Coinbase cryptocurrency exchange programmer Alex Wormuth connected a simulation of the nervous system of the fruit fly Drosophila to a trading terminal and taught it to make decisions based on market data. The experiment uses the Drosophila MaleCNS v1.0 brain map, which includes 166,700 neurons and about 25.6 million synaptic connections. Quotes for the BTC-USDC pair from Coinbase are converted into an RGB image and fed into the neural network through 3,335 brightness channels and 811 R8 color inputs. Buying is encoded as a positive stimulus, while selling is a negative one. After processing the activity, a fixed reader forms one of three commands: buy, sell, or hold the asset. An intermediary between the model and the trading system checks the set limits and, if necessary, sends a spot order through Coinbase Advanced.
Currently, the project operates only in a safe environment and does not use real funds. The experiment uses current BTC-USDC prices, but the fly's portfolio consists of virtual $100. According to the developer, a similar experiment can be launched with real money by creating a separate account with a maximum of 100 USDC and an API key with permissions for viewing and trading without the ability to withdraw funds. An additional set of restrictions is set for the neurofly: the size of a single transaction, including the commission, should not exceed $10, and the number of operations is limited to 24 per day. The system does not support leverage and short positions, and after the portfolio value decreases by $20, new orders are automatically blocked. This architecture is intended to limit potential losses during the experiment.
The model's behavior also implies a learning mechanism through changes in the activity of individual neurons. If the portfolio shows a positive result, 15 PAM11 dopaminergic neurons are stimulated. Loss-making operations activate two aversive PPL101 neurons, forming an avoidance reaction to such actions. However, Wormuth emphasizes that the experiment has not yet proven the fly brain's ability to independently develop a profitable trading strategy. The mere fact of synaptic changes does not mean that the model has learned to consistently earn on the market. Therefore, the project remains a research experiment rather than a ready-made trading system. Previously, in another experiment, a bot on Solana turned 0.227 USDC into approximately $696,000 in a single transaction within seconds amid a sharp drop in the value of the ANB memecoin.