Artificial intelligence and board games is not an entirely new combination. In the past, computers have already played against world chess champions or beat "Go" world champions, today board games with app support are one of the ways to bring digital and analogue together. The company "DeepMind" nevertheless goes one step further and teaches the AI how to play the board game Diplomcy. Why is this so interesting? Continue reading!
Board games with a computer touch often work simply: one or more human players compete against an AI opponent. This is the case with chess, Go, Mah-Jong or even tabletop games with software support, such as Lord of the Rings - Journey through Middle-earth or Imperial Assault Asmodee, typical Hybrid games. There one side wins: player or computer, sometimes there is a tie. However, it is always a matter of so-called zero-sum games, i.e. systems in which you play against each other to win.
Diplomacy: How diplomatic can computers be?
But what happens if a board game is not a pure zero-sum game, but focuses on a game process in which interim cooperation is a key to victory? The tech company DeepMind is breaking new ground in this regard and trying to teach artificial intelligence how to work together* – with the board game classic “Diplomacy”.
The title has been around for a few years and many publishing houses, originally comes from the late fifties and is aimed at two to seven players with staying power. A game lasts around four hours, which is long for humans and not a problem for computers. It's about a classic domination concept: the players simply try to gain dominance over Europe. However, this does not succeed by sending out masses of armies, but sometimes through diplomatic channels. Exciting: There have already been some adaptations of Diplomacy for the PC, but with moderate success. The simple reason: negotiations as a core element of the board game were directly related to the "quality of the decisions" of the AI players.
Behind the seemingly simple victory condition - you have to control players 18 out of 34 territories - there is a clever trick: The highlight of Diplomcy is that you always rely on the help of other players, winning on your own is almost impossible. And so individual players come together, agree non-aggression pacts or join forces against other parties. You have to know: Diplomacy has worked hard for its reputation of destroying friendships over the decades.
For the experts of the tech company DeepMind, the cooperation of several artificial intelligences has to work and that is completely different than with classics such as chess. Research Engineer Tom Eccles explains: “AI systems have already proven that they can outperform the best human players in zero-sum games. In this gameplay there can only be one winner and one loser. Differing from that, Diplomacy is about building alliances and fostering cooperation”. In particular, the mixture of cooperative and competitive elements makes diplomacy an interesting research challenge, adds researcher Tom Anthony.
AI learns gunboat diplomacy
The AI must be able to organize itself in a complex way, taking many factors into account. The researchers call this synergy – and it is precisely these easy and difficult decisions that artificial intelligence must bring together to form a solution. “Artificial intelligence is increasingly being applied to more complex tasks. This could mean that different autonomous systems have to work together, or at least in the same environment, to solve a task,” explains Eccles.
It is important to note that the basis for the researchers is the rule variant of Diplomacy, in which players are not allowed to speak to one another. So it's about the game decisions made in secret and the then implemented movement of the pieces. So the researchers face many questions and challenges.
However, the experts at DeepMind have an eye on how powerful AI action can be: "We start from the premise that all AI applications should remain under human control and be used for socially beneficial purposes," explains Tom Anthony. “Our teams working on technical security and ethics want to ensure that we are constantly anticipating short- and long-term risks, looking for ways to prevent those risks from occurring and finding ways to eliminate them when they do occur. "
The final goal is frightening: If Artificial Intelligences manage to interact on a level of action previously reserved for humans, what is the next step?
[stextbox id='author' caption='We are looking for reinforcements for the editorial team']We're looking for news writers in the fields of gaming, books, board games, as well as films and series. Do you want to participate? then > here. < click and apply.[/ textbox]
