Artificial Intelligence Learned A Popular Board Game And Is Now A Top Ranked Player
DeepNash, the latest artificial intelligence from the talented researchers at DeepMind Technologies Ltd., has upped the game-playing seen in AI by learning to play a classic board game. Described as more complex than chess and Go and craftier than poker, DeepNash has mastered the board game Stratego.
DeepMind is excited to announce that their artificial intelligence DeepNash, has learned, from scratch, how to play the war game known as Stratego. Considered challenging to master, Stratego requires two players, where the AI learned the game’s strategies by playing against itself a reported 5.5 billion times. During the research trials, DeepNash has beaten almost all human players in the game, which requires the opponents to set up a strategy to capture the enemy’s flag with confidential information.
Stratego is a strategy board game requiring two players, each responsible for 40 pieces representing individual soldiers in their army. The object is to capture the opponent’s flag or capture enough of the enemy’s players to halt their ability to make another move. The opponents cannot see the ranks of the opponent’s pieces, making it a game of imperfect information.
Artificial intelligence has previously mastered chess, Go, and poker, but Stratego is much more complex. According to The Byte, Go has ten to the power of 360 possible game states, much more than chess or poker, but Stratego has ten to the power of 535. In other news, some forms of AI are working on de-aging actors or turning people into pocket monsters.
This new advancement in artificial intelligence surpasses gameplay. These new findings will go far in the scientific community toward AI systems that are able to maneuver real-world scenarios with limited information. DeepNash could then be used in uncertain situations to develop balanced outcomes for complex problems.
Stratego is a game of bluffs but also uses strategy to complete the task and determine the winner. Each player must arrange the 40 pieces which are hidden from their opponent at the beginning of the game. This is where the strategy comes in. Each player must visualize every possible outcome when setting up the board and eventually make their turns with limited information.
As in poker, the ability to bluff is vital in Stratego, which is something artificial intelligence must learn. But the knowledge of poker will only take the AI so far in Stratego, as this game lasts much longer than a hand of poker, sometimes with hundreds of moves before a winner is declared.
An amazing factor in these findings is that artificial intelligence developed an “unpredictable strategy.” This allowed the AI to keep its human opponent guessing, which is instrumental in this game. Vincent de Boer, the co-author of a piece published in the journal Science, stated he had never heard of an AI Stratego player that could win against a human opponent.
DeepMind has reported that its artificial intelligence, DeepNash, has become so advanced in the playing of the game Stratego that it has achieved a top-three ranking among human experts on Gravon, the most extensive online Stratego platform. The AI played 50 times, earning an 84% winning record. This is truly a fantastic accomplishment in the world of artificial intelligence.
