I thought that the new phrenology (localizing some cognitive activity spatially somewhere in the brain) is dead. Late William Uttal provided empirical arguments to the effect that new phrenology is dead.

This said research is about mouse. Love to see this experiment performed on many human beings, at many sites, but by different research teams, with the the same spatial identity. Meta analysis on these as well. Until then, I am skeptical.

Well, there are certainly specialized brain regions for some tasks - e.g., we know certain structures in the visual cortex are responsible for extracting low-level image features (edges, lines, shapes, etc as well as attributes such as speed and direction). We even have a pretty good idea how they do that - good enough that early neural network research was inspired by it.

I'm highly skeptical as well though that this works for high-level cognitive tasks.

I'm also a bit confused by the way the article describes the task. They describe it as "the link between seeing and doing" as if that were some low-level, easily predictable reflex. But if you put it that broadly, wouldn't that entail almost all cognitive decision making?

Actually, localization is true for the early phases of input systems (sensory) and for the later phases of output systems (motor).

From the published paper:"This strongly suggests that PPC neurons encode choice-related signals related to decision formation or motor planning." 'Decision formation', if seen in isolation, sounds like a higher cognitive process; however, it is 'motor planning'. Here, one has enough freedom to fall back on the motor system and its dependence on the sensory system as well.

If we assume your hypothesis is correct, wouldn’t this research still be useful in the context of a particular mouse? I imagine researchers could identify specialized brain areas first, and then dissect that particular mouse’s brain to see just how those neurons had been spatially organized (assuming this is useful? I’m only a layman).

Admittedly, this sort of technique wouldn’t work well for human subjects.

The method you suggested--"identify specialized brain areas first, and then dissect that particular mouse’s brain"--was popular in old days. Lesion study as it was called. It did not shed any light on higher cognitive processes. Same thing with brain injuries, etc: yes, any damage or lesion is local, but does it settle the issue about distribution vs localization.

Imagine that your popular program starts misbehaving. Then, you remember that your laptop fell from 20 feet high. You see that one of the memory modules is not functioning properly. Sure, you can come up with a test case of this bug. What is the relationship between the behavior and the bad memory module: is it many-to-many, one-to-one, many-to-one, or one-to-many? If it one-to-one, yes, it is localization; otherwise, it is some kind of distribution.

Imagine lobotomy of old days. This was a result of how brain injuries impaired cognitive processes; then some guys thought that 'creating injuries' (or lobotomy) can fix problems on patients with problems.