Let us now compare the remembering of experiences in given contexts with the rote memorization of random words: toast, jelly, dinner, breakfast, etc. In this case, the memorization mechanism is completely different because it is devoid of context and meaning. This contrast is even higher if we compare the memory of having breakfast with the memorization of nonsensical words like those used by Ebbinghaus: TOC, MIF, REP, etc. Clearly, there is an enormous difference between the memory mechanisms he studied and those we use in our daily lives. Despite this, Ebbinghaus’s two fundamental principles are universally accepted as true: we have two types of memory, short-term and long-term, and repetition aids memory consolidation. What Bartlett’s vision adds to our understanding of memory is the importance of extracting meaning, or, to use Bartlett’s own terms, the construction of a schema.
The experimental procedure used by Bartlett was simple and mostly descriptive; unlike Ebbinghaus, Bartlett concerned himself only with elucidating general principles and did not use quantitative data. Basically, Bartlett had Cambridge students read a Native American folk legend, “War of the Ghosts,” and then asked them to repeat the story to him.10 From these experiments, Bartlett concluded that the recollections of the story tended to be short and simplified, and that each student modified it based on his or her personal interpretation. When he asked the students to repeat the story at different time intervals (weeks, months, and even years after having read it), Bartlett observed that the subjects tended to change the story each time they repeated it, and that, in some cases, after many repetitions, the recollection had very little relation to the original story. More than the story itself, the subjects remembered the schema of the story that they had constructed based on the interpretation and associations they made at the time they read it. Using that schema as a starting point, they reconstructed the story in a different way each time, forgetting many details and unconsciously inventing and adding others. Based on these results, Bartlett concluded that memory is a creative process, and that the consolidation of a memory, far from being the etching on wax envisioned by Plato, reinforces a schema—a subjective representation that often changes the memory itself.
Just as the process of seeing differs greatly from the pixelated representation produced by a camera, memory differs greatly from a reproduction of our recollections as in a movie. That is precisely the reason why we dwelt at length on the description of the elementary principles of vision—because the same principles apply to memory. There is in fact a deep resemblance between Helmholtz’s sign construction, which we sketched in the previous chapter, and Bartlett’s construction of a schema. In one case we referred to vision and in the other to memory, but the processes that occur in our brains are essentially the same: they imply constructing a meaningful reality starting from unconscious inferences and then using this meaning, this sign or schema, instead of reality itself; it implies making abstractions based on selecting information and discarding innumerable details. In the previous chapter we saw how unconscious inferences give rise to visual illusions; similar inferences give rise to fabulation—the consolidation in our memory of incidents that do not correspond to actual experience.
An astounding example of the malleability of our recollections was given by psychologist Elizabeth Loftus, who carried out a simple but conclusive experiment.11 Loftus showed different subjects a video of a traffic accident and then asked them to estimate the speed of the cars that were involved in it. But now comes the interesting part: she asked one group of subjects to estimate the speed of the cars when they hit each other; she asked another group for the speed when they collided; a third group had to estimate the cars’ speeds when they smashed into each other; with the fourth group she used the word contacted; and with the fifth, she used bumped. The surprising result was that all subjects saw the same video in the same conditions, yet those who were asked using the word smashed gave the highest speed estimates, followed by those who heard collided, bumped, hit, and, finally, contacted. Even more surprising was that, a week later, Loftus asked the same subjects if they had seen broken glass at the scene of the accident. Thirty-two percent of the subjects who heard the word smashed answered (incorrectly) that they had, while only 14 percent of those who heard the word hit did.
Loftus’s results show how fragile our memories are and how they are prone to manipulation during the consolidation process—all it takes is changing a single word in a single question. Beyond their scientific interest, these discoveries are of enormous practical importance because they highlight the subjectivity of eyewitnesses at trials, and how readily their testimony can be manipulated by the way in which questions are asked.12 It is estimated that, in the United States alone, more than 200 innocent people have been sentenced to prison after being incorrectly identified by eyewitnesses. Particularly notorious is the case of Ronald Cotton, which deserves to be recounted in detail for the telling evidence it provides of the fragility of memory.
In 1984, Jennifer Thompson, a college student in North Carolina, was raped by a person who broke into her home. With a knife to her throat, unable to escape, Thompson decided to focus on the rapist’s face and remember every one of his traits and features so that someday, if she survived the attack, she would be able to identify him and secure his conviction. Thompson helped a police sketch artist construct an initial likeness of the rapist. Police assembled a group of six suspects and showed their photos to Thompson to see if she could identify him. According to the detective in charge of the case,13 Thompson examined the photographs for about five minutes before identifying Ronald Cotton. Two days later, the detectives placed Cotton in a lineup and, after hesitating between two suspects, Jennifer Thompson identified him again. At that moment she was convinced that she had identified her rapist—even more so when she was told that it was the same person whose photograph she had previously singled out. There was no doubt in her mind. And yet she was wrong.
