Generalization is basically a pattern discovery process. In neural implementation, pattern is a coincidence of multiple inputs, or presynaptic spikes. Sufficient number of coincident spikes triggers Hebbian learning: “fire together, wire together“ between simultaneously spiking neurons. More precisely, a synapse is strengthened if pre-synaptic neuron fires just before the post-synaptic one.
Most of neocortex is connections between neurons (dendrites and axons), plus their life support. Given limited resources within a skull, there must be a tradeoff between total number of connections and their average length. In other words, a network can be relatively dense, with more connections of shorter average length, or sparse, with fewer total connections of greater average length.
The choice of coincident inputs becomes exponentially greater with the length of connections. Hence, stronger patterns (greater number and closer timing of coincident input spikes) can be discovered. But that range must come at the cost of having fewer total connections, thus less detailed memory. Which requires greater selectivity in learning: longer reinforcement to form and strengthen synapses.
So, other things being equal, there must be a tradeoff between speed and detail of learning, and scope and stability of learned patterns. Relatively dense hierarchy prioritizes speed and detail, I call it a “specialist bias”, while sparse hierarchy selects for scope and persistence: my “generalist bias”.
However, his interpretation that glia a main information processing component in human brain is implausible, I agree with mainstream opinion that they mainly provide support for neurons.
Greater proportion of glia would reduce density of neurons, but enable higher activity and longer-range connections for each of them. Again, that means a sparser neural network.
Very roughly, cortical hierarchy consists of four sub-hierarchies, listed from the bottom up:
- spectrum of primary-to-association cortices, within each of sensory and motor cortices
- posterior sensory and anterior motor cortices, the latter is somewhat higher in generalization
- lateral task-positive and medial default-mode networks, the latter is somewhat higher
- right and left hemispheres, the latter is somewhat higher
“At the lower level, representation is highly concrete and localized, and thus highly vulnerable. Local damage leads to well-delimited sensory deficit. In unimodal association cortex, representation is more categorical and more distributed, in networks that span relatively large sections of that cortex… In transmodal areas representation is even more widely distributed… P. 82: “Thus a higher-level cognit (e.g., an abstract concept) would be represented in a wide network of association cortex…”
In my terms, wider networks imply “sparse bias” on higher levels of generalization.
Another study by Uri Hasson stated: “It is well established that neurons along the visual cortical pathways have increasingly larger spatial receptive field.” and found that “similar to cortical hierarchy of spatial receptive fields, there is a hierarchy of progressively longer temporal receptive windows”.
On the other hand, performance in business, politics, social sciences, and literature (prefrontal cortex?) doesn't peak until late in life. This is probably even more true in philosophy, but performance metrics there are questionable. Also supportive is the observation that cortical development sequence is delayed by several years in subjects with ADHD. Obviously, effective generality of discovered concepts, thus also development of higher association areas, depends on attention span.
Buschman & Miller of MIT, ref:“ have found two types of attention in two separate regions of the brain. The prefrontal cortex is in charge of willful concentration; if you are studying for a test or writing a novel, the impetus and the orders come from there. But if there is a sudden, riveting event—the attack of a tiger or the scream of a child—it is the parietal cortex that is activated. The MIT scientists have learned that the two brain regions sustain concentration when the neurons emit pulses of electricity at specific rates—faster frequencies for the automatic processing of the parietal cortex, slower frequencies for the deliberate, intentional work of the prefrontal."
I think lower frequency here is due to longer feedback loop of higher levels.
Left hemisphere represents higher-generality and long-term-goal- associated concepts, while the right one mostly searches in the background, for lower-level contextual patterns (Cortex & Mind, p. 184, Split Brain, Gazzaniga). According to my premise, left hemisphere should be relatively “sparse”, which is supported in “Cortex & Mind“, p 185: “Pyramidal cells in language areas have been found to be larger on the left than on the right (Hayes & Lewis, 1995; Hustler & Gazzaniga, 1997)”. Their dendritic trees also extend further than those of right-hemisphere pyramids (Jacobs & Schneibel, 1993).
Hemispheres are densely interconnected by Corpus Callosum. This is partly for sensory-motor field integration and duplication (fault-tolerance). But greater “lateralization” in humans, vs. other primates, suggests that our hemispheres also combine in a deeper hierarchy of generalization. Finish study found that ambidexterity (correlated with lesser lateralization) doubles the risk of ADHD and lower academic performance in children.
Related study "Comparison of the Minicolumnar Morphometry of Three Distinguished Neuroscientists and Controls" by Casanova is reported in "Minicolumns, Genius, and Autism". The connectivity pattern of the neuroscientists appears to be similar to autistics in the density and size of minicolumns, but different in better inhibitory isolation between adjacent minicolumns. This should compensate for smaller size, while enabling greater number of minicolumns.
The thesis of local vs. global connectivity bias in autism is also supported in Exploring the Folds of the Brain--And Their Links to Autism by Hilgetag and Barbas: "in autistic people, communication between nearby cortical areas increases, whereas communication between distant areas decreases".
Such cortex should be more reliant on cortico-thalamo-cortical vs. cortico-cortical connections, which might be the implication in Partially enhanced thalamocortical functional connectivity in autism.
Either way, more numerous synaptic spines increase density of connections in the cortex, which must ultimately come at the expense of their effective range.
Truly pathological autism probably requires more than increased synaptic density and activity. The ultimate cause might be something as basic as pre|post- natal viral infection or retroviral expression, combined with low vitamin D levels (as is likely the case for schizophrenia and bipolar disorder).
Duke University study found a more direct “sparse” risk factor for schizophrenia: increased synaptic pruning "‘Spine pruning theory is supported by the observation that the frontal brain regions of people with schizophrenia have fewer dendritic spines, the tentacles on the receiving ends of neurons that process signals from other cells". But this increased pruning happens during puberty, probably secondary to increased testosterone, vs. reduced pruning at 3-4-year-old in autism.
Schizophrenia seems to be a uniquely human disorder, and there must be a reason these risk factors evolved. Other things that are unique for humans are large neocortex, complex society, and long life.
I think this risk and benefits are closely related: decreased density leaves more space and resources (such as astrocytes) for remaining neurons and connections, so they may grow longer. Which enables global intellectual integrity, thus dynamic social coordination, and long-term planning in general.
More specific “sparse disorder” may be dyslexia. This connection was also made by Manuel Casanova: “Autism and dyslexia: A spectrum of cognitive styles as defined by minicolumnar Morphometry“, although there is a lot less research on that. Basically, he thinks that dyslexia is caused or exacerbated by a “lossy” cognitive style, secondary to sparse connectivity, at least in language-oriented cortices.
- On one hand, speed & precision was more important for survival in the wild, which may explain why apes seem to have photographic memory, superior to humans: Chimps beat humans in memory test.
- On the other hand, more recent functional differentiation of modern society once again requires increasingly “lossless” knowledge acquisition. Social positions that do require higher generalization are relatively few, including law, management, politics and related academic disciplines.
Needless to say, this write-up is motivated by introspection.