Eon Systems recently published an interesting example of what happens when research on AI and the brain moves from modeling neural activity to modeling an entire sensory and motor loop. The company has created a virtual fruit fly that combines an existing computational model of the Drosophila brain with an existing simulated body. The result is a digital fly that can receive sensory information from a virtual environment, process that information through a brain model, produce motor outputs, and receive new sensory information as its body moves.
The demonstration is impressive, but Eon is also explicit about its limitations. This is not a complete digital reconstruction of a biological fly. Instead, it provides a useful look at what embodied brain emulation currently involves and why communicating this kind of research accurately matters.
Why Should We Care About AI and the Brain?
The most interesting part of Eon’s work is not simply that researchers simulated a fruit fly. Researchers have already developed computational models of the Drosophila brain and sophisticated virtual models of its body. Eon’s project integrates these components so that the simulated brain and body interact continuously.
The brain model is based on the adult Drosophila central-brain connectome and contains approximately 140,000 neurons and roughly 50 million synaptic connections. The virtual body comes from NeuroMechFly, which models the fly using 87 independent joints and a three-dimensional mesh created from an X-ray microtomography scan of an actual fruit fly. The body operates within the MuJoCo physics engine.
Connecting these systems creates what Eon describes as a closed sensory-motor-sensory loop.
Essentially, the process works like this:
- Something happens in the virtual environment.
- That event activates identified sensory neurons or pathways.
- The brain model processes the resulting neural activity.
- Selected descending neurons produce motor commands.
- The virtual body moves.
- That movement changes what the fly senses, beginning the process again.
Eon currently synchronizes the brain and body every 15 milliseconds, although the researchers acknowledge that this interval may be too slow for some behaviors.
What Can the Simulated Fly Actually Do?
The demonstration focuses on a small set of recognizable fly behaviors.
For example, the virtual environment can contain appetitive taste cues associated with food. These cues activate gustatory pathways in the brain model and can cause the fly to slow down, turn toward food, and feed. The researchers can also simulate dust accumulating on the fly. This activates mechanosensory pathways associated with antennal grooming, causing the virtual fly to stop and clean itself before continuing toward food.
The model also contains a simulated visual system. However, Eon notes that its current visual activations are largely “decorative” because they do not yet substantially influence most behavioral outputs. The researchers have also examined neural activity associated with escape responses to looming visual threats, but escape behavior has not yet been implemented in the virtual body.
This distinction is important. The fly is performing recognizable behaviors, but that does not mean every part of those behaviors emerges directly from the simulated brain.
Is This Really a Brain Upload?
Eon describes the project as the first “embodied fly upload,” but the company is careful to explain what it means by that term.
According to Eon, an upload does not have to be a perfect digital replica of an organism. The researchers instead describe emulation fidelity as a spectrum. Under that definition, they consider an earlier unembodied Drosophila brain model to be the first fly upload and their current project to be the first embodied version because it closes the sensory-motor-sensory loop.
Eon also acknowledges criticism of this terminology. The behaviors currently depend heavily on existing body controllers, several mappings between neural activity and movement were selected manually, and the model’s internal dynamics have not yet been validated against several known biological processes.
For technical communicators, that distinction matters. Calling something a “brain upload” can imply much more than the underlying research demonstrates. When discussing AI and the brain, terminology should make clear whether we are talking about a complete biological replica, a computational neural model, an embodied simulation, or some combination of these systems.
What Are the Limitations of the Virtual Fly?
Eon provides a substantial discussion of what the current model does not reproduce.
The computational brain uses simplified leaky integrate-and-fire neurons rather than detailed biological neuron models. It does not represent many aspects of neuronal structure and dynamics. Internal states and processes such as hunger, satiety, arousal, learning, plasticity, hormonal changes, recent sensory history, and neuromodulation are also largely absent.
The connection between brain activity and movement is similarly simplified. Rather than modeling the fly’s complete biological motor hierarchy, Eon uses a relatively small set of descending neurons as control signals. Existing lower-level controllers then translate those signals into physical movements such as turning, walking, grooming, or feeding. Some mappings between neural activity and movement have been chosen manually rather than derived directly from biological evidence.
Eon therefore explicitly warns against treating the project as proof that the structure of a brain alone is sufficient to reproduce an organism’s complete behavioral repertoire. The company describes the current system as a research and demonstration platform rather than an endpoint.
What Do Technical Communicators Need to Know?
For technical communicators, Eon’s simulated fly provides an interesting case study in communicating emerging technologies.
First, we need to distinguish the demonstrated system from the larger claims that terminology such as “brain upload” might suggest. The model contains substantial biological information, but it also contains simplified neural models, hand-selected mappings, existing motor controllers, and a limited behavioral repertoire.
Second, limitations are part of the story. In this case, understanding what has not been modeled is essential to understanding what has been accomplished.
Finally, technical communicators need to pay close attention to how researchers define terms. Eon explicitly defines uploading as a matter of degree rather than an all-or-nothing achievement. Accurately communicating the research therefore requires communicating that definition rather than treating “upload” as a universally agreed-upon category.
Research involving AI and the brain is likely to attract attention precisely because the concepts involved are so provocative. Eon’s embodied fly is interesting without turning it into something that the researchers themselves say it is not. At this stage, it demonstrates how a connectome-constrained computational brain can be connected to a physically simulated body and participate in a closed sensory and motor loop.
That is already plenty interesting.
Read the Full Eon Research Update
Eon Systems provides a much more technical explanation of the virtual fly, including the neuroscience research and computational models on which the project depends.
How the Eon Team Produced a Virtual Embodied Fly