Joscha Bach: Why We Still Can't Simulate a Mouse's Brain

Hinzugefügt: 2026-05-12

[00:00:01]We mapped every single neuron in a worm's brain, 309 of them, and the simulation still doesn't work. Here's why. [music] First of all, I suspect that it wouldn't work. I suspect that the current models of neuroscience do not accumulate.

[00:00:19]I don't think that if you were able to fully capture the connectome of an organism and try to run it in a computer simulation, it's not going to reproduce anything that looks like the behavior of the organism.

[00:00:31]And that's not because I'm woo or think that we need new physics or something like this, but I think that there is a misunderstanding about the role of neurons in an organism. And I'm conscious that this is very heretical and said by an outsider who is not actually a neuroscientist. My um um knowledge of neuroscience doesn't go uh beyond uh that of an undergrad student in the field.

[00:00:53]But uh it's uh When I look at this as a computer scientist, what I mean I look at an individual cell, and for every cell it's true that that it's able to send conditional messages [music] to other cells.

[00:01:07]And this means if I look at a multicellular organism, I look at a Turing [music] machine, at a general computational system that can in principle execute whatever program.

[00:01:16]You don't need neurons for this. Whereas neuroscience is largely working with the simplifying assumption that only neurons are computing and the information is only exchanged [music] by a spike trains and the content of memories is stored somehow in the connections between the neurons. There are experiments which show that you can teach some things to a caterpillar that uh the um butterfly knows.

[00:01:40]And in between the brain of the caterpillar nervous system of the butterfly gets liquefied, its connectome gets dissolved, and then gets reassembled in a different shape. So, how is this information being preserved?

[00:01:52]There is some indication that uh uh, memory might be preserved to some degree in RNA, which means within the cell.

[00:02:00]And also possibly exchanged via RNA across cellular boundaries. Once you are an animal that is able to eat plants to maintain the energy budget to drive your telegraph network, that is very useful to have it because it allows you to control your muscles very quickly because you can send information very quickly through these wires through the organism.

[00:02:18]Building wormholes in the three-dimensional topology of the space of the organism. There's a price that you have to pay for doing this. Well, the signals do not degrade over these long distances. You cannot just send chemicals or look uh, mechanical vibrations or small EM fields to your neighbors [music] as adjacent cells would be doing. Instead, what you need to do is you encode everything into Morse code into spike trains, so it doesn't degrade over long [music] distances. It's a bit awkward, but it pays off because you can basically send it very far and so as a result, it's much faster.

[00:02:47]>> [music] >> And once you move your muscles so fast, you also want to make perception and decision-making at the same rate, so you build yourself a second information processing system to the normal body and the information transmission from cell to cell over long distances. Right? You have this second information system that is [music] much faster and decoupled from the first one because it uses a different language, a different code to translate the information. [music] And so the thing that is talking to you right now is this telegraph network. But the telegraph network would not be functional without all the local operators that are connected to it because it's actually about what's happening around us. And a lot of the information processing is going to [music] happen in the areas around these uh, neurons. And so far I'm skeptical about uh, companies that are promising that they will soon be able to run the connectome of a mouse in a simulation of a fruit [music] fly because we cannot even run C. elegans in a simulation. Like we have pretty good models of the connectome of C. elegans because it's only like 309 neurons and a few thousand connections between them.

[00:03:47]But the simulations of C. elegans in this simulator don't work very well to my knowledge. Maybe it's updated in the last few months, but as far as I know, they they don't produce worm-like [music] behavior. And I think that's because they the other cells are important, too. Right? So, metaphorically speaking, the neuroscientist might be like an alien civilization that has discovered Earth 100 years ago, and they look at the planet, and they discover from their vast distance that they have to Earth that there is this telegraph network that spans the planet, and they can able to intercept signals on the telegraph network and to figure out parts of the Morse code even from first principles, and then they say "Very soon we'll be able to run a simulation of the human telegraph network, and thereby being able to predict and simulate human civilization."

[00:04:34]Yoshua just told me the entire field of neuroscience might be mapping the telegraph network and calling it the civilization. If he's right, what is consciousness?

[00:04:44]My conversation with Donald Hoffman goes even deeper. Watch that next.