25 September 2026
Heard In AI

A mapped fly brain is not an uploaded fly, and Alex Wissner-Gross, who helped found Eon, wants virtual flies treated humanely

On Moonshots, recorded September 16, 2026, computer scientist Alex Wissner-Gross said he supports "uploading" fruit flies. He also said he opposes making uploaded flies day-trade Bitcoin or play Doom. He helped found Eon Systems, one of the two projects he discussed. Google's September announcement describes a wiring map of a male fruit fly's nervous system. Eon's March account describes a simplified brain model driving a simulated body. That model has no learning and relies on many hand-chosen links between the brain model and the body. Neither team's account shows that a virtual fly can suffer.

A briefing reports one development at a point in time. We may correct or clarify it later; a new development gets a new briefing. How our formats work

Based on Moonshots with Peter Diamandis, episode published 19 September 2026

The question from a listener was short: what did the hosts think of "the Connectome fruit fly"? On an episode of Moonshots with Peter Diamandis, recorded on September 16, 2026 and published three days later, computer scientist Alexander Wissner-Gross gave a short answer. It mixed a claim about his own company, general enthusiasm and an ethical line he does not want crossed.

Wissner-Gross assumed the question was about what he called "the Google-uploaded fruit fly." He then pointed out his own interest. Eon Systems, "a company that I helped found," had announced what he described as "the first multi-behavior fruit fly upload" months earlier, "actually beating by many months the Google fruit fly upload."

His position is easier to follow once you separate the two projects he grouped under one word. Google's project maps a fly. Eon's project runs a fly-like model in a virtual body. Neither shows what, if anything, it is like to be that fly.

What Google mapped

A connectome is a wiring diagram: a map of which nerve cells, or neurons, connect to which. In an announcement dated September 3, Google described a years-long collaboration with HHMI Janelia, researchers at Cambridge and others. The team mapped the brain and ventral nerve cord of an adult male fruit fly. The ventral nerve cord is roughly the insect's equivalent of a spinal cord. The reconstruction contains more than 166,000 neurons, sorted into 11,691 neuron types.

To build it, researchers imaged very thin slices of tissue, reconstructed the cells by computer and had experts classify them. Because the brain and nerve cord are connected in one map, researchers can trace routes from sensory input toward movement. These include circuits for courtship that differ between males and females. According to Google, the map also supports companion studies of vision, taste and social behavior.

This is anatomy. It shows where the connections run, and scientists use it to study how the nervous system works. The map on its own does not move, eat or respond to anything. Google's page was updated on September 21, after the episode was recorded.

What Eon simulated

Eon's project, described in a technical account dated March 10, makes a different kind of claim. The team did not build the brain model itself. It started from an existing simplified neuron model published by Shiu et al., built from the FlyWire fly connectome resource, and connected it to NeuroMechFly, a previously published simulated fly body running in MuJoCo, a physics engine that calculates how bodies move and collide. Eon describes its system as an integration of these previously published components.

The interesting part is the loop between brain and body. Sensory signals enter the brain model. Selected output signals, like those that would travel from a real brain down to the body, influence how the virtual body moves. That movement then changes what the model senses next. Eon showed the simulated fly feeding and grooming.

Eon's own account lists important simplifications:

  • Hand-built connections to the body. The model controls movement through a small set of neural outputs. These feed body controllers that had already been trained by imitation, and many of the links between brain signals and body actions were chosen by hand.
  • Limited vision. Activity related to vision did not yet substantially affect behavior, and escape behavior was not built into the body.
  • No learning. The model had no learning or plasticity, meaning its connections could not change with experience. Hormonal effects and much of the fly's internal state were also missing.
  • Unchecked internal activity. Eon had not yet compared several important features of the model's internal activity against measurements from real flies.

The authors call the work an early integration and a testbed for improving how brain models connect to bodies. So "multi-behavior upload" is Wissner-Gross's term for a system in which a model built from a wiring map produces a few recognizable behaviors in simulation, with considerable human engineering in between.

The dates also need care. Eon's account came out in March and Google's in September. But one project is an anatomical map of a male fly and the other is an embodied simulation. The earlier date does not mean one team beat the other at the same task.

Wissner-Gross's line: no Bitcoin, no Doom

Wissner-Gross said he is "broadly supportive of uploaded fruit flies," adding, "I think we don't upload enough fruit flies." He went further: "we need to upload substantially all of our biosphere."

He then drew a line. "I am not supportive of torturing uploaded non-human animals," he said. He described writing in his newsletter about uploaded fruit flies "being forced to day-trade Bitcoin," and said some were "being forced to play Doom," the classic video game. He opposes both. His condition for support was that uploaded flies be "treated well and humanely or fruit fly-ly." If that holds, he said, "this is terrific and we should do a lot more of it."

The Bitcoin and Doom examples come from his own account. The episode did not say who built those systems or how they worked. The ethical point also stands apart from the technical record. Neither Google's map nor Eon's simulation, as described by the teams, shows that a virtual fly has experiences or can be harmed. Eon's model lacks learning and much of a real fly's internal state.

That leaves three claims that are easy to blur together. A wiring map shows anatomy. A simulation connected to a body shows that some behavior can be reproduced. Whether that simulation has anything like experience is a separate question that neither project addresses. Wissner-Gross's condition speaks to that third question without waiting for an answer: his support for building virtual animals from real nervous systems depends on how those animals are treated.

The exchange ended quickly. Host Peter Diamandis joked that Alex is "the patron saint of agents and food supplies" and turned to co-host Dave Blundin, who went back to an earlier listener question about AI and mathematics.

Share this article

Go to the original

Sources & further reading

  1. 01
  2. 02

Connected ideas and articles

From the conversation

Podcast episodes

Article history

Updates to this article

Tags

Rothblatt expects a court case to grant AI legal personhood by the 2030s

On Moonshots with Peter Diamandis, Martine Rothblatt said she believes today's AI models are already conscious to a degree, and predicted that a court will recognize a "cyberconscious" individual as a legal person no later than the 2030s. She sketched the test she imagines—documentation that the system is conscious and values its own life—and forecast that digital minds will eventually outnumber biological ones. Physicist Brian Greene, covered earlier, reads the same machines differently: he doubts chatbots have feelings now, accepts machine consciousness may be possible later, and offers no timetable.

5 min read

An agent built a simulation inside its simulation — and the panel argued over what it proves

On Moonshots with Peter Diamandis, the hosts played clips of three demonstrations attributed to Matt Schumer: a prompt-built Manhattan, agents that started talking to each other in order to cooperate, and an agent that sat at a simulated computer and made its own simulation. The panel split over whether nested worlds shift the odds that we live in one, what would follow if they did, and whether the characters inside eventually deserve consideration.

· Updated 7 min read

GPT-6 Astra tops a simulated vending-business test, but Dave Blundin says founders still have to manage the work

On an episode of Moonshots with Peter Diamandis recorded September 16, 2026, the hosts discussed three tests of AI-built businesses. The first was a GrokBot company-building livestream that Peter Diamandis described. The second was Andon Labs' Vending-Bench 2, in which GPT-6 Astra averaged about $15,515 at the end of a simulated year that began with $500. The third was the Build with Gemini XPRIZE, which requires real users and revenue. Salim Ismail stressed that the vending result came from a simulation, not a real company. Dave Blundin argued that AI fails at big tasks until a human provides a framework and breaks the work into small pieces. In his view, that makes entrepreneurs more powerful but leaves them with the job of managing the work.

7 min read