encapsulate, one could say that AlphaGo's strategy embodies a spirit of flexibility and open-mindedness: a lack of preconceptions that allows it to find the most effective line of play. As the following two games will show, this philosophy often leads AlphaGo to discover counterintuitive yet powerful moves.

Although Go is a game of territory, most decisive battles hinge on the balance of power between groups, and AlphaGo excels in shaping this balance. Specifically, AlphaGo makes masterful use of "influence," or the effect of existing stones on surrounding areas. Although influence cannot be measured exactly, AlphaGo's value network enables it to consider all stones on the board at once, endowing its judgment with subtlety and precision. These abilities let AlphaGo convert local regions of influence into coordinated global advantages.

In this game (Dia. 1), Black (AlphaGo) has little secure territory, while White has three corners, but Black's influence radiates across the entire board. In particular, while the marked exchange solidifies White, it also improves Black's potential. Go players usually shy from such exchanges, which pay a definite price for uncertain profit, but AlphaGo combines its sterling judgment with a keen sense of risk and reward to make such moves possible.

However, the value of influence depends entirely on context, and AlphaGo relinquishes influence freely when it can be effectively mitigated. In the the game displayed in Dia. 2, one of the most surprising in its oeuvre, AlphaGo has just played an incredible six stones along the second line. Go players have a saying: on the fourth line there is influence, and on the third line there is territory, but on the second line there is only defeat. AlphaGo's play at first looks deserving of such censure, as these moves give White both strength and influence in exchange for Black's paltry 4 points of side territory. Most players, unwilling to bear the ignominy of playing the marked stones, would reject this line in an instant. Yet AlphaGo judges it worthwhile to keep White's stones separated, and in the following exchanges, slowly erodes White's influence from the top and bottom to secure a winning advantage.

New Moves, New Patterns

AlphaGo has also played several opening novelties in its recent games, the most salient being the early 3-3 invasion and a new variation of the "Magic Sword". Each defies conventional theory, but proves sound on deeper inspection.

The Early 3-3 Invasion

One of the most territorial joseki (corner sequences) in Go is the 3-3 point invasion, shown in Dia. 3.

This invasion immediately secures the corner, but the textbook sequence shown in Dia. 4 has long been disparaged as unsuitable for the opening, as it gives too much influence.

AlphaGo's innovation is to omit the marked exchanges, leaving the corner unsettled as shown in Dia. 5.

Though slightly less secure, Black retains miai (options) to escape on the left or finish the joseki later, and has gained territory while ceding only moderate influence. This strategy has created a great stir among professionals, and at least one has already tried it in an official game (Dia. 6).

The New Magic Sword

Originally trained on human data, AlphaGo knows modern joseki and usually plays accordingly. However, in the "Magic Sword," a famously complex joseki family named for the cursed sword of Muramasa, it diverges.

Starting from the position in Dia. 7, the usual result exchanges the corner for the side as shown in Dia. 8.

However, AlphaGo often prefers to sacrifice outside access for territorial compensation (Dia. 9).

Most Go players would not consider playing this variation, as it gives Black a powerful wall, but White's follow-up approach declares that Black's influence is not as valuable as it looks. If Black does not reinforce the wall, it may even become a target. Kim Jiseok, one of Korea's top professionals, recently played this line in a tournament game (Dia. 10), which he went on to win.

More to Come

AlphaGo's innovations show great potential for impact in the world of professional Go, and we hope to present many more opportunities for collaborative research at the upcoming Future of Go Summit in Wuzhen. We look forward with great excitement to AlphaGo and human professionals striving together to discover the true nature of Go.