The Neuroscience of Intelligence
Page 12
All this time, I was applying for federal grant money to fund a brain-imaging program to study possible influences on intelligence. As explained in the last chapter, intelligence research from a biological viewpoint was viewed with some suspicion, and my applications were going nowhere. So, I decided to shift emphasis a bit and I was able to get a grant to study Down’s syndrome, a genetic disorder typically associated with low IQ. These individuals would be of inherent interest and so would the requisite normal control group. Federal agencies are more inclined to fund research on disease and syndrome categories (and stupidity is not yet a category recognized by the National Institutes of Health, so there is no national institute to study it), especially if the grant application barely mentions IQ. By the way, this is still largely the case, although there is an emerging exception for projects that propose to increase IQ in disadvantaged children by means of cognitive training. We will discuss this more in Chapter 5.
We had been wondering if low-IQ individuals might have inefficient brains, possibly due to a failure of neural pruning, the normal developmental reduction in excess or extraneous synapses starting about age 5 years. We were interested in scanning people with Down’s syndrome who had IQs between 50 and 75, and of course, control groups of people without Down’s syndrome who also had IQs in the same low range for no apparent genetic or brain-damage reason. We also had other controls with IQs in the average range (Haier et al., 1995).
At the time, most researchers predicted that PET scans of low-IQ individuals, especially those with known brain abnormalities like those found in Down’s syndrome, would show lower activity because some kind of brain damage was assumed to be responsible for low IQ. A failure of neural pruning, however, was consistent with earlier research in Down’s syndrome showing a higher density of synapses (Chugani et al., 1987; Huttenlocher, 1975). Based on the efficiency hypothesis and a possible lack of neural pruning, we were open to the possibility that we might see higher activity in the low-IQ groups. Figure 3.3 shows this is what we found.
Figure 3.3 PET images in two low-IQ individuals showing higher brain activity than a control with average IQ. Red and yellow show greatest activity in units of glucose metabolic rate (courtesy Richard Haier).
The two PET images on the left show more activity (red and yellow) throughout the brain in both low-IQ groups compared to normal controls on the right. We saw this as more evidence for the efficiency hypothesis, although we recognized alternative interpretations, including compensation for possible brain damage (Haier et al., 1995).
3.3 Not All Brains Work in the Same Way
By this time, I had negotiated for one free PET scan for every scan I paid for with grant money, so we turned to a different way to investigate the efficiency hypothesis. Recall in Chapter 1 I talked about the Hopkins study of mathematically precocious students that Professor Julian Stanley started in the 1970s. The early talent searchers found many more young boys than girls with high SAT-Math scores. In 1995 I decided to use PET to see if men and women showed equal brain efficiency in the same brain areas while they solved mathematical reasoning problems. Mathematical reasoning is a more specific mental ability than the g-factor, so this would expand the bounds of the efficiency hypothesis. I worked on this project with Professor Camilla Benbow, another former Hopkins graduate student who had worked with Professor Stanley.
We recruited 44 male and female college students from my university (UCI) based on their SAT-Math scores at admission (Haier & Benbow, 1995). We selected four groups: men with high SAT-Math scores over 700; women with equally high scores over 700; men with average SAT-Math scores in the 410–540 range; and women with average scores in the same 500 range. There were 11 students in each group (44 participants was all we could afford, but this still was one of the largest PET studies at the time). Each person completed a PET scan while they solved actual SAT-Math reasoning problems. We expected to see lower brain activity in both the high SAT-Math men and the high SAT-Math women compared to the average groups, consistent with brain efficiency. We also thought that the men and women matched for high math reasoning might show efficiency in different brain areas because there are sex differences in brain size and structure, although at the time the evidence for these differences was not as compelling as it is today (Halpern et al., 2007; Luders et al., 2004).
Here’s what we found. In the 22 men, statistical analysis showed that high math ability went with greater activity in the temporal lobes (the lower side parts of the brain that include important memory areas like the hippocampus) during the problem-solving. This was just the opposite of efficiency. In the 22 women, we found no systematic statistical relationship between mathematical reasoning ability and brain activity. How the brains in the high SAT-Math women were working to solve the problems could not be determined, even though they were solving the same problems as the men equally well. And the men showed the opposite of what we expected. And that is how research often goes.
