# OpenAI Discovers Decades-Old Algorithm Rivals Modern AI Training Methods

OpenAI has announced that evolution strategies (ES), an optimization technique that has existed for decades, can match the performance of standard reinforcement learning on challenging AI benchmarks while offering significant practical advantages.

The research team tested ES on popular benchmarks including Atari games and MuJoCo physics simulations—the same tests used to evaluate cutting-edge reinforcement learning systems. Surprisingly, the older approach performed just as well as modern RL techniques.

More importantly, evolution strategies overcome several limitations that make reinforcement learning difficult to work with in practice. While OpenAI didn't detail all the specific advantages in the tweet, ES typically offers better parallelization across multiple computers and doesn't require backpropagation, making it simpler to implement and scale.

This discovery matters because it suggests the AI field may have overlooked simpler, more practical solutions in favor