The Neuroscience of Intelligence
Page 16
At this point, you may be finding it difficult and confusing keeping in mind all the different brain areas related to intelligence. I know the feeling. Here is something to help. It would be nice to have a table showing each Brodmann area and what that area does. There used to be such tables, but as more data became available, it became clear that any one area typically is involved in more than one function. This had been observed for the g-factor in early neuropsychology studies of lesion patients (Basso et al., 1973) and more recently formalized as multiple demand theory (Duncan, 2010). So how could it be helpful if there is no simple correspondence between one brain area and one cognitive function? Doesn’t that make a complex situation worse? And, it is also the case that the way brain areas are defined is not exact and boundaries can be quite different from one brain to the next. Remember our first law: nothing about the brain is simple. In my view, there is no need for you to memorize all the things any brain area does. Think just about the fact that we are at the point where we can identify a set of brain areas related to intelligence. We are identifying the individual instruments in the orchestra. Learning how they work together to create the symphony of intelligence is a new challenge that requires even better technology and we will discuss one, the magneto-encephalogram (MEG), in Section 4.2.
Let’s recap brain network findings. Two main hypotheses proposed from the first phase of neuroimaging studies from 1988 to 2007, discussed in Chapter 3, were that intelligence was related to brain efficiency and involved multiple areas distributed throughout the brain, especially in a parietal–frontal network. The second phase of neuroimaging studies summarized so far in this chapter has applied more sophisticated image acquisition and analysis in much larger samples to test these ideas. Overall, the weight of results across multiple studies provides considerable, if not overwhelming, support for the parietal–frontal distribution hypothesis (albeit with some modifications) and some tentative support for the efficiency hypothesis based on measures of brain connectivity. Next, we turn to a more detailed examination of brain efficiency including testing actual information flow among brain areas while individuals solve problems on intelligence tests.
4.2 Functional Brain Efficiency – is Seeing Believing?
Following the observation of inverse correlations between intelligence test scores and glucose metabolic rate in the cortex (Haier et al., 1988), we formulated the hypothesis that high intelligence was related to efficient brain activity. In that report, the concept of efficiency was general and included possible characteristics of brain networks, neurons (especially mitochondria), and/or synaptic events. We also speculated that the decreased cortical activation we observed following task practice might result from the brain learning which areas not to use while task-relevant areas worked harder (Haier et al., 1992b). From this rather inexact beginning, it is not surprising that demonstrating a relationship between brain efficiency and intelligence has produced inconsistent results over the years. A subsequent review of the research literature concluded that brain efficiency was moderated primarily by type of task and by sex (Neubauer & Fink, 2009). Up to that time, most brain efficiency studies were based on EEG methods. The graph analyses summarized in the last section provided indirect evidence that structural and functional brain network efficiency was related to intelligence, but the story became more complex as more variables were identified that apparently influenced efficiency.
Two small-sample fMRI studies investigated brain efficiency by comparing cortical activations between high- and average-IQ participants (Graham et al., 2010; Perfetti et al., 2009). These studies are noteworthy for selecting participants for differences in intelligence; high- and low-IQ groups are compared. Most cognitive imaging studies avoid using intelligence as an independent variable because there is a general assumption that all human brains basically work the same way so comparing groups with different IQs would not be meaningful. The validity of this assumption, however, is quite doubtful. When intelligence is considered in the research design of imaging studies, differences are apparent. Both these studies reported generally consistent results. One study concluded that, “When complexity increased, high-IQ subjects showed a signal enhancement in some frontal and parietal regions, whereas low-IQ subjects revealed a decreased activity in the same areas. Moreover, a direct comparison between the groups’ activation patterns revealed a greater neural activity in the low-IQ sample when conducting moderate task, with a strong involvement of medial and lateral frontal regions thus suggesting that the recruitment of executive functioning might be different between the groups” (Perfetti et al., 2009). Similarly, the other study concluded that, “Whether greater intelligence is associated with more or less brain activity (the ‘neural efficiency’ debate) depends therefore on the specific component of the task being examined as well as the brain region recruited. One implication is that caution must be exercised when drawing conclusions from differences in activation between groups of individuals in whom IQ may differ” (Graham et al., 2010). Unfortunately, cognitive studies like these that use intelligence as an independent variable are still exceptions (one additional example will be discussed later in Section 4.4).
Two more recent studies tested the efficiency hypothesis directly with fMRI data. The first one studied 40 teenagers (20 males and 20 females) and incorporated sex, task difficulty, and intelligence in its research design (Lipp et al., 2012). These 40 were selected from a pool of 900 so that the male and female samples were matched on intelligence scores (general intelligence and visual–spatial scores) and, to avoid restriction-of-range problems, each sample included a broad range of scores. During fMRI, each participant solved a set of spatial rotation problems along with control problems. The visual–spatial task activated mainly frontal and parietal areas but, contrary to the efficiency hypothesis, there was no basic finding of inverse correlations between intelligence and brain activation during the task. Activation in the posterior cingulate and precuneus was related to intelligence. These are two of the default network areas proposed as additions to the PFIT (Basten et al., 2015). The authors interpreted this as a possible indication that deactivation in areas of the default network might indicate greater task demands for less intelligent participants. They also found that in the females, higher intelligence was related to greater activity in task-related areas for more difficult problems. In short, as predicted by our first law, the results reinforced the complexity of the efficiency concept.
