G is general intelligence, a common property we see in the similar test scores that people show for different cognitive tasks. But just what is this common property that makes some people generally smarter than others? There have been attempts to identify g with a specific brain characteristic. Unfortunately, as Anderson (1995) notes:

In general, these attempts have all been failures. It has always been possible to show a disassociation between any putative single psychological process and measures of general intelligence. For example, while it is possible to show correlations between g and memory measures, it is also possible to show normal intelligence in people with severe amnesia, thus eliminating the possible equating of memory and intelligence. While vocabulary measures correlate with IQ, subjects with global aphasia can have normal IQ and subjects with mental retardation can show semantic proficiency and precocity.

Whatever g is, it seems to be some general property and is not eliminated by damage to one brain area. Miller (1994) suggested that g might correlate with myelin, i.e., the fatty sheath that surrounds neurons. More myelination means faster nerve conduction, quicker reaction time and, ergo, higher intelligence. Regrettably, this hypothesis no longer seems tenable:

While IQ correlates with reaction time (RT) encouraging the hypothesis that neuron conduction velocity accounts for the individual variation in IQ, there is also discouraging information. Correcting IQ-RT correlations for neural conduction velocity does not diminish the strength of the relationship and neural conduction velocity does not correlate with RT. (Anderson, 1995).

Barrett and Eysenck (1993) have also failed to find a significant correlation between measures of nerve conduction velocity and IQ.

In his review of the literature, Anderson (1995) discounts other candidates: neuron number (no empirical support); cerebral cortical cell number (does not correlate with problem-solving performance in rats); and volume of the cerebellar granule cell layer (does not correlate with attention to novelty in rats. He concludes that the likeliest candidate seems to be the range and extent of neuronal processes that alter brain connectivity:

Dendritic arborization has been correlated to educational attainment and been shown to be more complex in brain regions critical for language. The molecular layer volume of the cerebellum, which correlated with attention to novelty in rates, is largely composed of Purkinje cell dendritic arborizations. Synapse number correlates with dementia severity in Alzheimer disease. Further, a change in connectivity can explain the IQ-RT correlation and the brain size-IQ correlation. (Anderson, 1995)

Thatcher et al. (2005) come to a similar conclusion in their comparison of EEG measurements to predict performance on the Weschler Intelligence test:

… it is hypothesized that general intelligence is positively correlated with faster processing times in frontal connections as reflected by shorter phase delays. Simultaneously, intelligence is positively related to increased differentiation in widespread local networks or local assemblies of cells as reflected by reduced EEG coherence and longer EEG phase delays, especially in local posterior and temporal lobe relations. The findings are consistent with a ‘network binding’ model in which intelligence is a function of the efficiency by which the frontal lobes orchestrate posterior and temporal neural resources.

Finally, we should not assume that IQ captures all variation in cognitive performance. In general, IQ tests involve answering a series of discrete questions over a limited span of time. Yet this kind of cognitive task is only a subset of all possible tasks that confront the human mind.

For instance, when something puzzles me, I may think about it over several days or longer. It will often sit in the back of my mind until it is re-activated by a piece of relevant information. Sometimes, I will get up in the middle of the night to jot down a possible answer. Then there are the lengthy, monotonous tasks: driving non-stop to Montreal, transcribing old hard copy into an electronic file, keeping track of different ‘things-to-do’ over the course of a day, and so on.

How well does a one-hour test measure performance on such tasks?