What emergent intelligence looks like without labeled data

Almost all of the recent progress in artificial intelligence has come from a specific recipe: collect a large dataset, label it according to a task humans care about, and train a model to reproduce those labels well. It's an extraordinarily effective recipe, and it has a specific limitation — the resulting model can't be more capable than the patterns already present in the data humans chose to label.

A different question

Genesis starts from a different premise. Instead of asking a model to reproduce labeled outcomes, it places digital organisms inside an artificial ecosystem with a simple survival condition and no predefined task at all. What those organisms need to do to keep existing is left for evolution, not annotation, to determine.

Evolution instead of annotation

Across generations, organisms that behave in ways that help them survive and reproduce persist; the rest don't. Given enough generations, this pressure produces behavior — cooperation, competition, division of labor — that was never written into the system by a designer, because nobody specified it. It emerged from the constraints of the environment.

Why this matters beyond the lab

The interesting result isn't any single organism's behavior, but the collective patterns that appear across the population — the kind of coordination that static, individually-trained models rarely exhibit, because they were never asked to survive together in the first place.