In the second study of efficiency, Basten and colleagues obtained fMRI in 52 participants while they performed a working memory task of increasing difficulty (Basten et al., 2013). They made an important distinction between two kinds of brain areas: task-positive, where activation increased during task performance, and task-negative where activation decreased during performance. They correlated intelligence test scores to activation in both kinds of areas separately. In the networks formed by task-positive activations, higher intelligence was related to less efficiency. In the task-negative networks, higher intelligence was related to greater efficiency. These opposing findings, similar in male and female subsamples, suggest that whole-brain analyses of the efficiency hypothesis may be more confusing than regional analyses.
Despite the initial appeal of the simple efficiency hypothesis regarding individual differences in intelligence, subsequent research continues to underscore a complex set of issues. On one hand, efficiency remains a popular concept for thinking about neural circuit activity and how it relates to complex cognition (Bassett et al., 2015). On the other hand, the concept has been characterized as so vague as to be useless, although it still has potential explanatory power if better defined and measured (Poldrack, 2015).
We are exploring an approach to measuring efficiency that uses the non-invasive neuroimaging technique based on the MEG. MEG detects minute magnetic fluctuations created as groups of neurons fire on and off. The spatial resolution of this technique is about a millimeter, but the time resolution of a millisecond makes this esp
ecially appealing for studying information flow in the brain. Magnetic signals also have less distortion than EEG signals as they pass through the skull, an advantage for spatial localization of activity. When MEG is acquired while a person solves a cognitive problem, millisecond-by-millisecond fluctuations related to neurons firing can be detected and tracked through the entire brain. There are a number of issues surrounding the interpretation of such fluctuations, but they can potentially provide insight into how individual brains process information during problem-solving.
One group of researchers, for example, used MEG during a choice reaction-time task to assess the timing and sequence of brain activations that might be related to intelligence (Thoma et al., 2006). The task was chosen because choice reaction time is correlated to intelligence (choice reaction time tasks require making a decision about which response is correct; simple reaction time tasks just require a response to a stimulus). Fast reaction times in choice reaction-time tasks, reflecting faster information processing speed, are associated with higher intelligence test scores in many studies; reaction time in simple tasks is not (Jensen, 1998, 2006; Vernon, 1983). The MEG results from 21 young adult males suggested that activation sequences involving a posterior visual processing area and a sensory motor area were related to scores on the RAPM test of abstract reasoning (described in Chapter 1). This was a pioneering use of MEG to study intelligence, but it did not have the advantage of a large sample or a model to test so the complex MEG results are necessarily tentative. Another MEG study of 20 university students investigated efficient information flow during a verbal memory task (Del Río et al., 2012). The results suggested that “an efficient brain organization in the domain of verbal working memory might be related to a lower resting-state functional connectivity across large-scale brain networks possibly involving right prefrontal and left perisylvian areas.” The PFIT was not tested directly, but the results are an encouraging example of the potential for MEG analyses for detecting the sequence and timing of information processing.
MEG is a tricky, expensive technology and not many MEG machines have been available to researchers. This contrasts with MRI methods now widely available. Many psychology departments have one or more MRI machines under their control along with legions of graduate students familiar with sophisticated image analysis software, developed by mathematical experts specifically for cognitive studies. MEG is still very much in development as a research tool. For example, one research group used MEG and fMRI in the same sample to study optimal methods for revealing network connectivity (Plis et al., 2011), and other groups have used data for graph analysis of connectivity (Maldjian et al., 2014; Pineda-Pardo et al., 2014), but intelligence was not a variable in any of these studies. Another research group studied reading difficulties and found MEG activations in three areas correlated to IQ scores, but the sequential timing among areas was not reported (Simos et al., 2014). At this time, the potential of MEG for investigating brain efficiency and intelligence has not been realized but we are working to do so (see Textbox 4.2).
