by Bor, Daniel
So, being heavily unconscious, such as under general anesthesia, is associated with a more orderly, slow, thumping march of neuronal activity, which is capable of carrying only minimal information around the brain. These neurons may really struggle to broadcast anything further afield than their immediate neuronal friends; they aren’t receiving anything very original to pass on anyway, as so many neurons are singing the same tune.
And here is an important initial clue as to why we can be conscious, while bacteria and plants cannot: Consciousness occurs when there is an active transfer and intermingling of information across much of the neural landscape. In contrast, any information processing in bacteria and plants is distinctly local to some small pocket of the entire protein-DNA machinery.
LEARNING ON THE OPERATING TABLE
My linking of consciousness with complex information processing is in accord with this unconscious general-anesthesia scenario, in which mechanisms that would usually support such processing are no longer available: Brain waves have become slow and lumbering, and the latest, smartest parts of the brain have been turned off. So, does it follow that all sophisticated learning is unavailable when anesthesia strips us of consciousness?
Problematically, responses to anesthesia can differ markedly from one patient to another. On top of this, thankfully very rarely, some patients have remained awake during their operations and have been traumatized by their very conscious, very intense pain. So before testing for any learning, you first need to ensure that the anesthesia is sufficiently deep that the patient is truly unconscious.
The main method to establish that the patient is fully unconscious is by using EEG to ensure that the brain waves are sufficiently slow and deep, with little or no neural communication across regions. What can be learned after such checks have been confirmed? From the consensus of studies on learning in anesthesia, where testing only occurred after appropriately deep levels of anesthesia were carefully detected and monitored, there is no evidence at all of subjects, on waking, recalling anything from their operation. If word lists to be memorized are read out during the operation, patients have no memory of any of the words. If instructions were repeatedly given in the midst of the procedure, say, to lift a finger when the patient hears the name “Rumpelstiltskin,” then nothing happens later in recovery when the experimenter gives the cue—patients do not lift a finger or recall having been given any instructions.
But, despite this, there are still some more subtle forms of learning that are possible. We actually learn all the time, in a weak sense of the word, without knowing that we’ve learned anything. For instance, as I mentioned in Chapter 1, if I say “artichoke artichoke artichoke artichoke artichoke” to you a hundred times in the next minute, you’ll spend the rest of the day predisposed to react more to this word than usual. You’ll read “artichoke” a little faster when you come across it, you’ll notice it quicker in the supermarket, you’ll think of artichoke a little earlier if I tell you to come up with as many vegetables as you can, and you might even be a little more inclined to buy an artichoke at the market. You won’t be aware of any of this, and you’ll have little control over it, but it happens nevertheless.
What occurs, from the neurons’ point of view, when you’ve been exposed to such a word? The neurons that collectively represent “artichoke” in your brain were reactivated, and in the process, their thresholds for firing again were tweaked. They therefore will be a little quicker to draw their neuronal firing gun the next time you hear about or see artichokes. In one sense, this is akin to a micro-muscle getting a little exercise and becoming stronger from the use. But, actually, far more is going on. This little collection of neurons firing together a little more readily in response to “artichoke” is in fact an important predictive computation. The neurons are effectively saying: “Aha—artichoke is around again, perhaps it’s now a bit more important and frequent, and we should reflect that fact by getting faster and louder to respond to it, or even anticipate it more keenly.” Every time we are exposed to virtually anything whatsoever, this neuronal fine-tuning occurs.
There’s no doubt that our unconscious minds are bubbling, spitting cauldrons of computations, based on this constant stream of neuronal tweaks that are dedicated to predicting the world around us. Our sensory perception is not a direct mirror to the world outside, but a series of computational steps designed to give our conscious minds the most pertinent information available. We move our bodies based partly on rich feedback from our senses as to where our limbs are and where the objects we’re reaching for are placed in relation to our limbs. All this occurs seamlessly, unconsciously. Many of these low-level lessons are hardwired. We are primed to predict that the sun is above us, that objects have edges, and so on. This hardwiring is the product of many millions of years of evolution and designed with ruthless efficiency, so that we humans, and our mentally simpler ancestors, can swiftly extract the dangers and delights of the environment—and react before the predator or a competitor does. Many other unconscious lessons may once have been born as conscious explorations of the world when we were infants, but now they are so embedded into our world picture, and so buried under a mountain of more meaningful ideas, that we never consciously acknowledge their existence.
In one sense, this unconscious machinery for predictive learning is indeed complex, but only because of the sheer number of simple statistical calculations occurring, not because of the grand truths that are being unpeeled before our eyes. These neuronal tweaks that occur under the surface can only ever be the servants of our understanding, because only when their largest lessons are combined in consciousness can we really learn the interesting, deep patterns of life.
