AI just recreated evolution
Researchers in Sweden built a synthetic world inside a computer and released simple artificial “animals” into it. At the beginning, they had no vision—no instructions, no predefined eye structures, and no programmed blueprint for how sight should develop.
Over generations, something remarkable happened.
The organisms first became sensitive to light. Then they developed directional awareness. Eventually, they formed functioning visual systems capable of detecting objects and navigating their environment. No one told the AI how to build an eye. Vision emerged through variation, selection, and iteration inside a digital environment.
How it worked
The virtual organisms were assigned basic survival tasks: navigating terrain, avoiding obstacles, and locating food. Each generation contained small variations, and those best suited to the environment passed their traits forward. Over time, simple light-sensitive patches evolved into structured visual systems connected to primitive decision-making networks.
What surprised researchers most was that the simulated eyes developed architectures already seen in biology—dispersed photoreceptors, camera-type eyes, and compound eyes—even though the digital world was intentionally simplified. Evolution followed familiar paths.
Why this matters
This experiment isn’t really about artificial animals. It’s about using AI to simulate evolution itself.
Instead of designing systems from the top down, engineers can define constraints, create environments, and allow structures to emerge. This approach enables testing evolutionary pathways and exploring potential solutions long before they exist in nature.
The same method could lead to more adaptive robotics, resilient infrastructure systems, efficient sensor architectures, and self-optimizing AI models.
The bigger shift
This marks a transition from programming systems to evolving systems. When intelligence is used to simulate natural selection, the engineer’s role changes. Rather than specifying every structure, we shape the conditions under which solutions arise.
That is more than artificial evolution. It represents a fundamental shift in how complex systems may be designed in the future.
More next week.
— The Engineering Brief