Google Research’s New AI, AlphaEvolve, Is Inventing New Mathematics

It’s not just solving problems humans find hard. It’s discovering fundamental mathematical objects that were completely out of our reach.

Photo by Karollyne Videira Hubert on Unsplash

We often hear about AI creating stunning images, writing human-like text, or generating code. That’s cool, but it’s becoming normal. What’s not normal is an AI diving into one of the most abstract and difficult fields of mathematics and computer science “complexity theory” and making genuine, novel discoveries.

I just finished reading a research paper from a team at Google DeepMind and UC Berkeley, and honestly, it’s one of those things that subtly shifts your perspective on what AI is capable of.

The paper is called “Reinforced Generation of Combinatorial Structures: Applications to Complexity Theory,” and it introduces an AI agent named AlphaEvolve. This isn’t just another language model. It’s a system designed to discover new mathematical structures that can help solve problems that have confused researchers for years.

No hype here. Let’s just break down what they did, how they did it, and why it’s a much bigger deal than it sounds.

First Off, What the Heck Is Complexity Theory?