A full digital replica of a fruit fly’s brain has been successfully transferred to a simulation, where it has been observed displaying natural behaviors
A startup based in Silicon Valley has declared what it calls the first “multi-behavior brain upload,” following the creation of a perfect digital duplicate of a fruit fly’s brain that operates a virtual body within a simulated world.
This milestone, revealed last week by Eon Systems, marks a major advance over traditional artificial intelligence.
In contrast to AI that acquires behaviors via training, the simulated fly’s activities—such as walking, grooming, and searching for food independently—stem from a precise, neuron-for-neuron replica of an actual biological brain.
“This is not an animation. It is not a reinforcement learning policy mimicking biology,” stated Eon co-founder Dr. Alex Wissner-Gross in a social media post announcing the development. “It is a copy of a biological brain, wired neuron-to-neuron from electron microscopy data, running in simulation, making a body move.”
This accomplishment is founded on groundbreaking 2024 research in which a global team charted the entire connectome of an adult fruit fly—all of its approximately 140,000 neurons and 50 million synaptic connections. Eon senior scientist Philip Shiu was a co-author on a Nature paper demonstrating that a computational model based on this wiring diagram could forecast real fly movement with 95% accuracy.
However, that model was essentially a brain without a body to control. Eon has now completed the system by connecting the digital brain to a physics-based simulated fly body using Google DeepMind’s MuJoCo engine.
Sensory information from the virtual world is fed into the replicated brain, neural signals travel through its entire connectome, and motor instructions power the simulated body’s actions. The digital entity’s behaviors are generated by its own internal circuitry, not by pre-written code.
Eon CEO Michael Andregg reported that the uploaded fly attains 91% accuracy in its behavior by relying solely on the connectome’s structure, basic neuron models, and “no hand-tuning, no additional learning algorithms.”
The team at Eon is currently collecting data to try a full mouse brain simulation—approximately 70 million neurons, which is 560 times larger than the fly’s brain. Looking further ahead, the team’s ultimate goal is to endeavor a complete human brain upload.
“The ghost is no longer in the machine. The machine is becoming the ghost,” Wissner-Gross remarked.