Fruit Fly's Neural Network Outperforms Advanced AI in Chess
Experiments using a digitized fruit fly brain have yielded surprising results, challenging the supremacy of leading artificial intelligence models.
The brain of a fruit fly, once fully mapped and digitized, has demonstrated capabilities that rival and even surpass advanced artificial intelligence systems, notably in the game of chess. This development stems from scientists mapping all 166,000 neurons and 25.6 million connections in an adult male fruit fly's brain, a dataset released online on September 3. This breakthrough, a collaborative effort by HHMI Janelia Research in Virginia and Google Research, has enabled a wave of creative experiments by researchers and enthusiasts.
One such experiment involved hooking the digitized fruit fly neural network to a chess engine and pitting it against Claude Opus 5, a leading AI model. The fruit fly-brain-powered engine achieved checkmate in just 11 moves, a feat that stunned observers. Creator Maxime Labonne noted that the fly developed strategies for playing and compressing chess that were beyond current human comprehension and had to be stopped from training further due to its performance against established models.
Beyond chess, the mapped neural pathways have been utilized in a variety of other applications. One engineer developed "Stonkfly," a project that equipped the digital fly with a $100 cryptocurrency portfolio and simulated dopamine rewards for successful trades, aiming to foster positive reinforcement. Another project involved a student creating a fighting game where one digital fly was controlled by a human player and another by the fruit fly's neural network. In a king-of-the-hill style competition, the human player lost all 12 bouts against the fruit fly's digital counterpart.
Further demonstrating the versatility of the fruit fly's neural map, one engineer created a blackjack game dubbed "FlyJack." In this simulation, the fruit fly's brain made bets against a digital dealer. The success rate in FlyJack varied with each experiment, mirroring the unpredictable nature of real-world gambling.
The digitized neural network has also been integrated into various custom-built robots, including animatronic cats, crab-like walkers, and small drones, showcasing the broad potential for applying this biological computational model to artificial systems. The accessibility of the fruit fly's complete neural map continues to inspire novel applications and research avenues.