Coinbase engineer’s fruit fly brain AI gains $1 trading Bitcoin

A digital simulation modeled after a fruit fly’s brain has achieved a modest gain in Bitcoin trading, in an experiment conducted by Coinbase software engineer Alex Wormuth. Wormuth allocated $100 to the project, which acts as an unusual trader by leveraging the neural wiring of a fruit fly mapped by recent neuroscience research.

AI-powered fruit fly simulation enters crypto trading

The experiment, named “Stonkfly,” utilizes a highly detailed digital model of an adult male fruit fly’s nervous system. This model is based on the MaleCNS v1.0 connectome, a comprehensive blueprint of the fly’s central nervous system released earlier this month by an international group of scientists, including Google Research.

While Stonkfly uses advanced neuroscience, there is no physical insect at a trading desk. Instead, the system operates by receiving live BTC-USDC market data from Coinbase. This real-time data is transformed into an RGB image, simulating how the insect would perceive visual information. The processed image is delivered to thousands of simulated sensory neurons, mirroring actual insect brain activity.

One day into the experiment, Stonkfly’s trades yielded a $1 profit. However, Wormuth noted that this minor gain is likely the result of chance rather than consistent learning, since the system’s trading decisions are closely linked to both market shifts and the randomness of its simulated neural responses.

Stonkfly uses the entire adult fruit fly connectome, mapping real-time Bitcoin price data to simulated sensory input. Trading outcomes trigger specific dopamine or aversive response neurons according to profit or loss, respectively.

Mini dictionary: Connectome, a comprehensive map of neural connections within an organism’s nervous system, often used in neuroscience to model and simulate brain activity computationally.

How Stonkfly’s simulated brain trades

Stonkfly’s mechanism involves activating 15 simulated PAM11 dopamine neurons when the trading portfolio records a gain, mirroring reward signaling in living insects. In contrast, if a loss is registered, two PPL101 neurons, which are linked to aversive responses in the fly, are triggered.

Despite these intricate mechanisms, the current system has not demonstrated reliable learning or decision-making autonomy. The initial $1 profit may simply reflect natural price fluctuations and random trades rather than true adaptation or strategy.

Background and future applications

The publication of the complete adult male fruit fly connectome marked a major advance in neuroscience. The map includes 166,700 neurons and defines the intricate wiring running through the insect’s brain, optic lobes, and ventral nerve cord. Connectome projects like this enable highly detailed simulations for scientific and experimental purposes.

Wormuth, who works at leading US crypto exchange Coinbase, has previously engaged with the MaleCNS model, most notably by linking it to the classic video game Doom in a project called DOOMFLY. Other developers have made similar integrations, connecting the fruit fly simulation with games such as Beat Saber, Super Mario 64, Minecraft, and Pong.

Although the simulation models intricate wiring, significant elements of biological brains remain beyond current technology’s grasp. Features like chemical neurotransmitters, gene expression, and broader modulatory processes are absent from these digital connectome models, and cannot yet be recreated outside of real organisms.