The Ravenous Brain: How the New Science of Consciousness Explains Our Insatiable Search for Meaning
Page 24
The prefrontal cortex is in some ways the most abstract general higher-order region that humans possess. Penfield’s sister had most of her prefrontal cortex removed, but only in one hemisphere, with the left prefrontal cortex intact and able to take on functions that both prefrontal cortices had previously carried out. But if you lose both prefrontal cortices, what happens? Such cases are rare, possibly because the patient is far too ill to take part in any scientific research. But one neurologist, Bob Knight, has come across such a patient. The patient was awake, but otherwise looked disturbingly like a zombie. He had no motivation of his own, just constantly sat immobile in a chair as he stared into space, and tragically, it seemed as if any meaningful awareness was totally absent.
It seems clear from all these patients with varying degrees of damage to the prefrontal parietal network that such regions carry out a combination of attention and working memory processes, and that when such functions are severely impaired, consciousness consequently shrinks to a very marked degree.
BRAIN-SCANNING THE PREFRONTAL PARIETAL NETWORK
But what is the consensus from brain-scanning studies exploring the role of the prefrontal parietal network? The main thrust of the research has indeed centered on working memory and attention, with the prefrontal parietal network closely associated with both.
For instance, the prefrontal parietal network will increase in activity if you increase the number of letters you place in working memory, the number of abstract relations between items of an IQ task you hold in working memory, and the number of spatial locations you have to remember. Likewise for attention. The prefrontal parietal network increases in activity, for instance, if you switch attention between tasks, or if you attend to visual changes on a screen.
In fact, as this is our most abstract, higher-order brain network, a common finding is that these areas are activated whenever we perform a complex or novel form of any task whatsoever, whether it involves short-term memory, long-term memory, mental arithmetic, or any other potentially challenging cognitive process. For this reason, these regions are also most closely linked with IQ.
This pattern of results implies two vital features of our thoughts and consciousness: first, the intermingling of cognitive processes, and second, how this amalgam of advanced mental functions is inextricably linked with awareness.
If you read a psychology textbook, even now, it will probably make neat distinctions between working memory, attention, long-term memory, mental arithmetic, reasoning, and so on, as though these were all independent processes. Increasingly, though, researchers are eroding the spaces between almost all types of thought and memory, and building broad links between them.
Aside from the fact that all such thoughts will activate the prefrontal parietal network, there are many other connections. For instance, it’s clear that working memory is intimately linked with attention, with working memory effectively being the output of attentional filtering. Manipulations in working memory, especially of some current goal, can influence which attentional filters are set. And if you fill up working memory with items, your attentional ability suffers. To demonstrate this, Nikki Pratt and colleagues recently gave volunteers a classic attentional task that involved responding to the direction of a central arrow while ignoring a distracting group of surrounding arrows pointing the wrong way—a surprisingly difficult test requiring firm attentional control. On some of the trials, volunteers also had to keep a set of letters in working memory prior to stating which direction the central arrow was pointing. This extra working memory load caused the volunteers to guess the arrow’s direction more slowly and less accurately—in other words, it reduced their attentional resources for the arrows. Pratt also found, using EEG, that brain-wave markers of attention for the arrows were weakened when working memory items were stored, further reinforcing the links between the two processes. Results such as these have led the most prominent current theories of attention to subsume working memory within an attentional framework.
Thus, this traditional picture that there are a group of largely independent brain areas and corresponding types of thought—say, one for attention, another for working memory, another for long-term memory encoding, and even another for consciousness—is one that increasingly clashes with the evidence. Instead, the most parsimonious way to describe the data is in terms of another distinction, that between static, automatic, unconscious processes, on the one hand, and highly dynamic, flexible, conscious ones, on the other. The automatic processes are those we’ve stored in our specialized memory and motor areas as overlearned habits and goals, and are usually the product of our roaming consciousness. And consciousness is an inextricably interlocked collection of processes carried out in the main by the prefrontal parietal network, with attention and working memory being the most prominent two functions. This consciousness machine is designed to kick in whenever the task cannot be achieved by our instincts or bank of unconscious automatic habits. It analyzes and manipulates the contents of our working memory in logical ways, drawing in further information from our specialist systems if necessary, and brings to bear this substantial portion of general-purpose cortex to solve complex or novel tasks, ideally by creating new automatic habits that the next time won’t require consciousness.
THE PREFRONTAL PARIETAL NETWORK, CONSCIOUSNESS, AND CHUNKING
But what of chunking and spotting patterns? While it is indeed a near universal finding that the prefrontal parietal network will be increasingly activated as the demands of a task increase, there is one clear and important set of results with the exact opposite picture.