Just how limited is our unconscious mind, when applying these simple neuronal calculations? If you tell a patient recovering from an operation to complete the partial word “ash,” she may be equally likely to say “ashcan” or “ashtray.” But if you repeated the word “ashtray” when she was deeply unconscious, she will in fact be more likely to say “ashtray” in recovery. She will have absolutely no knowledge that she has just heard the word “ashtray” an hour or so ago; nevertheless, some part of her brain will remember. Some family of neurons has been tweaked to reflect this recognition, and she will respond accordingly. In other words, she has been “primed” to repeat a word to which she has already been exposed. Examples like this show that at some very superficial, unconscious level at least, words can be noticed.
But there are multiple features of a word—for instance, there’s the sound, the grammatical features, the linguistic relationship a word has to others in various ways, and there’s also the meaning of the word. Meaning is the first main level at which our minds structure information, since meaning requires relationships between items in a dense web, hierarchies of categories, and so on. In this way, meaning is an especially pertinent test for the argument that consciousness is equated not only with information processing, but especially with structured information.
Can patients be primed for the meaning of the word as well as the sound? For instance, if asked after the operation to complete the same partial word “ash,” are patients more likely to say “ashtray” instead of “ashcan” if they were exposed during the operation to the word “cigarette”? If so, this would mean there had been a deeper triggering of activity, so that the neuronal population had spread from the sound of the word to activating any neurons coding for a related meaning. This is precisely what some researchers tested, and it turns out that, when under a sufficiently deep anesthesia, this form of learning is beyond us.
So, when profoundly unconscious, as happens under deep levels of general anesthesia, we can faintly learn under the radar of our consciousness that a certain word has just been presented to us, but that’s about the limit of it. Anything remotely more complex, such as a word’s meaning, is beyond our unconscious selves and requires at least some level of awareness. And, of course, for anything that we are actually conscious of learnin
g—creating a strategy, memorizing a list, learning from instructions, or any of the myriad forms of information we manipulate every day—consciousness is certainly required.
UNCONSCIOUS BETTER THAN CONSCIOUS?
General anesthesia is the gold-standard method for studying just what our unconscious minds are capable of, because it puts us in a situation where we are fully unconscious, even though our brains are otherwise quite healthy and capable. Therefore, the limitations on unconscious processing described above should be taken as definitive. But science is always improved when it adopts multiple approaches to examine the same question. Indeed, various other techniques have been used to explore unconscious learning.
For instance, what about when we’re already awake? Could it be that the deep ocean of our unconscious minds, when fed material from our conscious gaze, can churn over vast swaths of information and quickly allow conclusions to surface that are far more advanced than the insights that our deliberate, slow, conscious cogitations could ever produce? A Dutch researcher, Ap Dijksterhuis, carried out a series of experiments to argue just this point. He claimed that there were many situations where we should follow our gut instinct, where the slow, integrative processing of our unconscious minds was vastly superior to our clunky, far more limited conscious thoughts.
In one of Dijksterhuis’s experiments (there are now quite a few, all of a similar mold), volunteers were asked to rank four imaginary cars in order of preference, based on a set of attributes they were given, one at a time. One car had 75 percent of its features set as desirable (for instance, “The Dasuka has cupholders”); another two cars were neutral, with 50 percent of facts good and the other half bad; and one car was the worst of the bunch, with only 25 percent of the attributes set as positive. Once participants were shown all the features for all the cars in turn, they were split into two groups. One group, the “conscious” group, spent 4 minutes thinking about all the attributes they’d just read about and tried to work out the most accurate ranking of the cars. A second group, the “unconscious” group, spent 4 minutes being distracted by anagram puzzles instead. This, according to Dijksterhuis, allowed their unconscious minds sufficient time to process and integrate the facts unfettered by their consciousness, which was adequately engaged in an irrelevant task. In fact, so Dijksterhuis’s theory goes, the more the conscious mind is distracted by some other task, the better it is for the unconscious understanding of complex information. In the final, crucial stage, after the 4 minutes were up, both groups had to pick their favorite car.
