DeepMind and Meta artificial intelligences are now beating human players in the strategy board games Stratego and Diplomacy. This is a new milestone for the technology, but some experts fear the risk of serious abuse…
For many years, artificial intelligence beat man at chess and the game Go. However, she has not yet been able to master the more complex board games.
AI just took a new step to become an expert in stratagem and diplomacy. These two strategy games are based on the idea ” imperfect information unlike chess and go where players see all the pieces on the board.
at Stratego, the identity of the parts is hidden until another piece hits them. Diplomacy, in turn, consists of agreements, alliancesand vengeance, the nature of which is kept secret.
These games therefore do not require calculation of victory paths, but require more subtle skills such as the ability to guess the opponent’s thoughts and adjust their strategy to thwart their plans. You have to bluff and convince.
These two board games have been mastered a few days apart by two AI models different, one develops by DeepMind and the other from Meta (formerly Facebook).
DeepMind’s DeepNash wins against Stratego
the Model playing Stratego and developed by DeepMind referred to as DeepNash. Instead of focusing on making smart shots, it’s designed to play in unpredictable ways.
This game has features that make it stand out more complicated than chess, go or poker already mastered by AI. Two players each place 40 pieces on a board, but cannot see the opponent’s pieces. The goal is to move the pawns to eliminate your opponent’s and capture a flag.
In total, Stratego can unfold at 10535 Manners different. In comparison, this number is 10360 for the game Go. Similarly, Stratego has incomplete information at the start of the game 1066 hidden positions possible around 106 at a poker game.
This AI is sometimes brave. In a game against a human, they sacrificed several high-ranking pawns and found himself outnumbered. It was actually a calculated risk to push the player to do their best. It then won by developing its strategy around that element.
This DeepNash model is good enough at Stratego to beat every other system and win every time 84% of games against experienced people. In 50 games played on the online gaming platform Gravon since 2002, she has risen to the third place among the best Stratego players on the platform.
To achieve this level of performance, DeepMind couldn’t use the same algorithms as for the chess and Go game, which weren’t adapted to that game at all, so the researchers invented a new algorithmic method entitled Regularized Nash Dynamics.
The DeepNash model combines an algorithm reinforcement learning With a deep neural network. To find the ideal action for each state of a game, this AI played 5.5 billion games against yourself.
Cicero: Meta’s AI has mastered the game of diplomacy
On his side, the AI masters diplomacy is developed by Meta and CSAIL and called Cicero. Despite the difficulty of this game, the model is able to compete with human players.
In Diplomacy, up to 7 players compete and represent each other each a European power before the First World War. The goal is Control supply centers by moving fleets and armies.
This game requires a sense of intrigue, a talent for treachery and false promises and a real Machiavellianism. The complexity is not in the world map or the counters, but in the strategy surrounding the agreements made. In addition, players must communicate verbally and convince of the sincerity of their intentions.
So again not not just a question of computing power. To beat the human at this game, Cicero follows a multi-step process.
First, the AI relies on the current status of the board and ongoing discussions make an initial prognosis actions of each player. She then perfects that prediction and uses it to formulate an intention for herself and her partners.
Then she generates multiple candidate messages based on the board’s status, dialogue and intentions. Candidate messages are then filtered to reduce nonsense, maximize value, and ensure consistency with intent.
This artificial intelligence was trained by the Data from 125,261 games on the online version of Diplomacy, combined with data from matches played against themselves. His Strategic Reasoning Module (SRM) has thus learned to predict the actions of the players and to choose an optimal action accordingly.
His dialog module, which is used to communicate intentions to allies, is based on a 2.7 billion parameter language model that was pre-trained with text found on the internet and then refined with messages from human-played Diplomacy games. Based on the intentions of the SRM, this module generates a chat message.
On webDiplomacy.net, Cicero managed to assert himself against his human opponents. She got up in second place in a ranking of 19 players and surpassed the score of most of them.
An AI that can start a war?
According to Michael Wellman of the University of Michigan, the speed at which different game features conquered or mastered by AI in recent years is quite remarkable “. The computer science researcher deals with strategic thinking and game theory.
As he points out: “ Stratego and Diplomacy are quite different and also present completely different challenges of games where similar success has been achieved “.
According to meta-AI researcher Noam Brown, these game AIs are able to interact with humans and accommodate sub-optimal or even irrational actions pave the way to real applications.
In his words ” when you build a self-driving car, you don’t want to assume that every other driver on the road is perfectly rational or will behave optimally. Cicero is a big step in this direction “.
He believes this technology could help to better understand virtual assistants what consumers want, where they go more compelling virtual Metaverse beings and realistic. The goal of these researchers isn’t to build an AI that can beat humans in games, but to work with them in the real world.
However, some experts are significantly less optimistic. According to University of Michigan artificial intelligence expert Kentaro Toyama, These AIs are scary and could be used for evil “. As Generative AIs worry artiststhis type of artificial intelligence also poses a threat.
He fears her Ability to hide informationThinking several turns ahead of the enemy and surpassing the intelligence of humans is a risk. In his eyes, this technology could be used to create more convincing scams or more realistic deepfakes.
the Cicero’s code is publicly availableand malicious actors could copy him and use his negotiation and communication skills to craft persuasive emails and blackmail their prey.
Worse, if someone were to train this language model on data like the diplomatic documents leaked by WikiLeaks, Toyama worries the system could impersonate a diplomat and to initiate communication with a foreign power.
According to this specialist AI is like the nuclear power of this era. It has colossal potential for both good and evil, but… I think if we don’t start regulating evil, all sci-fi about dystopian AI will become scientific fact »…