Google mapped a fly's brain with AI, so the internet did the logical thing: making it play Doom and trading crypto.
With the help of AI, Google was able to map the entire brain structure and central nervous system of an adult male fruit fly. While the company advertises the model consisting of over 166,000 neurons as a "major milestone in neuroscience," the internet considers it the ultimate playground. Turning Google's map into a digital simulation on a computer, creative engineers made the fly trade stocks, take driving lessons, and play video games.
Google's Latest Breakthrough
In order to even begin to understand the enormity of Google's undertaking, one has to go back to the beginning: Google Research and the HHMI Janelia Research Campus, along with other scientific partners, have collaborated for decades to essentially complete three tedious steps. They first had to make individual neurons and the microscopic synapses connecting them visible by physically slicing the fly's brain into microscopic sections. The researchers then used Electron Microscopy to produce millions of high-resolution 2D images that were aligned into a seamless, virtual 3D block using AI. The final step hinges on a highly specialized AI-system called Flood-Filling Networks: Their task is essentially to track which "wire" belonged to which cell, reconstructing the complex 3D shapes of all 166,000+ neurons and identifying the 125 million synapses.
Open-Source Release as a Playground
The open-source release of the MaleCNS v1.0 dataset gave developers and hobbyists the chance to download the 3D connectome and experiment with it. By mapping game pixels to visual sensory neurons and routing motor outputs to digital controls, they port the simulated neural network into modern computational environments to run these 166,000 virtual neurons locally. This opens the doors to all kinds of fun experiments: Enthusiasts have used the neurons as alternative processing engines to play Doom and Super Mario 64, trade cryptocurrencies, and simulate driving tests.
However, the actual use of the dataset goes far beyond playing video games: the "wiring diagram" can advance medical research by serving as an ideal testbed for modeling human cognitive ailments. Furthermore, it can advance the development of energy-efficient, robust AIs.
What is the first thing you would hook this neural network up to? Leave a comment!
