# OpenAI's Curiosity-Driven AI Achieves Human-Level Performance on Notoriously Difficult Game

OpenAI has announced a breakthrough in artificial intelligence exploration with Random Network Distillation (RND), a new method that uses curiosity to help AI agents learn more effectively. The system has become the first to surpass average human performance on Montezuma's Revenge, one of the most challenging games for AI to master.

The innovation centers on "prediction-based rewards" that encourage AI agents to explore their virtual environments driven by curiosity rather than just pursuing explicit goals. This mimics how humans learn through experimentation and discovery when facing unfamiliar situations.

Montezuma's Revenge, a classic Atari game from 1984, has long been considered a benchmark for AI difficulty. The game requires complex planning, exploration, and memory—skills that traditional reinforcement learning struggles with because rewards are sparse and