One Fly Closer to a Digital Mind
One of the primary goals of artificial intelligence is to synthetically recreate the key capabilities of the human mind: perception, learning, and reasoning. It might seem like the day science achieves this is a long way off, but it may be closer than we think.
Recently, a team from EON Systems PBC took a computational model of an entire brain of an adult fruit fly, complete with 125,000 neurons and 50 million synaptic connections, then set the model loose in a virtual environment.
The results were astounding: a virtual fly that exhibited natural fruit fly behaviours, including moving around in a natural manner, wiping dust from its digital antennae, and snacking on virtual banana slices.
The video went viral, causing some transhumanists to believe that this is proof that human consciousness may one day be uploaded for virtual immortality. Most scientists, however, say that there are a lot more steps to go before we get there.
The Background
In a 2024 study published in Nature, EON senior scientist Philip K. Shiu and his colleagues assembled a central brain connectome – a comprehensive map of connections in the brain – for an adult Drosophilia melanogaster fruit fly. This “wiring map” contained more than 125,000 neurons and 50 million synaptic connections, providing a template for examining the brain’s sensory processing. The researchers said that their complete computational model of the entire fruit fly brain would enable them to study circuit properties of feeding and grooming behaviors.
It did. the 2024 study showed that activating gustatory neurons in the computational model, for example those that sense water or sugar, accurately predicted the neurons that respond to taste, and which are required to initiate feeding behaviors. Researchers also found that using the model to activate neurons in the feeding region of the Drosophilia brain predicted those that elicit motor neuron firing. The researchers validated the latter idea by optogenetic activation and behavioral studies.
The researchers also applied the model to mechanosensory circuits, and found that the computational activation of mechanosensory neurons predicted activation of the neurons of the antennal grooming circuit.
Uploading the Fly
“For decades, whole-brain emulation has been the tantalizing counterpart to artificial intelligence,” wrote EON Systems cofounder Weissner-Gross. “Copy a biological brain, neuron by neuron and synapse by synapse, and run it.”
This is essentially what they did. They also gave the brain a body and an environment to explore.
Researchers took the connectome created in the 2024 study, paired it with an AI algorithm that could match the firing of the virtual neurons to those found in an actual fruit fly to within 95 percent, and the used a digital replica of it to power the fly’s digital body. In addition to the whole-brain computational model from the 2024 study, the project also made use of the NeuroMechFly v2’s embodied simulation framework developed by Sibo Wang Chen et ali, and Özdil et al’s research on centralized brain networks underlying body part coordination.
Now embodied and having its own environment, the virtual fly displayed multiple distinct fruit fly behaviors driven by the emulated brain’s own circuit dynamics. On a light brown field of sand, surrounded by grassy, tree-dotted hills, the simulated fly darts around, then stops to rub its legs together and clean its antennae. It encounters banana slices, which it appears to consume with gusto.
Weissner-Gross writes,
“Sensory input flows in, neural activity propagates through the complete connectome, motor commands flow out, and a physically simulated body executes the output, closing the loop from perception to action for the first time in a whole-brain emulation.”
The Implications
Before this project, researchers had been able to create models of brains without bodies, or bodies without brains. A recent project from DeepMind and Janelia used reinforcement learning to control a simulated fruit fly body, rather than connectome-derived neural dynamics. OpenWorm, which focussed on emulating the C.elegans worm, attempted embodiment of a model brain, but with a much smaller brain–around 300 neurons, rather than 125,000
EON’s project is the first example of a complete emulated brain, derived from a biological connectome, powering a physically simulated body, to engage in multiple naturalistic behaviors.
Writes Weissner-Gross, “If a fly brain can now close the sensorimotor loop in simulation the question for the mouse becomes one of scale, not of kind.”
Of course that scale can be daunting. A mouse brain, with some 70 million neurons, is a lot more complex than that of a fruit fly. And a human brain has somewhere around 86 billion neurons. To give an idea of the complexity of the task, a team from Princeton recently announced that they’d mapped the vision centers of a mouse brain – a task that had taken nine years.
But EON is working on it. According to Weissner-Gross, their team is working on amassing the connectomic and functional recording data they will need to attempt a complete mouse brain connectome. They’re combining expansion microscopy to map every neural connection with tens of thousands of hours of calcium and voltage imaging to capture how those networks activate in living tissue. And they hope to achieve this ambitious goal in the next two years.
Weissner-Gross writes, “The ghost is no longer in the machine. The machine is becoming the ghost.”
But has the experiment actually recreated true consciousness? Is the virtual fly experiencing its digital life in the same way a natural fruit fly experiences its own life?
Two experts say, not so fast.
According to Karl Friston, a neuroscientist and consciousness researcher at University College London, it’s a “category error” to think that the simulation of a conscious system means the existence of a subjective experience. Likewise, another neuroscientist, Anil Seth, says that inferring subjective experience from a simulation is like “someone watching a computer simulation of a rainstorm and concluding that the computer itself must be wet on the inside.”
Still, giving the virtual fly a body and a home represents a significant leap forward. And who knows where the technology could take us from there?