The day has come when one Artificial Intelligence is capable of playing war. At least to the war of diplomacy and, literally, to play. An AI called Cicero, created by Meta, has managed to outplay the players of diplomacya popular online game based on a board game with a lot of history.
The objective of Artificial Intelligence created by Mark Zuckerberg’s company is to emulate human performance in decision making. diplomacy It is the perfect place to put it to the test. It is not the first time that an AI has competed in games against human beings. The battle in the field of chess, Go or the ladies. However, in this popular online game based on political relations in Europe there is a new variable. Its operation starts from the discussion between players. Pure diplomacy.
One that had the objective of testing the analysis of natural language by Artificial Intelligence and then making a binding decision. This is one of the biggest barriers of this technology. Although it was capable of reproducing human text, it was not fully capable of understanding how natural language worked, let alone reproducing it. In the case of Diplomacy, in addition, the factor of interactivity, the effect on third parties and subsequent decision-making are added.
Opponents will have to debate to create links and alliances. Later it will be seen if they are breached or broken, jeopardizing international diplomacy –at least in the game–. The objective of the game is to seize the power of the cities considered supply centers in order to survive.. Later, each turn, you will have to decide whether to attack or support your opponents.
Cicero, Artificial Intelligence that learns from human errors
In the case of Cicero, the Artificial Intelligence of Meta has been able to study whether his opponents will comply with the agreed diplomatic agreements or if, on the contrary, they will end up breaking them. All starting from the base of the game: conversational negotiation. On this Cicero has the ability to negotiate tactical plans, reassure an ally, discuss the game’s broader strategic dynamics, or even engage in small talk, about just about anything a human player could discuss.
Cicero’s performance on webDiplomacy.net has been overwhelming. Artificial Intelligence managed to close 40 games, of two hours each, against 82 humans. Games in which, by the way, he always maintained his anonymity so as not to reveal his status as a machine. Finally, he obtained a score twice the average of the rest of the players. Significantly above average.
How do i do it? Like any other Artificial Intelligence, Cicero began by learning from his closest opponents. Which means that he began by replicating the most intrinsically human failures: he lied, and a lot. And besides, he was doing it wrong. Later he was polishing the technique until reaching the balance between what he would have to comply with and what not. Also in his ability to influence the decision of third parties. Before putting it into practice online, the team of researchers in charge of this Artificial Intelligence provided the system with thousands of game data in Diplomacy. A total of 125,261 anonymous games, 40,000 dialogues and 12 million messages.
For Meta, the objective of this negotiating Artificial Intelligence has a clear focus for its future, and uncertain by the way, metaverse. To create more immersive, social experiences and to connect with those around us in a more natural way. In fact, they point on your websitethat the fact of making Cicero’s code open has the objective that third parties rely on its Artificial Intelligence technology to make responsible use of it both inside and outside its ecosystem.