In a very fascinating study, Posani et al. analyzed Neuropixels probe recordings of mouse cortical regions during complex tasks. They found that most cortical neurons are not members of neat functional categories, but multi-purpose generalists that organize in populations reflecting their anatomical connectivity.
Except for early sensory areas, neurons exhibited very diverse responses, simultaneously encoding high-dimensional information about multiple distinct variables (e.g. color, shape, orientation). Representation dimensionality appears to systematically increase along the cortical hierarchy.
Paper: Rarely categorical, highly separable representations along the cortical hierarchy | Nature
Accompanying article: https://zuckermaninstitute.columbia.edu/upending-decades-debate-scientists-discover-most-neurons-are-jacks-all-trades
Some very interesting tidbits from the article:
The scientists did find that based on a neuron’s pattern of responses to a given task, they could pinpoint with surprising accuracy which specific module in the brain it belonged to. However, these neurons still generally display generalist behavior.
“For example, compare this to maps of voter opinions,” Dr. Posani said. “There are clear clusters where people generally vote the same way. But when you zoom in, you see mixes of opinions.”
A key implication of these findings is that while each neuron generally encodes multiple variables, it is difficult to decode what any of these variables are from an individual neuron. It is only upon examining populations of neurons, like the brain does, that variables can be decoded. This is a major shift in thinking for the neuroscience community, “which for decades focused on one neuron at a time, discarding all neurons whose responses could not be understood,” Dr. Fusi said.
These results are quite consistent with multiple aspects of TBT. 