Milner’s result demonstrated the existence of a distinct form of memory for motor tasks, part of what today is known as procedural or implicit memory—the memory we use to ride a bike, tie our shoelaces, or drive a car. This type of memory does not depend on the hippocampus, which had been surgically removed from H.M.’s brain. In contrast, declarative or explicit memory, the memory for facts and events, things that can be named and purposely recalled, does depend on the hippocampus, and so was severely compromised in H.M.
Figure 7.4: Taxonomy of memory
Long-term memories are divided into declarative and nondeclarative (sometimes called procedural). Only the first kind, which in turn are divided into episodic and semantic, depend on the hippocampus.
To the duration-based classification of memories, we can thus add one based on their type. Declarative memory is in turn divided into semantic memory (of people, places, and concepts—the memory that allows me to remember the name of France’s capital) and episodic memory (of events and experiences—the memory that allows me to remember what I did on my last trip to Paris). These are intimately related, since, on the one hand, semantic memories form largely from repeated patterns in episodic memories (I see a university colleague at the pub, at various seminars, walking down the hall, and finally, though I may have forgotten most of the episodes in which I met him, I form the concept of my colleague), and, on the other hand, episodic memories tend to form by combining concepts—in other words, from semantic memories (for example, to remember having seen my colleague at the pub, I generate an association between those two concepts).
Nondeclarative memory is composed of many subtypes, among them motor-skill memories (which encode different abilities, like the movements necessary to ride a bike or serve a tennis ball) and what is known as emotional memory, which involves an area adjacent to the hippocampus called the amygdala and allows us to draw on past experiences to recall (mainly unconsciously) that we like or dislike a certain smell, place, or kind of food. The emotional charge of a specific event, either positive or negative, is in fact tightly linked to its probability of being remembered. When this charge is very strong, the memory is burned into the brain as if with a branding iron, and it becomes what is called a flashbulb memory: the memory of Neil Armstrong walking on the moon, of the attacks on the World Trade Center, or of Maradona’s World Cup goal against England. Curiously, we may remember these events in great detail but have no idea of what happened in the days before or after their occurrence.
Finally, we can further classify memories according to the type of sensory information they involve. We have visual memories, like the features of a familiar face (which reside in the part of the cerebral cortex dedicated to the processing of visual stimuli), auditory memories, like the timbre of a trumpet (which reside in the auditory cortex), and so on. Different aspects of a single memory may be stored in different areas of the brain according to the sense they involve. The information provided by the various senses can be combined into multisensory memories (for example, when we remember both the lip movements and the sound produced to say the word “mom”) and converge in the hippocampus. There resides a much more advanced representation of memories, to which our next chapter is devoted: memories of concepts.
Chapter 8
HOW DOES THE BRAIN REPRESENT CONCEPTS?
In which we discuss the visual perception pathway and the recording of individual neurons in humans, the discovery of the “Jennifer Aniston neurons,” and the critical role these neurons play in the formation of memories
In outer space there is only silence, an eternal, immutable silence that is not even disrupted by the explosion of a supernova. Sound exists only in a very small fraction of the universe; notably, this fraction includes Earth, our planet. An astronaut spacewalking about his space station would hear absolutely nothing if the station were destroyed by a meteor shower. He would witness the events as though watching a silent movie. Sound, as we experience it, is created by variations in air pressure. Strictly speaking, sound doesn’t even exist as such in the atmosphere. Sound—the voice of a friend, a nocturne by Chopin, the crack of a thunderbolt—is a construction generated by the brain from the vibrations of small hairs in the ear that transform pressure variations into nerve impulses. If a Martian were to materialize suddenly upon our planet, it would be pointless to try to talk to him, and not because he would not understand Spanish, English, or Arabic: he would simply be unable to hear, to perceive or interpret subtle variations in air pressure, because there is no air on Mars and he would not have evolved structures like the ear.
Just as with sound, color doesn’t exist as such around us; what actually exist are electromagnetic waves that strike our retina, and color is just our interpretation of these. In the initial chapters we gained a general understanding of the way the brain extracts meaning from what we see. As Aristotle and, later, Aquinas argued, we generate images based on external stimuli, and these images in turn give rise to the formation of concepts, which are the basic units of human thought. But what exactly is the process that generates these constructions of increasing sophistication? What is its physical, neural basis? This fascinating topic has been dominant in neuroscience in recent decades, and I have been lucky enough to be involved in investigating it.
