We've uploaded a fruit fly. We took the @FlyWireNews connectome of the fruit fly brain, applied a simple neuron model (@Philip_Shiu Nature 2024) and used it to control a MuJoCo physics-simulated body, closing the loop from neural activation to action.
A few things I want to say about what this means and where we're going at @eonsys. ๐งต
It's crazy that this worked. The uploaded fly has 91% behavior accuracy with only 4 things:
- the graph of connections
- the weights as determined by the number of synapses connecting those neurons
- a map of excitatory and inhibitory neurons
- leaky-integrate-and-fire
This shows how much information is captured by the architecture itself, rather than the neuron model, which is great for the feasibility of full emulation.
Some caveats: We can't trace the actual motor neurons because the body was not scanned. However we do know what the brain does when it wants to move in certain ways and that's what we connected to the NeuroMechFly. This is a real limitation of the FlyWire connectome, which is why we plan to scan both the brain and the body.
Another limitation is that we're using Leaky Integrate-and-Fire, which doesn't have any kind of plasticity rules. This fly cannot form long-term memories atm.
Now let's think about the fly.
This is, in our view, a real uploaded animal. We don't know what its experience is โ nobody does. But we take the possibility seriously, and we're working to give it a rich environment, not just a test box.
If you're going to build this technology, you have to care about the beings you create with it. That starts now, at the smallest scale.
But the fly is just the beginning. Here's the bigger picture.
Emulation helps with three things:
- Understanding the brain and treating diseases
- Discovering intelligence algorithms that evolution invented in the most expensive training run in history
- Uploading
We're entering an era of artificial superintelligence. The biggest question isn't whether ASI will arrive โ it's what form it takes and who gets to participate in it.
Right now the default path is: a few labs build opaque AI systems, and the rest of humanity hopes they're aligned. I think there's a better option.
A successful hi-fi upload should feel like you. You that is robust, free from illness and death; editable, can run faster than real time and keep up with AI (transistors are a billion times faster than neurons); and most importantly aligned, with your values, memories, relationships, and moral intuitions.
Brain emulation allows humans to flourish in a world with superintelligence.
We're a small team in San Francisco working on what I believe is the most important technology of this century. Reach out if you want to help build this! ๐ eon.systems
Thank you to the neuroscience community, we're standing on the shoulders of giants! Especially Janelia, @FlyWireNews, @srinituraga et. al, @neuromechfly โค๏ธ
Also thanks to our advisors @robinhanson, @geochurch, Stephen Larson, @KordingLab, @anderssandberg, Ken Hayworth, and many more!