Textbox 4.2: Seeing intelligence at work in the brain
The PFIT hypothesizes that intelligence is related to a specific sequence of activation across specific areas during problem-solving. Generally, the sequence starts in posterior sensory processing areas, travels forward to parietal and temporal association areas where information is integrated, and then moves on to frontal lobe areas for hypothesis testing and decision-making. How often this sequence might be repeated while solving a particular problem could be a key variable related to individual differences in intelligence. The exact areas involved in the sequence could also differ among individuals and so could the timing of the sequence. MEG provides a means to assess the actual sequence and compare it to what the PFIT predicts. For example, individuals with high intelligence test scores might engage a different set of PFIT areas than individuals with average intelligence test scores. Perhaps fewer areas would define a sequence in the high-score group, consistent with efficiency. Or, individuals may engage the same set of PFIT areas in the same sequence irrespective of intelligence, with high intelligence related to a faster speed of engaging or repeating the sequence of areas. We are investigating these hypotheses using MEG in a sample of 32 young adults. Each participant performs four different cognitive tasks adapted for computer administration in the MEG machine (paper folding simple, paper folding complex, inductive reasoning, vocabulary). Each task includes dozens of problems, each with four possible answers. Each item takes a few seconds to solve so the millisecond time resolution of MEG can chart brain activity changes during this short period.
As a first step, we have looked at MEG signals from 300 detectors during the time period of 500 milliseconds just before the participant answers an item correctly by pressing a button indicating which of the four choices is correct. Figure 4.6 shows one problem from the simple paper-folding task (see also Animation 4.2 on this book’s website (www.cambridge.org/us/academic/subjects/psychology/cognition/neuroscience-intelligence) for a demonstration of a paper-folding problem). Figure 4.7 shows a screenshot of MEG activation patterns 500 milliseconds just before the button is pressed for high- and low-IQ groups. The idea is to see whether PFIT areas are engaged in the final stages of thinking as the person decides on the answer. Animation 4.3 shows the MEG changes in the brain activation patterns for these groups while the problem is solved (see website, www.cambridge.org/us/academic/subjects/psychology/cognition/neuroscience-intelligence). MEG movies are visually compelling and show large amounts of data qualitatively. They speak to the maxim that a picture is worth a thousand words. But is seeing believing? The challenge is how to use these images to test the PFIT hypotheses quantitatively, especially person by person.
Figure 4.6 Paper-folding problem example. In the first step, a folded piece of paper appears on a computer screen as on the left side of the middle rectangle. In the second step, a hole is punched in the folded paper as on the right side of the middle rectangle. At that point the third step is the appearance of four multiple-choice answers in the circles that show a hole pattern if the paper were unfolded. Only one of the four is correct. The person presses a button to indicate which of the four they think is correct. Courtesy Richard Haier. (See also Animation 4.2 on the website, www.cambridge.org/us/academic/subjects/psychology/cognition/neuroscience-intelligence) courtesy of The Intelligent Brain, © 2013 The Teaching Company, LLC.
Reproduced with permission of The Teaching Company, LLC: www.thegreatcourses.com.)
Figure 4.7 Screenshot of MEG activation in high (top) and low (bottom) IQ groups at exactly 500 milliseconds before signaling the correct answer to the problem shown in Figure 4.6. Colors show locations of PFIT areas. See also Animation 4.3 on this book’s website (www.cambridge.org/us/academic/subjects/psychology/cognition/neuroscience-intelligence) that shows millisecond changes in brain activity during paper-folding problem-solving (see Figure 4.6) (courtesy Richard Haier).
Here is one way. Figure 4.8 shows the activation average for four individuals with high IQ and the average for four individuals with low IQ in bar graphs every 10 milliseconds for the 500 milliseconds before the button press. Even without a statistical analysis, you can see major differences between the groups – the average group shows more activated PFIT areas. Even within groups the activation patterns are quite different for each individual as shown in Figure 4.9. Such differences are always ignored when group data are averaged and compared. In our view, insights about brain processing and intelligence are more likely if imaging data like these are presented person by person as illustrated in the bar graphs shown in Figure 4.9.
Figure 4.8 Average number of MEG activations (y-axis) in high- and low-IQ groups out of nine left-hemisphere PFIT areas every 10 milliseconds over 500 milliseconds before pressing button for correct answer for a paper-folding problem (see Figure 4.6). The low IQ group generally shows more areas were activated during problem-solving (courtesy Richard Haier).
Figure 4.9
MEG activation in nine left-hemisphere PFIT areas summed over the 500-millisecond period before button press for correct answer to the paper-folding problem (see Figure 4.6). Top row shows two individuals with high IQ (both 132); bottom row shows two lower-IQ individuals (87 and 95, respectively) (courtesy Richard Haier).
As this book goes to press, we are still analyzing the data from this study. We need to analyze the 500 milliseconds after the problem appears to test whether the earliest sensory processing activation sequences are related to intelligence. We may discover that other time epochs are more informative than 500 milliseconds. We also need to examine different items from all four tests and look for any commonalities related to IQ. There is much to do before we can even consider replication in an independent sample. It is too early to characterize the results and many interpretive issues remain. Nonetheless, the use of MEG, especially person by person, illustrates a new approach to investigate whether intelligence relates to individual differences in brain efficiency at the network level. The visual representation of such complex data is powerful, but without a quantitative analysis, caution is required because seeing does not always mean believing.