💡 Scientists Taught a Digitized Fruit Fly Brain to Control a Virtual Fly

In 2024, the FlyWire Consortium released the complete connectome of an adult fruit fly—a detailed map of all 139,000 neurons and roughly 50 million connections between them. Now, that digital brain has been put to work.

Researchers turned the fly's actual neural wiring diagram into a graph neural network called FlyGM. They first trained the model to mimic another AI system that controls a simulated fly, then fine-tuned it with reinforcement learning.

💻 As a result, under FlyGM's control, the virtual fly learned to walk, turn, and even fly—without redesigning the network for each new task.

Even more surprisingly, distinct functional regions emerged within the network independently. The researchers didn't explicitly program this behavior; the fly's brain structure effectively guided the system in organizing information processing. FlyGM also trained faster and performed more efficiently than conventional AI architectures.

Can we call this virtual fly alive?

❤️ — Yes, it's digital life
🔥 — No, it's just code
😎 — We're already in a simulation anyway

@hiaimediaen