In 2024, the FlyWire Consortium unveiled a comprehensive connectome of an adult fruit fly, detailing all 139,000 neurons and approximately 50 million interconnections. Now, this digital representation of the fly's brain has been activated for practical use. Scientists converted the actual neural wiring diagram of the fly into a graph neural network called FlyGM. Initially, they trained this model to replicate the functionalities of another AI system designed to operate a simulated fly, subsequently refining it through reinforcement learning. The outcome was remarkable; under the governance of FlyGM, the virtual fly acquired the ability to walk, turn, and even fly, all without needing to redesign the network for each different task. Even more astonishingly, distinct functional regions arose within the network autonomously. This behavior was not directly programmed by the researchers; instead, the inherent structure of the fly's brain influenced the system's capability to organize its information processing. Moreover, FlyGM demonstrated quicker training times and greater efficiency compared to traditional AI models. This raises an intriguing question: can we consider this virtual fly to be alive? Some may argue it represents digital life, while others might contend it is merely a collection of code. There are also those who suggest we might already be living in a simulation.
Informational material. 18+.