Actually, this finding was one of the first clear indications from imaging data that men and women may process information and problem-solve with different brain networks. Remember, in this study the men and women were equally matched on SAT-Math score, and they solved the same problems during the scan equally well. Their brains, however, showed apparently different patterns of activity. To us, this meant that not all brains work the same. This may seem obvious and even trite to you, but most cognitive researchers are interested in discovering how brains work in general, assuming that all brains basically work the same way. A focus on individual differences and the idea that not all brains work the same way was not so popular then or now. Also, remember, mathematical reasoning ability is a more specific factor; it’s not g. Brain efficiency may be related to g, but for specific abilities like mathematical reasoning, better performance may require more brain activity. Along these lines, another PET study about the same time in eight middle-aged individuals reported increased activation during the performance of a perceptual maze task, a measure of visuospatial reasoning, which also is a more specific factor of intelligence than g (Ghatan et al., 1995).
Confused? My purpose in describing these studies in the chronological order they occurred is to give you a feel for how researchers go about their work and sort through apparently discrepant findings. Remember my three laws. Repeat after me: No story about the brain is simple; no one study is definitive; and it takes many years to sort out conflicting and inconsistent findings and establish a weight of evidence. The next chapter will bring some clarity to imaging results and, unsurprisingly, raise new questions.
But before we continue to other early imaging studies of intelligence, I want to mention one more PET study we did. By the year 2000, it was still the case that very few other intelligence researchers were using PET or other imaging. We were still interested in brain efficiency, but we also started to wonder about whether efficiency would be related to intelligence even when the brain was not solving problems. In other words, could a smart brain be distinguished even when it was not working to be smart?
Our next PET study looked at eight new college students while they passively watched videos with no problem-solving required (Haier et al., 2003). This was a project on emotional memory so some videos were more emotionally loaded than others, but as a separate analysis, we looked at whether intelligence, assessed by the g-loaded RAPM test of abstract reasoning, was related to watching the videos irrespective of their emotional content. We correlated brain activity during this non-problem-solving condition to RAPM scores. Significant correlations were apparent in several areas. None were in the frontal lobes. Most were in the posterior areas of the brain where basic information is perceived before it is processed by association areas more toward the front of the brain. This suggested that people with higher RAPM scores seemed to be viewing videos with different brain activity than lower RAPM people. We think this means that smarter people are more engaged and actively processing the video information differently. In other words, the s
marter brains were not so passive. This is more evidence that not all brains work the same way, perhaps even while watching television.
Several other PET studies related to intelligence were done in this early period. Collectively they reported activations in areas throughout the brain while performing different tests of deductive/inductive reasoning (Esposito et al., 1999; Goel et al., 1997, 1998; Gur et al., 1994; Wharton et al., 2000). The individual differences approach, that is, looking for correlations between test scores and degree of activation, was not systematically reported, but all these studies found that multiple areas across the entire brain were activated during reasoning. The evidence was mounting that intelligence was not just a function of frontal lobes.
3.4 What the Early PET Studies Revealed and What They Did Not
The PET studies we’ve covered in this chapter so far represent the first attempts to use high-tech functional brain imaging to investigate intelligence. The overall point is that even these earliest studies helped shift intelligence research away from predominately psychometric approaches, and the controversies about them, to a more neuroscientific perspective because imaging provided a way to determine how psychometric test scores were related to measurable brain characteristics like glucose metabolism.
Here’s a summary of four key observations that emerged from these early functional imaging studies:
1. Intelligence test scores are related to brain glucose metabolism. This helps validate that the test scores were not meaningless numbers representing a statistical artifact. In fact, as neuroimaging studies of intelligence continue to increase, old criticisms about intelligence test scores having no meaning are less and less meaningful, if they were ever meaningful at all.
2. Early on, we had the unexpected and counter-intuitive finding that higher intelligence test scores were associated with less brain activity. The resulting efficiency hypothesis encouraged many subsequent studies and it is still viable, although as we will see in the next chapter, the story gets more complex as more studies are done, the same progression of progress found in all science.
3. Learning some tasks is associated with the brain becoming more efficient as indicated by lower brain activity after practice. This raises the question of whether intelligence can be enhanced by mental training. We will discuss this possibility and our deep skepticism of recent efforts in detail in Chapter 5.
4. PET scan differences between men and women solving problems, and PET differences between high- and average-intelligence watchers of videos, indicate that not all brains work the same way. We’ll be discussing this concept in Chapter 4.
