After a Fruit Fly Brain Was Put Into a Computer, It Began Tasting Sugar and Cleaning Itself

According to information posted by Greg Isenberg on X, scientists copied the entire brain of a fruit fly into a computer and connected it to a virtual body. No one wrote it a script to "find sugar to eat" or "clean its body." As described in the original post, it began responding to sugar and performing grooming movements.

First, let's draw the boundaries of the facts clearly. What is copied here is the connectivity structure of the fruit fly brain. Uploading a fruit fly's consciousness to a computer is another matter. The virtual body, sensory input, and motor output still had to be built by the experimenters. A fruit fly brain has roughly over 100,000 neurons, while a human brain has about 86 billion, a huge difference in scale. This demonstration cannot prove consciousness, nor can it be directly extrapolated to humans.

What is worth looking at is the way the behavior appears. We are used to writing objective functions, reward functions, and action libraries for AI systems, and then training them. This demonstration draws attention back to the neural wiring diagram itself: first there is a connectome, then a virtual body is provided, and we see which behaviors emerge on their own. The original post emphasizes that "no one programmed it to do these things"; this sentence should be understood in terms of specific behaviors. The experimenters of course built the simulated environment, but feeding and grooming were not written as scripts.

In the video, the virtual body responds to a sugar stimulus, and then grooming-like movements appear. For ordinary readers, it looks like a little bug doing two things without instructions. For people working in AI, the more worthwhile question is: which behaviors come from circuit structure, which from sensory feedback, and which from the dynamics of the virtual body. Only by separating these three layers can we know what exactly "emergence" is emerging.

This type of work most easily makes people skip the intermediate steps and think directly of consciousness uploading. The intermediate steps are precisely the point. Connectome data, neural simulation, virtual body, sensory encoding, and motor output—each layer involves human design. Behavior looking autonomous does not mean the system has subjective experience. The fruit fly is far from the human brain, but the question is very close: given a neural wiring diagram plus a body, how much behavior can grow out of it?

The original post is here: Greg Isenberg's tweet. The video is also placed above. Next, one can look at the more detailed experimental setup: how the sugar stimulus is encoded, how grooming movements are judged, and what feedback the virtual body has. There is no rush to talk about consciousness; first answer these questions clearly.