At the age of twenty-two, Ronald Cotton was sentenced to life in prison. Some time later, by chance, the prison admitted a serial rapist who hailed from the same town as Cotton and bore a passing resemblance to him—Bobby Poole. Cotton heard through the grapevine that Poole was the man who had actually raped Jennifer Thompson. Cotton managed to have the case reopened, but as she faced the two suspects, Jennifer Thompson once again, and with complete certainty, identified Cotton as the perpetrator and asserted that she had never seen Poole before. Finally, after spending nearly eleven years in prison, Ronald Cotton was exonerated when a then-new tool, DNA testing, proved his innocence (and Poole’s guilt) once and for all.
It is noteworthy that, despite having made a concerted effort to remember the face of her rapist during the attack, Jennifer Thompson wound up reshaping her recollection and remembering someone else, and was unable to identify the true culprit when he was in front of her. Even after being told of the incontrovertible evidence provided by the DNA test, it was Cotton’s face she saw when she thought back on the attack. So, how could she have been so convinced of something that just wasn’t true?
It is easy to argue in hindsight, but it is telling that Thompson hesitated for five minutes before choosing one of the six suspects from the photo array; had she been truly certain, her decision would have taken no more than a few seconds. She then dithered between two suspects at the police lineup. After that, she unconsciously consolidated an incorrect memory that went on to become her unquestionable truth.14 Jennifer Thompson did not act in bad faith or take lightly the matter of sending someone to prison for life; she simply acted in accordance with what she (erroneously) remembered.15
Summing up, we’ve seen that our memories are shaped and stored based on our interpretations of them. We described two experimental approaches to the study of memory: on the one hand, Ebbinghaus’s systematic and quantitative study of the number of nonsense words that he could remember at different time intervals, and, on the other, the efforts of Spiller and Bartlett, which, despite being more descriptive than exact, show cl
early just how little we remember. Is there a way to blend these two approaches? Can we make a more reliable estimate of our memory capacity without resorting to experiments with nonsensical words or our vague and subjective remembrance of experiences?
In the 1980s, Thomas Landauer, an American psychologist, set out, in the spirit of Bartlett and Spiller, to estimate the amount of information that we remember, but using a more quantitative experimental approach.16 To that end, he studied the number of words people remembered a few minutes after reading a text sample; the time interval was long enough to render short-term memory negligible and allow the examination of long-term memory alone. Assuming an average reading rate of 180 words per minute, Landauer estimated that his subjects stored in memory around 1.2 bits of information each second. This result is not restricted to textual memory, since Landauer obtained a similar number, between one and two bits per second, when he estimated the number of visual images the subjects could remember a short while after having seen them. These estimates led to several interesting conclusions. Assuming that people are awake sixteen hours a day, and taking into account the fading of memory with passing time (using data similar to Ebbinghaus’s forgetting curves), Landauer estimated that a seventy-year-old person stores about 109 bits of memory. In other words, in the course of a lifetime, we can accumulate no more than 125 MB of information. This estimate was based on remembering text and images, but Landauer argued that the information required by other types of memory (spoken dialogue, musical passages, etc.) is of the same order. The exact amount may be larger or smaller, but the undeniable conclusion remains that we remember very little of our lifetime experiences.17 Following Landauer’s calculations, a 128 GB flash drive, whose chip is smaller than a thumbnail, can store one thousand times the memory that a human brain accrues in a lifetime. Does this mean a thirty-dollar flash drive is more powerful than a human brain?18 Obviously not. But as we study these estimates further, they begin to illustrate what distinguishes our memory from that of flash drives and computers.
In Chapter 2, we saw that with eight bits (that is, one byte) of information it is possible to represent the 256 ASCII characters used in text. Thus, if we assume an average word length of five letters19 and an average reading rate of 180 words per minute (that is, three words per second), we come up with an information flow of 120 bits per second (bps). But this presupposes that we process text letter by letter. If instead we consider a more elaborate storage representation, for example one based on words as the smallest unit processed, then the quantity of information drops to approximately 45 bps.20 The intriguing fact is that we do not store 120 bps, or 45bps, but just 1 bps—because the representation we generate is much more sophisticated than that conveyed by letters or words. It is even more interesting to consider the amount of information we store from images. In previous chapters, we saw that the retina transmits approximately 10 Mbps of information to the brain through the optic nerve. Thus the visual information that we end up remembering (according to Landauer, on the order of 1 bps) is less than a millionth of that transmitted by the eye to the brain, which, as we saw earlier, is in turn much less than the information present in our field of view. In other words, the memory we keep of all the images we see in a lifetime amounts to approximately as much information as that sent by the eye to the brain in just two minutes.