There’s another important inference based on something we did not see. These early data did not show any one area in the brain that could be called the center of intelligence. In fact, the early PET imaging data reported that many areas distributed throughout the brain were associated with intelligence test scores. In 2000, however, one group of researchers claimed their PET study showed that the neural basis of the g-factor was derived from a specific lateral frontal lobe system and downplayed the importance of other regions (Duncan et al., 2000). They imaged blood flow over a 2-minute period as 13 subjects (with the wide age range of 21–34) performed a small number of problems that varied in g-loadings. Blood flow increased during the tasks, but only the frontal activations were noted as common to the high and middle g-loaded tasks. This publication, in Science, received considerable attention, but many researchers in the field were quick to point out several major flaws in design and interpretation (Colom et al., 2006a; Newman & Just, 2005). Design questions included the omission of any description of the subjects in terms of sex and IQ. Also, they were apparently recruited at a distinguished university, so a severe restriction of range of g-scores is likely, limiting correlations. Imaging occurred during a few problems while subjects worked at their own pace so the task reliability was low and averaging over subjects could minimize any differences due to differences in speed of responding to the problems. As for interpretation, none of the previous PET studies showing distributed areas related to high g-scores were cited, so there was no acknowledgment or discussion of inverse correlations or a distributed network of areas other than in the frontal lobe. And their own data for the high-g task showed activation in multiple areas outside the frontal lobes. Subsequently, Dr. John Duncan, the research leader, apparently abandoned the frontal lobe-centered model of intelligence and came to the view that other areas also were involved (Bishop et al., 2008; Duncan, 2010), consistent with virtually all other studies available at that time (Jung & Haier, 2007). So we will not dwell on this short-lived detour other than to note that its publication in Science brought important attention to imaging/intelligence research and validated again that g-scores could be studied scientifically, a proposition that had been surprisingly controversial. So even this flawed paper had some positive effect. At the time, journals like Science were reluctant to publish intelligence research, owing in large part to the controversies of the 1970s and 1980s concerning average group differences (see Chapter 2). The late Constance Holden, a science writer working at Science, lamented the prejudice against intelligence research she saw from the inside and did her utmost to cover intelligence research with a journalist’s combination of integrity and skepticism. Following her untimely accidental death, The International Society of Intelligence Research (ISIR) sponsors a presentation by a journalist as the Constance Holden Memorial Lecture at its annual meeting in recognition of her efforts.
3.5 The First MRI Studies
By 2000, a new imaging technology was becoming available much more rapidly than had PET. PET is based on positrons colliding with electrons and requires injection of radioactive tracers. Magnetic resonance imaging (MRI) does not require radioactive injections so no cyclotron or hot lab is required, making MRI scans considerably less expensive than PET (about $500–$800 versus $2,500). MRI is based on the effect of magnetic fields on spinning protons and hydrogen molecules. Because it produces high-resolution images of the entire body that have many important clinical uses without radiation exposure, MRI quickly became a must-have technology for most hospitals, especially ones associated with universities. This allowed many cognitive psychologists access and, in fact, within a dozen years or so, most psychology departments at major universities had their own MRI scanner, a multimillion dollar expense once unthinkable for a psychology department (although acquisition of imaging equipment by psychology departments was predicted by at least one prescient researcher; Haier, 1990). MRI studies of cognition have grown exponentially since the year 2000 and MRI analyses are now a mainstay of cognitive neuroscience research.
Here’s how MRI works. Protons naturally spin around an axis and the spinning creates a weak magnetic field. Each proton axis has a different, random north–south orientation. If protons enter a strong magnetic field, they snap into the same north–south alignment. When a radio wave is pulsed on and off rapidly into the magnetic field, the protons snap out of alignment and then back in. This pulsing can be done many times per second. As the protons snap in and out of magnetic alignment, the shifts give off weak energy, and this energy can be detected and mapped showing where the protons are if the magnetic field is applied along a gradient of different intensities. This sequence of events is called magnetic resonance imaging, or MRI (the original name of this technology was nuclear magnetic resonance, but was changed to avoid a “nuclear” connotation). Hydrogen protons are abundant in water and most of the body, especially soft tissue, is made of water, so MRI gives beautifully detailed images of the body and the brain.
MRI scanners are large, donut-like devices that contain a very powerful magnet. When a person lies on the scanner bed and the head or whole body goes into the center tube-like area surrounded by the magnet, radio waves are rapidly pulsed into the magnetic field, and the protons in the body snap in and out of alignment. The person has no sensation of this snapping. The shifting energy patterns formed by all this snapping are detected and m
athematically turned in to picture.
The illustration in Figure 3.4 shows an example of a basic MRI of brain structure in great detail. The figure shows side view slices (sagittal slices). Even whole-brain 3D images can be viewed. These, of course, are mathematical slices not actual slices. Like a picture printed in a newspaper, each brain image is made up of many individual dots called pixels. With MRI, the pixels actually have three dimensions so they are called voxels and they have volume rather than just area. There are millions of voxels in a brain image. Each voxel has a value determined by the imaging technique. In this case the value is the amount of energy detected by protons snapping and this can be interpreted as amount of gray matter, for example.
Figure 3.4 Structural MRI scan (sagittal view).