My colleagues and I at Cambridge carried out one of the first experiments explicitly to demonstrate this exception. Volunteers in the fMRI scanner would view an array of 16 red squares arranged in a 4 by 4 matrix. Four of these red squares would blink blue in a sequence, and a few seconds following this, the participants would try to correctly recall this sequence of locations. This is so far a standard spatial working memory test. The twist is in the fact that we actually gave volunteers two different types of sequences—ones that looked entirely random, just like the conventional spatial working memory tests, and another that took advantage of the 4 by 4 structured array to make the sequence form squares, triangles, or other symmetrical and regular paths (see Figure 8). This made the sequences easily chunkable. Subjects consciously detected these chunks and talked after the experiment about how these sequences were easier because of the patterns.
If difficulty in every single situation drives the prefrontal parietal network, then the unstructured, more difficult sequences should generate more activation in these regions. But if instead there is something special about chunking processes, then perhaps these easier structured sequences will drive the prefrontal parietal network more than the harder, unstructured alternatives.
In fact, very robustly, the easier patterned path trials did light up the prefrontal parietal network more brightly compared with the more difficult, unstructured sequences. So in some cases, at least, task demands don’t drive this network—at least when chunking is involved.
Because this was a rather unexpected result, we repeated the same experiment, except this time with digits: Subjects heard 8 single digits in the fMRI scanner, and then after a few seconds had to say back the sequence in the order that they’d heard it. Some sequences were distinctly structured, such as 8 6 4 2 9 7 5 3 (descending even numbers, then descending odd numbers), while other sequences were deliberately made to be as random as possible. Just as with the spatial structured and unstructured trials, these structured digit sequences were easier because subjects could chunk them—and they, too, activated the prefrontal parietal network more.
One outstanding question we had with this work was whether it was simply memory for preexisting chunks, such as squares or single-digit, even-number sequences, to which we are all exposed from early childhood—that was driving the participants’ responses, or if participants were instead noticing the chunks on the fly, s
potting the pattern as a powerful new rule to apply on each trial. So we devised a new experiment, sticking with verbal working memory and digits, but this time moving on to sequences of double digits. This way we could give volunteers sequences they were very unlikely to have seen before, to make sure we were dealing with mathematical and novel pattern spotting. For instance, subjects might have to remember the sequence 57 68 79 90 over a few seconds (each number going up by 11 each time). We also had a standard, totally unstructured sequence that subjects couldn’t apply any chunking strategies to (say, 31 24 89 65). And there was also a memory-based chunking condition, to see if that would also activate the prefrontal parietal network, and if so, how strongly. To enable a memory-based chunking condition, I trained my volunteers for at least 4 hours each to memorize 20 different unstructured 4-digit chunks before I scanned them. I told them they had to imagine they were playing a game where they were a new receptionist at a medium-sized company, and as part of their induction they had to memorize the faces, names, and phone extension numbers of 20 key members of staff. So, after a series of graded training exercises, they each had a relatively naturalistic set of 4-digit numbers cemented in their memories. When it came to the scanning part of the study, they might have seen the novel sequence 21 05 81 63 to recall in working memory, where 2105 would be one phone extension number they’d comprehensively learned during the previous week, and 8163 would be another. So we had three different fMRI digit-based working memory conditions—one involving mathematical structure, a second involving mnemonic structure (made up of the phone extension numbers they had learned so thoroughly), and a third with no structure (so subjects just had to rely on their working memory and not chunking). With these three, we could distinguish the brain regions activated for mathematical chunking against those for memory-based chunking. In fact, we also had two controls—one to match the memory content and a second to match the mental arithmetic content, but in each case without the additional digit sequences task, and thus the opportunity to benefit from chunking.
As usual, the sequences that could be chunked, because of strategies involving either memory or mathematics, were easier for the subjects to recall compared with the unstructured sequences. And both kinds of chunking sequences, as before, lit up the prefrontal parietal network considerably more brightly than did the unstructured sequences, where chunking wasn’t an option. Additionally, the condition involving mathematical sequences still robustly activated the prefrontal parietal network compared to its matched control condition (where subjects carried out equivalent mental arithmetic tasks, but without any chunking component). A similar pattern of prefrontal parietal activity was seen when the condition involving the memory-based chunking sequences was compared to its own matched control (with the same level of memory recall, but no chunking aspect to the task). This demonstrated that these regions were not just being driven by mental arithmetic or memory recall when subjects carried out the chunkable sequences, but something more—quite specific to the act of chunking. I was most struck, though, that the condition involving mathematical chunks still robustly activated the prefrontal parietal network when compared to the memory-based chunking condition. This placed the mathematical chunking task as the firm winner in the game of driving activity in the prefrontal parietal network. Ask a scientist in the field to generate a list of those processes that will most reliably activate the prefrontal parietal network, and she will most likely include working memory, long-term memory, and mental arithmetic. But in this experiment, the mathematical chunking task activated the prefrontal parietal network more strongly than all of these, as well as compared to the memory-based chunking condition. In other words, this experiment demonstrated that the prefrontal parietal network will activate for many complex tasks, but it will be most excited when subjects are actively searching for and finding entirely new patterns.
