EA tests self-learning AI agents in Battlefield 1

EA tests the use of virtual players who in Battlefield 1 teach themselves the basic skills virtually from scratch by using artificial intelligence. 

In this blog post, Magnus Nordin from SEED details how his team, inspired by Google’s work with old Atari games, wondered “how much effort it would take to have a self-learning agent learn to play a modern and more complex first person AAA game like Battlefield”. The AI-controlled troops in the game learned how to play after watching human players, then parallel training against other bots. The AI soldiers even learned how to pick up ammo or health when they're running low, much like you or I do.

The results are an 'agen'” that, while inferior to human players, “is pretty proficient at the basic Battlefield gameplay”. The agent changes behaviour if it’s low on health or ammo, and while more complex behaviours like knowing the details of each map are beyond it (at the moment), EA has found that “while the human players outperformed the agents, it wasn’t a complete blowout by any stretch.”

Nordin noted that AI agents can hold their own against humans when they're pitted against live players in a simple game mode that's been restricted to handguns. The next step, says Nordin, will be training the network to deal with more complex strategies "like teamwork, knowing the map and being familiar with individual classes and equipment." While the agents still get confused or stuck in behavior loops, EA sees promise here for the future of neural networks and machine learning in games.