Neuroscience is an experimental discipline which studies "neurons," the electro-chemically active cells in nervous systems and brains. If one wishes to understand a brain, that approach is problematic for two reasons.
The first is that neuroscience claims to be founded on the axiom of "the neuron doctrine," which originally meant that each neuron is a distinct cell metabolically controlling its own filligree of inter-neuronal connections ("neural processes"), as opposed to being part of a continuous mesh. Unfortunately, many researchers in practice assume not only that neurons are distinct cell types (which seems obvious), but that neurons themselves are the atomic units of computation. That cannot be true, because a brain's primary task of continuous 4-D simulation requires a continuous and continuously evolving 3-D simulation medium, a task which cannot be met by discrete units alone.
The second problem is that neuroscience cannot be a principled discipline, because its axiomatic elements are neurons, whose very existence (much less definition and structure) is determined by experiment. This lack of unassailable principles hamstrings theoretical neuroscience in more practical ways as well, because new theories must be made to include neurons (somehow), and theorists must beg their funding and data from experimentalists. These interactions are circular; it is at present very difficult for a brain theory or theorist to become credible unless neurons figure prominently.
But neuroscience is nonetheless a legitimate scientific discipline with a remarkable record, so its findings too must be conformable to any mathematical theory of humanity. The key is to understand that neurons are in fact essential computational elements, merely not the smallest ones. As I argued in Elastic Nanocomputation, the problem of 4-D representation requires a continuous, pseudo-elastic 3-D substrate ("simulatrix") which is not neural. But it also requires elements akin to hydrophones which detect significant amplitude-excursions in that medium (by producing all-or-none events whenever signals cross a threshold) and transmit those events with minimal jitter to other parts of the simulation medium. These elements are realistically neural, and will necessarily produce realistically irregular spikes, because the simulatrix-waves which cause each spike are uncorrelated. (Contrast this realistic irregularity with the unrealistic regularity of spikes produced by the dominant neural model, the "integrate-and-fire" neuron. See Simple vs. Efficient Codes for more detail on the neural-noise paradox.)
While a mathematical theory of humanity finds it difficult to predict any particular property of neurons besides their irregular spikes, it does instead, by construction, accommodate the most self-evident experiment of all, the "experiment" every human engages when feeling her body or looking at the world. If any one of us is confident in the truth of our interpretation of our memory of a measurement performed elsewhere by fallible researchers, vetted by biased referees, and then reported in a profit-seeking journal, we ought to be that much more confident that the world in front of us right now exists in three dimensions. That accurate perception of a 3-D world is not only far more computationally difficult than neuroscientific tasks and stimuli, but it is the only experiment that really matters, both to the theory and to the individual humans whose lives the theory would explain.