Virtual animals grow vision from scratch

Evolution inside a machine

The researchers designed simple digital organisms and placed them in virtual environments. Each creature had a body, basic sensors and a simple “brain.” They were given tasks such as moving through a maze, finding food or avoiding harmful objects.

Those that performed better were more likely to pass on their traits to the next generation. Small random changes were introduced over time, mirroring natural selection. What happened next surprised even the researchers.

Depending on the task, the artificial animals developed different types of visual systems. Creatures trained to navigate obstacles evolved multiple small light sensors spread around their bodies, giving them wide awareness of their surroundings. Those trained to recognize specific objects developed forward-facing, higher-resolution eyes that concentrated on detail.

From pinholes to lenses

In another experiment, the team explored how more advanced eyes might emerge. At first, the digital creatures could only adjust the size of their pupil, similar to a simple pinhole camera. Smaller openings improved sharpness but reduced the amount of light.

When the researchers allowed the creatures to evolve light-bending elements similar to lenses, a breakthrough occurred. Over generations, the system discovered shapes that focused light more efficiently. The virtual animals were then able to see clearly while still collecting enough light, overcoming a major limitation of simple eyes.

The findings suggest that lenses in real animals may have arisen as a natural solution to the tradeoff between clarity and brightness.

Bigger brains, sharper sight

The study also examined the link between eye quality and processing power. The team found that improving visual sharpness only boosted performance if the artificial brain grew as well. Simply increasing computing capacity without better visual input did not help.

Across tasks such as navigation, object detection and tracking moving targets, performance followed clear patterns. High-precision tasks required both sharper vision and more processing resources. In tasks that relied on motion over time, short-term memory could partly make up for smaller brains.

These results echo patterns seen across animal species, where advanced visual systems are often paired with larger brains.

A new tool for biology and engineering

Rather than replaying the exact history of life on Earth, the system models the basic forces that drive evolution: variation, selection and adaptation. By changing one factor at a time, researchers can test “what if” questions that are impossible to explore in nature.

The team says the approach could help biologists better understand why certain eye designs are common while others never appeared. It may also guide engineers in building more efficient and adaptable vision systems for robots and other machines.

Earlier, Qazinform News Agency reported that your AI may be poisoned and you would never know.