I fully support the last sentence of this abstract from @profdata.bsky.social :
https://elifesciences.org/reviewed-preprints/99047
"...the complexity of the brain should be respected and intuitive notions of cell type, which can be misleading and arise in any complex network, should be relegated to history."
🧠📈 🧪
https://elifesciences.org/reviewed-preprints/99047
"...the complexity of the brain should be respected and intuitive notions of cell type, which can be misleading and arise in any complex network, should be relegated to history."
🧠📈 🧪
Comments
One of the least useful activities in #neuroscience is what I call "stamp collecting", i.e., finding neurons with yet-to-be-observed response properties and giving them a name, like "border cells", etc.
It is better to understand the responses as part of a distributed code.
Rather, I think this paper shows that any natural factorizations in the data will be respected in the network (under certain conditions).
BUT, it doesn't show the negative, i.e. that you will *never* get cells that entangle the variables.
But, I am claiming that this way of characterizing neurons leads to misconceptions amongst neuroscientists, and it would be better to resist the urge to focus on imposed functional categories like this.
But, IMO the importance was due to the fact that these were first steps in getting a handle on the function of these circuits.
It was *not* the act of naming these cells, nor the subsequent "stamp collecting" the field did, that was useful.
Rather, I would argue that analyzing everything as distributed codes is actually a *closer* approximation to the bio-reality, and the "stamp collecting" I described above is more like the spherical cow.
That exercise is us imposing names and interpretations on the data.
I’m in the distributed camp myself. But afaik the cell type people don’t claim to explain the entire brain, they focus on reporting what they observe. They also study the relationship of cell types, connections to other regions, collaborate w computational folks.
I think a lot of modelling work has been led astray by assuming that every cell in the hippocampus or EC is of some well-defined cell type (place cells, grid cells, etc). But, most of the cells in real brains don't adhere to these categories!!!
To be clear, the OP was expressing a concern about "functional cell types", e.g. things like "simple vs complex cells", "place vs grid cells", etc.
I think those sorts of categories are us imposing an artificial interpretation on the data.