A.I. learns through new computer games

To further push the abilities of artificial intelligence, researchers and developers are teaching through various video games. A.I. has reached the point where being able to learn from pre-programmed choices is no longer cutting it. Many Gen Zs interact with A.I. on a daily basis, emphasizing the need for A.I. to better collaborate with humans, even in regards to creativity.


Researchers from Google Brain and DeepMind are teaching A.I. to play Hanabi, a card game that is considered to be harder than Go and Chess. Hanabi is a game that requires theory of mind, or awareness that not everyone’s understanding and mental processes are the same. The game requires collaboration between players, meaning the A.I. has to understand information it receives from other players based on their moves. If researchers can successfully teach A.I. to play this game, they will be closer to developing A.I. that interacts effectively with humans.


Created by The Allen Institute for Artificial Intelligence, Iconary is a game similar to Pictionary, only the player works alongside an A.I. player instead of a human opponent. Players draw and arrange icons to form phrases for the computer to guess, or they can try to guess arrangements made by the computer. What makes Iconary stand out in the world of A.I. is that the computer isn’t trained to recognize the drawings, it is learning to recognize relationships between the images as it goes. There are 75,000 phrases in the game and developers are adding more regularly. Researchers claim Iconary is currently the most significant example of A.I. collaborating meaningfully with humans.


Unity, a real-time creation platform, has created Obstacle Tower as a game meant to only be played by A.I. The game is packed with 100 different levels that change and scale in difficulty every time they’re played. The purpose of the game is to judge the level of sophistication of an A.I. based on its ability to learn the game on the go. Different levels will test the computer software’s vision, virtual locomotion, planning, and spatial awareness. Unity hopes this game will serve as a benchmark tool to evaluate self-learning software and help researchers better understand how the mind learns to solve new problems.

Have you heard the latest episode of The Cassandra Daily Podcast? Listen and subscribe here.