According to Dijksterhuis’s results, if there were only 4 facts per car, and 16 in total, then both groups did pretty well at choosing the best car, with the conscious group marginally on top. His interpretation was that, with few items, our conscious minds cope fine, because their very limited memory resources aren’t too taxed by the moderate weight of facts to assimilate. But if you make the task vastly more difficult and have 48 attributes in total, 12 per car, then only about a quarter of the conscious group chooses the highest ranked car correctly—in other words, no better than a totally random guess. Meanwhile, a tremendously healthy 60 percent or so of the unconscious group makes the correct choice. This is a staggering difference. It led Dijksterhuis to conclude that in almost every sphere of decision making, be it political, managerial, or whatever else, “it should benefit the individual to think consciously about simple matters and to delegate thinking about more complex matters to the unconscious.”10
There is one small problem with Dijksterhuis’s conclusion, though: It is utterly wrong. If you stop and think about the task, as if you were a participant, it isn’t difficult to see why. So you’ve just sat down in the small, stark testing room in front of a computer monitor. The environment is rather alien and you know that this is a psychology experiment, which makes your pulse race a little faster. What are these weird scientists going to test? Are they really going to look at what they will claim to be looking at, or will you be tricked in some way? Are you going to appear stupid? Socially awkward? A man with only half a brain? You’re told to pick the most desirable car out of a bunch of facts, so you immediately try to do your best. First you’re told, “The Hatsdun is very new.” That’s good, isn’t it? Or does the car need to be broken in? Are some positives better than others? You haven’t been given any instructions about this, which is disconcerting and confusing. You haven’t seen any other attributes yet, so what are you meant to do? You decide the Hatsdun is provisionally in the lead anyway. Then you’re told, “The Kaiwa has a poor sound system.” Okay, that’s bad, the Kaiwa is now at the bottom of your list (even though you only listen to the stereo for the news, and don’t really care about the quality of the sound system—but you’ll ignore this thought for the moment). It all feels a little overwhelming, with 48 facts to try to remember, but you do believe after a short while—only about 10 facts, actually—that one car clearly has accumulated more positive facts than the others. You can’t see why you need to see all the facts to make your decision, but you keep paying attention and thinking about your choices, ready to modify them at any stage, and by the time all 48 facts have been presented, you’ve made up your mind. You could give your decision now, and start to open your mouth helpfully to volunteer the information—but that’s not what the experimenter seems to want. They get you to delay your decision by doing anagrams, maybe to distract you, but you make sure you keep remembering your favorite car throughout, giving yourself a little reminder every so often. It’s the same if you were asked to think for 4 minutes. It’s actually pretty pointless, as it doesn’t change your mind, as you’d already decided somewhere in the middle of seeing these facts. But you do what the nice researcher wants, so you get your money, or your course credits, and keep him happy.
This kind of experiment, as the above scenario illustrates, doesn’t have much to do with the unconscious mind at all. It is a distinctly conscious task, and in a very similar study from another lab, this time carried out by Laurent Waroquier and colleagues, 70 percent of subjects reported explicitly applying various strategies while they were being presented with the facts to make up their minds, and had already made up their minds before they had seen all the facts. In other words, they admitted that their minds were fixed by the time the facts were all presented, with the other 4 minutes of distraction or deliberation being totally superfluous. So as an experiment designed to tap the unconscious domain, the test disastrously missed its target.
Furthermore, it failed the critical scientific test of replicability. If you add together all the studies that use this design, on average they find absolutely no advantage for unconscious processing. In fact, even some of Dijksterhuis’s own experiments failed to find an unconscious-advantage effect. Other researchers, including a colleague of mine, Balazs Aczel, with whom I collaborated on an attempted replication, have found, if anything, that the opposite result was true: that even with long trains of facts, conscious deliberation provided an advantage over distraction. This is probably because being distracted, which fills up your conscious memory with other items, is more likely to dislodge your previously set correct memory for the order of the best cars. These dissenting papers, unfortunately, will never be splashed on the covers of any newspapers. But to me, the beauty and rigor of a sequence of carefully controlled experiments, doggedly seeking the truth, are far more exciting and fascinating than many sexy papers that spark an inferno of media interest by their novel, unreplicated results.11
So again, the message is clear: There’s absolutely nothing superior about the forms of learning possible when only our unconscious minds can perform the calculations.
FEELING YOUR WAY TO KNOWLEDGE
A more established and reliable body of evidence in this field involves performing some surface task, and unconsciously, accidentally, absorbing the underlying information within it. If I write the phrase, “Yesterday, the prime minister drive past me in the street,” we all would know that it was grammatically wrong, but many tim
es, with grammar, words just feel wrong, and we aren’t necessarily clear about which explicit rule has been violated. It is possible to experimentally reproduce this impression that we can know that some item is correct or incorrect, without knowing why. The main format for such experiments is for researchers to show subjects incomprehensible sets of long letter strings—for instance, “XMXRTVM.” The experimenter then tells these poor volunteers, rather sadistically, to memorize every one they see. After this part of the test, the participants are informed that the order of the letters in each string obeyed a complex set of rules, a kind of artificial grammar, but they will not be told what these rules are. For instance, one rule might be that the letter strings will always end with an M if there’s an R in the mix. The subjects will now be shown more letter strings and have to say whether they obey the same rules or not. Even though subjects think they are entirely guessing much of the time, it turns out that they still get significantly more answers correct than they would by chance. It’s nowhere near 100 percent, but then again, we had years to learn our native language’s grammar with many thousands of examples, and these people just have a few dozen instances over 20 minutes or so.
Just how complex can these rules be? One of the most impressive examples is that we can unconsciously detect the connections between musical notes (when told just to memorize them in the same way as in the experiment above) and transfer that learning to the same connections in letters, all while still believing we are guessing randomly—but again I should stress that this learning is statistically above chance, but not by much, hovering around 60 percent (where chance between two choices is 50 percent), so it’s miles away from infallible knowledge.