## TL;DR
AI in gaming spans from NPC behavior control to procedural world generation and competitive gameplay. AlphaStar mastered StarCraft II; SIMA learns to follow instructions across multiple 3D games without game-specific engineering.

## Core Explanation
Traditional game AI: finite-state machines, behavior trees, pathfinding (A*). Modern: reinforcement learning for adaptive NPCs; neural network-based animation blending; procedural content generation (levels, quests, dialogue). DLSS (NVIDIA) uses AI to reconstruct high-resolution frames from lower internal resolution.

## Detailed Analysis
AlphaStar architecture: transformer-based policy processes entity list (units visible on screen) → outputs actions every 0.4 seconds. Trained on TPU pods for 44 days of real-time gameplay. Google's SIMA (2024) represents the shift toward generalist game agents — a single model playing 9 different games.

## Further Reading
- Unity ML-Agents
- OpenAI: Dota 2 Bot
- Game AI Pro Book Series