谷歌绘制果蝇神经系统，开发者用它玩《毁灭战士》

Serving tech enthusiasts for over 25 years. TechSpot means tech analysis and advice you can trust. The big picture: Google has mapped the brain and central nervous system of a fruit fly, and the dataset is already being used in an unlikely way: to train a simulated fly brain to play Doom and Super Mario 64. The experiments show how quickly a scientific dataset can become a development project once researchers make it publicly available. The MaleCNS v1.0 dataset, announced by Google Research late last week, includes more than 166,000 neurons from an adult male fruit fly. Google worked with HHMI Janelia Research Campus and other researchers on the project, using AI to turn millions of 2D images into 3D reconstructions of neural structures. The work is a major addition to fruit fly research, which has long been used in genetics and neuroscience. A complete map of the brain and central nervous system could help scientists study how neural circuits work and how they influence behavior. It has also given developers a new technical challenge. Software engineer Alex Wormuth connected the fly connectome to Doom, the classic first-person shooter. In Wormuth's setup, each game frame stimulates sensory neurons in the model, which then translates the resulting neural activity into game controls. – Alex Wormuth (@nftechie_) September 6, 2026 // Related Stories Someone got Doom running on a 2009 Canon DSLR camera, and it's surprisingly playable Scientists built a battery that charges almost instantly, but there's a catch "Each Doom frame stimulates sensory neurons. Neural activity is mapped to game controls. Damage triggers a stimulus to two PPL101 dopamine cells as reinforcement," Wormuth wrote. The idea is to see whether the reconstructed neural system can learn to respond to the game environment. When the player takes damage, the model receives a stimulus intended to act as reinforcement. Wormuth has made the code open source and set up a live page showing the training process across the roughly 166,700-neuron model. "Will the fly learn to survive?" Wormuth asked. Developer Jessica Paquette is running a similar experiment with Super Mario 64. A short clip Paquette posted shows Mario repeatedly flying into a wall, suggesting that the system is still in its early stages. Paquette described the project as "literally 100% vibe coded with GPT Astra... just for fun." The code for that experiment is also available. Neither effort should be confused with putting a real fruit fly in front of a game. The developers are working with a digital reconstruction of a biological nervous system. The projects feed game data into mapped neural circuits and then use the resulting neural activity to trigger in-game controls. – Jessica Paquette (@barrelshifter) September 7, 2026 That is what makes the experiments interesting. The connectome was created as a scientific resource, but its release also gives developers a detailed biological neural network they can test in software. Instead of building an artificial neural network from scratch, researchers can examine how a known set of biological connections behaves when exposed to input and feedback. Google said the fruit fly map could accelerate brain research. The immediate Doom and Mario experiments demonstrate another result of releasing such a large dataset: it can quickly become a public testbed for developers. It remains to be seen whether either project will produce a simulated fly brain capable of learning enough to navigate a game. For now, the experiments offer an early look at what can happen when a detailed biological map becomes accessible to software developers. Share this article: See more TechSpot in Google Add us as a preferred source and our reporting shows up first when you search. Add TechSpot Featured on TechSpot An AI-developed drug appeared to make patients biologically younger in a clinical trial Scientists built a battery that charges almost instantly, but there's a catch