"AI" has been used by gamers for years as a generic term to describe "computer players" but that's not really an accurate assessment. In actuality, "computer players" are typically pre-programmed opponents that follow a distinct set of instructions. They don't truly think for themselves or come up with their own ways of doing things.

Developers, however, are in the process of changing this.

At the Game Developers Conference this week, EA said it has been training AI agents to play Battlefield 1. As The Verge highlights, EA is training its agents using a combination of two methods.

The first, imitation learning, involves the AI agent observing human players and attempting to mimic their actions. According to EA, that constitutes about two percent of an agent's knowledge and gets its set along the right path. The second phase is called reinforcement learning and involves the agent figuring out the rest of the game on their own, collecting rewards for completing various tasks along the way.

Despite their advanced learning capabilities, the AI agents aren't as good as human players. Magnus Nordin, head of EA's Search for Extraordinary Experiences Division (SEED), said agents played against internal DICE developers who, while are good, aren't professionals. The humans beat the bots easily, Nordin said, but it wasn't a total blowout.

Nordin said the bots got pretty good at certain tasks like dodging bullets and adjusting to compensate for gun recoil but lacked in other areas, particularly their "scanning" behavior. "When they don't see any enemies they just start scanning for something to do," says Nordin. The Verge likened it to a dog looking for a treat because it is bored.

EA's hopes the AI research will lead to improved games with tougher, more realistic enemies. It won't happen in the next Battlefield game as that's just around the corner, Nordin said, but perhaps the next one as a hybrid of classical AI and neural networks.