On the other hand, we saw in Chapter 1 that there are 100 billion (1011) neurons in the human brain. Considering that each neuron can encode one bit of information (by being either silent or active), then the brain would be able to store dozens of gigabytes. Some scientists go further and estimate that the brain can store one bit of information in each synapse, of which there are some 1015; this corresponds to around 1,000 TB or a petabyte of information.21 Again, whichever estimate we prefer, it is clear that the storage capacity of the brain greatly exceeds the amount of information that it actually stores (125 MB, according to Landauer’s results). This is because the brain’s machinery stores information in a very redundant way—with sets of neurons encoding, in parallel, specific aspects of the same piece of information—in order to derive meaning. This is precisely what distinguishes our human brains from flash drives or computers. A computer’s hard drive can store and faithfully reproduce scores of text passages, photographs, or videos, but it cannot understand them. The human brain, on the other hand, concentrates its resources on attributing meaning to the paltry amount of information that reaches it from the senses.
As Helmholtz and Bartlett argued, meaning is constructed by way of assumptions based on previous experience. A few years ago, an extraordinary magician and great friend, Miguel Ángel Gea, gave a talk on magic at a packed lecture hall at my university. He started by remarking that the people in the audience, mostly academics and university students, were cultured and intelligent, and thus . . . very easy to fool! Gea continued by explaining that “intelligent people” are constantly making assumptions about reality, and that the magician’s art consists of flouting these very assumptions as they go about their tricks. In fact, it is not a coincidence that magic tricks for children are completely different from those for adults: children notice details that grown-ups have learned to ignore completely with time.22
These assumptions, what Helmholtz called unconscious inferences, are part of our everyday lives, whether we are watching a movie, listening to music, crossing the street, reading, or playing sports. For example, an important aspect of music theory involves the use of tension and resolution. Tension is generated by building expectations (leading us, for example, to expect a tonal chord after a fifth) that are resolved at a time of the composer’s choosing.23 We may admire the genius with which a composer breaks with classical musical structure by inserting dissonance, changing the key or rhythm, etc., yet too much of this disruption renders us unable to predict what might come next, and as a result we generally find it unpleasant listening. Even musical styles that we think of as being chaotic and disordered follow recognizable musical conventions.
We make inferences every time we watch a movie as well. Horror and suspense movies, in particular, manipulate our expectations and generate tension by leading us to predict that something is about to happen based on the music, the setting, or the length of a scene. Of course they also play with the surprise of the unexpected, but most of the tension they generate results from our prediction that something dramatic is about to occur—even, or especially, if we aren’t sure of the exact moment it will. Alfred Hitchcock, the master of cinematic suspense, once said that an explosion does not cause fright, but its expectation does; according to Hitchcock, showing a bomb before it explodes generates much more tension (and is thus far more spine-chilling) than having one explode unexpectedly.
Examples of the use of expectation and inference abound in sports: A goalkeeper sees the stance of the kicker and predicts where he will shoot the penalty kick; a tennis player sees how his rival moves as he hits the ball and predicts where the ball will go. Successful athletes avoid giving away clues that may allow rivals to make such predictions—or actively encourage false predictions by manipulating their opponents’ expectations.
Similar principles apply to everyday situations. If I get a response to a job application that begins, “Unfortunately …,” I do not have to keep reading to know whether I’ve been hired. More fundamentally, the fact that, as noted previously, we process written information in a more sophisticated form than what is conveyed by letters or words underlies how we read; this is why an adult can read so much faster than a child: children read syllable by syllable, whereas adults tend to skip words by making use of unconscious assumptions. Similarly, I can infer the tone and to an extent even the content of what someone is saying just from his expression—a tool we often use when we are having a conversation in a loud room or in a language we do not fully understand. If I am at home and hear a voice that I do not immediately recognize, I do not compare it to the voices of everyone I have met; I automatically sort through a much-narrowe
d set of possibilities. I assume it is someone from my family because the probability that it is someone else is very low. Likewise, if I hear a train, I know the sound is coming from the radio or the television, because even if I have a state-of-the-art audio system and the sound is identical to that of a train at the station, I know that there are no railroad tracks close to my house.
In short, the brain makes decisions about how to interpret the information the senses provide by making inferences based on previous experiences.24 Just as we think we see everything in detail, but in truth see only a fraction of the scene in our field of view and infer the rest, we’ve learned that we remember astoundingly little. We think we remember past experiences in detail, but in reality we remember only a few concrete facts and fill the gaps between them with assumptions. I think I remember what I did yesterday: I went by bike to the office, made myself tea as I read emails on my computer, then discussed some results with one of my students, had lunch, and so on. However, from all of these events, I may really remember only some of the conversation I had with my student—and then only if something about it was novel or notable. Everything else is part of my daily routine, to which I pay no attention and which I do not encode in my memory but rather presume based on experience. This is exactly the process that led Bartlett to find that his students remembered a shorter and more coherent version of the story they were given to memorize. They could not remember everything; they remembered a limited number of concrete facts and inferred the rest. The construction of a schema based on such inferences—remembering our subjective interpretation of reality rather than reality itself—is precisely the source of false memories that lead us to be certain of events that never happened.
The Forgetting Machine Page 5