Machine Learning Could Create the Perfect Game Bosses

It is behind you, breathing down your neck. You weave between pillars to collect orbs while avoiding it. It looks like you have lost it for a moment. Your opponent appears around a corner. You are in it.

Something different appears just below the familiar Game Over screen. Please tell us the difficulty level. You didn't last long in this hair-raising chase, so you tapped "hard." It will be less aggressive next time.

Artificial intelligence is joining the battle.

The game I just described at Unreal Fest was demonstrated by two machine- learning specialists. It combines machine learning with an Unreal Engine feature to let developers use spatial data to make decisions.

Behavior trees are usually used to handle this. The machine-learning model is used to drive the behavior of the artificial intelligence. The machine-learning model becomes the brain of the artificial intelligence and decides how it should respond.

The game is not as frightening as I made it sound, mostly because of its top-down presentation, but it is still a fun game that requires players to collect orbs strewn across a map. The ghosts behaviors are no longer scripted.

If a developer decided to use a stronger chase mode, the only thing that could be done was to increase the reference value in the tests. It would be up to a game designer to decide which game variables need to be changed in order to change the difficulty in a game.

Up to a game designer is the key phrase. Designers, programmers, and other developers have to back-and-forth to fine- tune behavior in a traditional behavior tree. Designers can model difficulty without going into branches of a behavior tree by tweaking a machine- learning model. Able to put that aside, designers may be better able to focus on making the game more fun and challenging.

A smarter boss isn't always a better one.

Machines can be used to create enemies. It isn't always desirable to raise the difficulty because it may make the game boring.

A multi-output model that learns to predict the player's score and cut them off was one of the ways that Trachel and Peyrot tried to use artificial intelligence. The enemies tended to camp on the orbs. We didn't show the results because it wasn't enjoyable to play against.