A team of Google Deepmind engineers has published a challenge aimed at the artificial intelligence (AI) developer community. They want to get an AI that is able to participate in a card game called Hanabi.
We have already seen the big ones fall in this area. The first was DeepBlue, from IBM, beating Chess Grandmaster Gary Kasparov, but there have been more like DeepStack and Libratus in the poker world and more recently AlphaGo, of Deepmind themselves, against the great masters of Go.
Why is it a challenge to develop a system to play a simple card game?
Hanabi, unlike classic games, is a cooperative game. That means that all players collaborate to reach a common goal. If not everyone wins, everyone loses.
This game requires collaboration and communication, which are different challenges to those posed by the other games, and therefore, capabilities that are not developed so far.
The AI must be able to establish a communication protocol with humans during the game. The added problem is that we humans tend to adopt conventions in an organic way, evolving even throughout the game, instead of establishing a set of fixed rules and sticking to them.
We’ll be updating this story shortly when we have more information.