Battlefield 1’s Self-Teaching AI Bots Don’t Revive Each Other Either

EA’s Search for Extraordinary Experiences Division (SEED) and DICE teamed up to build a self-learning AI-agent in Battlefield 1. A single neural network controls all the characters seen in the video below. But SEED didn’t throw the AI right into the deep end. They let it observe 30 minutes of human gameplay to kickstart its learning in a process called imitation learning.

And it looks like the AI picked up on one Battlefield tendency we are all familiar with. There isn’t a single revive in the video below.

Kidding aside, the tech on display here is cool. Sure, the bots start wandering in circles if there’s no action going on – but some of them are downright lethal. Now, if only they’ll learn that aiming down sights will make them better. Hey, baby steps.

SEED’s Technical Director Magnus Nordin doesn’t expect these Battlefield 1 bots to start out playing the best Battlefield 1 players anytime soon. And that’s not the end goal anyway.

“Our aim is to help create new experiences that enhance games and make them more fun,” says Nordin. “Getting owned by a superior AI isn’t necessarily that fun for players in the long run.”

So, how can developers use this technology now and in the future? In the short term, Nordin sees self-teaching AI as a great tool for quality assurance and testing. Let the AI bots handle the grunt work of gathering crash reports, finding bugs, and hunting down glitches. As deep learning improves, it could bring us NPCs that change over time as we interact with them. Or, be used to help developers create content for their games.

It’s going to be crazy to see where this tech goes in the next five or ten years. Hell, I’m just pumped to see how this tech could improve games from a quality assurance standpoint.