Coinbase Engineer Conducts $100 BTC Trading Experiment Using Fruit Fly Brain

A software engineer at Coinbase has unveiled an open-source experiment that entrusts $100 worth of Bitcoin trading to a system that mimics the neural network of a fruit fly. The project, named 'Stonefly', recreates the neural connections of the fruit fly in a computer system to generate trading signals. The experiment utilizes the 'MaleCNS v1.0' model, which replicates the entire neural network of male fruit flies, comprising approximately 166,700 neurons and 25.6 million connections. The system converts Coinbase's BTC/USDC candlestick chart into an RGB image for visual stimulation, which in turn generates buy, sell, and hold signals. When a profit is made in trading, it stimulates dopamine neurons to create a reward signal, while a loss triggers an avoidance signal. The default setting is for simulated trading, with actual orders limited to about $10 each and a maximum of 24 attempts per day. New orders are halted if cumulative losses reach $20. Currently, there is no evidence that this method can yield stable profits. Stonefly has been released on GitHub under the MIT license, showcasing the potential for combining AI and automated trading.