Read on Twitter
Some musing on this response from @jpillowtime to the thought-provoking paper on representation by @TBBake @benlansdell & @KordingLab
Other have asked they give a concrete example; so I'll take the example of place cells, and look at each of their criteria - and suggest issues https://twitter.com/jpillowtime/status/1361345263751864320
Other have asked they give a concrete example; so I'll take the example of place cells, and look at each of their criteria - and suggest issues https://twitter.com/jpillowtime/status/1361345263751864320
Criterion 1: correspondence
No problem, place cells correlate with something in the world - their spike count increases at specific locations in the physical environment of the animal
No problem, place cells correlate with something in the world - their spike count increases at specific locations in the physical environment of the animal
Criterion 2: causality
Paper says representation must be causal. I say "could be". Place cells are not causal for the function of navigation. I can navigate using distal cues. Navigation using cues does not need a representation of self-location. But...
Paper says representation must be causal. I say "could be". Place cells are not causal for the function of navigation. I can navigate using distal cues. Navigation using cues does not need a representation of self-location. But...
...place cells will still be active while I am navigating using cues. They are representing the world, but they need have no causal role right now. Hence "could be" causal. (And also representations that are purely acausal e.g. daydreams)
Which bring us to...
Which bring us to...
Criterion 3: exclusivity.
Paper says the neural basis of a representation must be causally unique. Place cells are not the only representation of place in the brain. The brain could in principle use any of those representations to solve a navigation problem. So...
Paper says the neural basis of a representation must be causally unique. Place cells are not the only representation of place in the brain. The brain could in principle use any of those representations to solve a navigation problem. So...
Either: all these are "the" representation, or we cannot have exclusivity. The latter, I think: different neural activity representations of place could be independently causal, and have the same behavioural effect.
(or, the general version: the brain is potentially degenerate, so we cannot demand that a representation be exclusively assigned to one set of neural activity.)
Criterion 4: teleology
The paper says a representation must serve a specific aim or purpose. Place cells seemingly do this - they represent place. But! There are strong arguments their purpose is *not* to represent place but something about state and/or time...
The paper says a representation must serve a specific aim or purpose. Place cells seemingly do this - they represent place. But! There are strong arguments their purpose is *not* to represent place but something about state and/or time...
...It just so happens that in rodents state and/or time are consistently linked to place in the tasks we study them in. So "place" is not their purpose, but they do represent place.
Closing...
Read the paper, it's great. And I'm sure the authors have excellent answers to all the above! (Some of which no doubt I missed in the paper).
Another win for preprints!
(here's the link again: https://arxiv.org/abs/2102.06592 )
End
Read the paper, it's great. And I'm sure the authors have excellent answers to all the above! (Some of which no doubt I missed in the paper).
Another win for preprints!
(here's the link again: https://arxiv.org/abs/2102.06592 )
End
