The relationship between artificial intelligence and games can be summed up in two words: cat and mouse. Since 1997 Deep Blue defeated Kasparov at chess, primal mouse, a long list of rodents have been presenting their candidacy to become the definitive game that proves or denies the intellectual superiority of the machine: Jeopardy, Starcraft, Poker Go, another aspiring classic, defeated in 2016, has just returned to the ring determined to make life difficult for the cat with the help of a team of scientists from Shanghai Jiao Tong University.

Those responsible for the new challenge have devised a modality inspired by the quantum physics of this ancient board game. In it, players make their moves with two interlocking tiles instead of one. Quantum physics can give the game the non-deterministic feature typical of games of chance that does not exist in the classical version, they explain in the resulting study. Furthermore, in this revamped Go, it is played under conditions of imperfect information ; that is, players can only know a part of the state of the game, while the rest is kept secret, like the cards in poker.

In classic Go, two players face each other to achieve control of the board and surround their opponent by strategically placing black and white pieces, called stones, on the intersections of a board of 18 by 18 squares. If a player occupies the four intersections that surround an opponents piece, he captures it. At the end of the game, which comes when the two players pass their turn for not seeing more possible moves, the one who has surrounded the most empty intersections with their stones wins. This seemingly simple dynamic allows long successions of plays and generates endless scenarios that, at least until 2016, made Go a challenge with the potential to exceed the capabilities of the machine and remain on the shrinking list of things that Humans still do better.

The quantum version is a twist that exponentially expands the possible states of the game by incorporating interlocking tiles into the game and confronting the algorithm with someone of its size: another machine.

The key to this new and convoluted game mode is that the position of the interlocking stones on the board is not final. As soon as the opponent places a piece on a vertex adjacent to either of the two that are interlocking, one of them will disappear and the other remains on the board. In this way, the player does not know if he has succeeded in his action until he has completed it. In the event that the tile next to which it has been placed disappears when the entanglement collapses, you will have wasted your time. In the same way, the unmasked stone that remains on the board will only be able to surround the enemy from that moment on. Lets say that each player places a pair of interlocking stones separate from their opponents. At that point, the board could have four different configurations depending on which one remains on it.

Who decides stays and who leaves? The permanence or not of the chips is obtained by creating a true quantum entanglement process. The scientists used entangled pairs of photons to extract a random series of measurements of 0 or 1 that were assigned to the paired stones.

Where we come from

Go was considered a worthy successor to the outdated chess for two main reasons. On the one hand, the greatest number of possible positions on your board complicates the tasks of searching for potential movements. On the other hand, the aspect of a victory in chess -capturing the king- is more limited than in Go, where any configuration of the board in which none of the players see more benefits to conquer, gives rise to the final count.

As it is, it is not surprising that, for decades, the machine was incapable of beating a human, whether it was this professional or amateur player. The hunt for Go jumped into the worlds newspapers in 2015, when Fan Hui took on Alpha Go, the algorithm developed by Deep Mind, in a first round from which the machine emerged victorious. In 2016, this artificial intelligence established itself as superior as far as Go is concerned after winning the former world champion, Lee Sedol.

Then Alpha Go Zero would come. And then Alpha Zero. The original learned to play over the course of thousands of games against players of varying levels. The second generation learned by playing against itself. And the third, also self-taught, also taught himself how to play chess and shogi.

Where we go?

It is not unreasonable to ask what humanity wants a quantum Go for. According to the Chinese scientists who have invented it, the aim is basically to raise the bar for the machine. Our results establish a paradigm for inventing new games with quantum features and resources and offer a versatile platform for both classical and quantum machine learning, they explain.

Alpha Zero, for example, brings to Deepmind collateral victories to those that occur on the board. An algorithm capable of assimilating the rules of three different games is a significant advance towards the creation of learning systems of general purpose and adaptable to changing situations. Putting the machines to pursue these new goals can have two results: that the algorithm becomes more sophisticated until it reaches them or that the mouse hunts the cat and we finally find the limit of artificial intelligence.

Go here to see the original:

Quantum physics: the trick to beat artificial intelligence to Go? | Innovation - Explica