Should we be worried about EA creating self-learning AI? Probably, but it’s very neat anyway. EA has set up a Search for Extraordinary Experiences Division (SEED) that is looking to explore the future of gaming, beginning with developing a self-learning AI agent that’s teaching itself how to play Battlefield 1 multiplayer.

A complex neural network is running the entire match in the footage below. Every single soldier is AI controlled, learning from its experiences and becoming better and better with each and every game it plays.

Hmm, I guess there’s still some learning to be done. The AI can get almighty confused while it learns, although this does help us understand the pain the truly terrible Battlefield players are going through. Looking at the scoreboards there are always three or four people with about 1 kill / 34 deaths who are clearly in need of some machine learning.

Anyway, SEED technical director Magnus Nordin is pulling the strings on this one, explaining that the team has “conducted playtests, pitting AI agents against human players in a simplified game mode, restricted to handguns. While the human players outperformed the agents, it wasn’t a complete blowout by any stretch.

“The agent is pretty proficient at the basic Battlefield gameplay, and has taught itself to alter its behavior depending on certain triggers, like being low on ammo or health. But Battlefield is about so much more than defeating your opponents. There’s a lot of strategy involved, stuff like teamwork, knowing the map and being familiar with individual classes and equipment. We will have to extend the capabilities of the agents further for the AI to be able to crack these nuts.”

The next major obstacle is to tackle the AI’s forward planning. In its current incarnation it’ll shoot on sight, or run towards an objective, but it’s not smart enough to hide and wait for a big group to go past, or to go out of its way to a quiet objective point. The bots here have been subjected to 300 days of total gameplay experience, although there’s a long way to go until they can become top tier players.

All of these problems should be ironed out with time though, and it could well inform offline bot-play for future DICE titles. SEED is thinking a little more outside the box with its plans though, and the first intention is to use these AI NPCs to perform quality assurance and bug testing, letting them loose in endless matches to farm far more crash reports than would be possible from human testers. After this, their “aim is to help create new experiences that enhance games and make them more fun. Getting owned by a superior AI isn’t necessarily that fun for players in the long run.”

Could self-learning bots be the future of offline play? If bots could mimic human behaviour, would there even be a need for human opponents?