Refining AI Training for a Better Player Experience

I’ve noticed the in-game AI seems to learn heavily from overly competitive or frustrated players (likely due to unbalanced matchmaking or gameplay issues). Instead of just criticizing, I’d suggest exploring methods—like clustering algorithms—to identify and filter out problematic player behaviors from training data.

Much like in real-world sports, not all behaviors are worth replicating as some are just noise. If the AI is trained traditionally, it’s probably training on the average player’s habits, including impulsive or aggressive tendencies (whether from younger playerbase, lag frustration, or other factors). Since it prioritizes high-frequency data, it may unintentionally reinforce toxic or one-dimensional playstyles.

A more focused training approach could significantly enhance the player experience. By prioritizing data from skilled, intentional gameplay, the kind that plays for fun and is fun to play with or against, but maintains disciplined like how all athletes play then we'd get much better results.

Would love to hear if others think this approach could work—or if there are better solutions out there.