Other research groups have also linked chunking with the prefrontal parietal network. In the long-term memory domain, Cary Savage and colleagues showed that chunking will both light up the prefrontal cortex and boost performance if a group category strategy is applied to word-list memorization (for instance, remembering all plants together in one group and all metals in another). Similar results have been found in the working memory field. For instance, Vivek Prabhakaran and his coworkers presented letters to participants with the instructions that these stimuli had to be recalled over a short delay. If volunteers chunked the letters using a strategy that involved binding a spatial location to each letter, then prefrontal activity increased. In another working memory task, Christopher Moore, Michael Cohen, and Charan Ranganath demonstrated that extensive training involving the categorization of abstract shapes enabled memory-based chunking for such stimuli, which improved subjects’ performance and increased activity in the prefrontal parietal network.
One recent intriguing, related study by Stanislas Dehaene and his team actually showed the transition from consciously spotting the pattern to then using it in a routine, automatic manner. Subjects had to discover a novel sequence from the letters ABCD. For instance, a participant might first try “A” and be told this was wrong; then she might try “B” and be informed that this was correct—so now she would know that the first letter in the sequence was “B.” She might try “C” for the second letter and be given feedback that this was wrong. At this point, she would have to start the sequence all over again, but at least she would now know to start with “B” and explore the second letter. Eventually, by trial and error, she would work out the sequence, which she would have to repeat between three and six times—now a very easy task for her. Then there would be a new sequence for her to work out, and the cycle would continue. During the initial search phase for each trial, there is massive prefrontal parietal network activity, but this quickly dies down as soon as the task becomes routine. The prefrontal parietal network then becomes virtually silent as the task is automated in the volunteers’ minds, and they need very little consciousness to complete the well-learned sequence. In other words, this is a beautiful illustration of the distinction between the conscious search for patterns, carried out in the prefrontal parietal network, and our largely unconscious automatic habits, which require specialist brain areas alone.
In all the examples above, the chunks tended to be so obvious that volunteers couldn’t help but be aware of them. What would happen, though, if the participant couldn’t consciously spot the chunks for some reason? Would the prefrontal parietal network fail to activate for these undetected structured sequences? Serendipitously, I was able to answer this question via a fascinating person who was effectively blind to all such number chunks. Daniel Tammet is a prodigy with a few similarities to the incredible Russian mnemonist Sherashevski, discussed in Chapter 2. For one thing, Tammet has a rather extreme form of synesthesia, just as Sherashevski did. Those with synesthesia commonly associate colors with specific single digits they read. Tammet, however, has a different experience for not just the first ten numbers, but the first ten thousand. Not only this, but he also adds not just color, but texture, shape, height, and even touch to his perception of numbers. This creates an incredibly vivid, structured experience for him whenever he reads a stream of digits. He is also diagnosed with a high-functioning form of autism known as Asperger’s syndrome.
When I tested him, Tammet was able to cram far more numbers into his short-term memory than any other volunteer I’d ever tested (though this was somewhat reduced when I deliberately confused him by coloring the numbers in ways that clashed with his synesthetic inner eye). He had just set the European record for memorizing the most decimal places of pi: 22,514—which he claimed was quite easy to do, the most difficult part simply being the 5-hour stint needed to recite all the numbers from memory. He also has prodigious mental arithmetic skills—for instance, being able to divide one double digit by another and give the answer to 100 decimal places. And he claims a deep facility for languages, with the ability to learn a new language in a single week. Alt
hough his Asperger’s syndrome probably allows him to concentrate more deeply than most, his exceptional abilities mainly arise from the intensely multisensory inner numerical world he experiences, with every number seeming so very vivid and distinct to him. Memorizing or mentally manipulating numbers comes easily and naturally, and to recall them, he simply has to convert the inner psychedelic mountain terrain back into digits.
We decided to investigate his brain activity when he carried out one of our chunking tests—where 8 single digits are presented to be retained in memory over a few seconds, with half the sequences structured, like 8 6 4 2 9 7 5 3, and half random and unstructured. When I asked him after the experiment whether any trials were easier than any others, he unsurprisingly said that they all were just as easy—because for him, unlike normal people, remembering only 8 digits really is very straightforward. But when I probed further, it turned out that, surprisingly, he’d completely failed to realize that some of the trials involved highly structured sequences of digits. He was in fact completely blind to the external structure. His brain activity reflected this: Totally unlike normal volunteers, he showed absolutely no increase in activity for the structured sequences compared with the unstructured sequences. He was neither aware of the structures nor in any way exploiting these patterns to chunk the sequences and reduce his working memory load. So Tammet showed, by failing to notice these chunks, and failing additionally to activate his prefrontal parietal network for these obvious forms of external structure, that you really do need to be aware of and use the chunk in order for the prefrontal parietal network to kick into action.