As we described in previous chapters, the visual process begins in the retina, where photoreceptors transform photons of light into the firing of neurons. The retinal ganglion cells, whose axons make up the optic nerve, encode local contrast—in other words, points that stand out from their surroundings—which are transmitted through the lateral geniculate nucleus in the thalamus to the primary visual cortex, or V1, located in the rear of the brain. It is in this area that David Hubel and Torsten Wiesel (disciples of Stephen Kuffler, whom we mentioned in Chapter 3) discovered, in experiments with cats and, later, monkeys, neurons that respond selectively to lines at specific points in space and with particular orientations—for instance, some to vertical and some to horizontal—a discovery that earned them a Nobel Prize in 1981. Much as the center-surround organization of retinal cells results in information about local contrasts, this selective neural organization in the primary visual cortex gives rise to information about the lines constituting an image. This information is then further processed by other regions of what is known as the ventral visual (or perceptual) pathway until it arrives at the inferior temporal cortex (IT), where, as was found in experiments with monkeys, there reside further specialized neurons that respond, for instance, solely to faces (and not to other images like hands, fruits, or houses).1 Thus, the neurons in the different areas along the ventral visual pathway encode increasingly complex information: we move from a representation of local contrasts in the retina, to one of borders in V1, to one of faces in the inferior temporal cortex.
Approximately 20 percent of patients with epilepsy have seizures that cannot be controlled with medication. Sometimes such episodes bring about a major decline in quality of life, and if the seizures happen to have their genesis in nonvital parts of the brain, a possible treatment is the surgical removal of the so-called epileptic focus. In the previous chapter we described the case of H.M., whose two hippocampi were removed in the 1950s in an effort to cure his epilepsy. As we saw, the procedure had a catastrophic outcome, for H.M. was unable to form new memories after the surgery. The hippocampus is often involved in the origin of epileptic seizures, but its surgical removal these days generally entails no significant collateral damage. The difference is that, today, no surgeon would remove the hippocampus from each hemisphere of the brain, as was done in H.M.’s case. Instead doctors remove just one hippocampus, never both, after first identifying which is the one causing the seizures.
Before attempting such surgeries, it is obviously critical to locate the epileptic focus precisely. In some cases, this can be done based on clinical evidence and magnetic resonance imaging. In other cases, this information is inconclusive, and it is necessary to implant intracranial electrodes
in the brain to localize the epileptic focus as accurately as possible. The decision of when and where to implant the electrodes obviously varies from patient to patient, but, given the above-mentioned prevalence of the hippocampus’s involvement in epilepsy, electrodes are often implanted there and in the surrounding structures, in what is known as the medial temporal lobe.
Figure 8.1: The perceptual pathway
Neurons in the primary visual cortex (V1) respond to lines of a given orientation (a vertical line, in this case). This information is transmitted to higher visual areas through the so-called ventral visual (or perceptual) pathway and ends up in the inferior temporal cortex (IT), where neurons have been found to respond to more complex stimuli, such as faces. The information from IT is then transmitted to the hippocampus.
Technological innovations developed at UCLA have resulted in intracranial electrode recordings that allow us to see the activity of individual neurons in the human brain. The chance to perform such studies was what led me to enroll as a postdoctoral researcher in California with one of my mentors in neuroscience, Christof Koch, and collaborate with Itzhak Fried, one of the neurosurgeons who established this line of research.2 Details aside, our initial experiments were in principle very simple: as we recorded the activity of up to around 100 neurons, we showed patients image after image on a laptop to see if any neuron responded to any of the images. Given the abovementioned responses to complex visual stimuli in the inferior temporal cortex, and given that these neurons then send that information on to the hippocampus and the structures that surround it, one would in principle expect a very advanced representation to arise in the hippocampus, a response more sophisticated than one to contrast, lines, or even faces. And yet, in spite of these expectations, what we found was beyond anything I had ever imagined. I still remember like it was yesterday the time I saw the first of these responses; I remember jumping from my chair and watching the computer monitor in amazement. I had seen, for the first time, a neuron that responded to a concept.3 And, curiously, this concept turned out to be no more and no less than Jennifer Aniston.
The Jennifer Aniston neuron, as it is currently known in scientific discussions and even in neuroscience textbooks, responded to seven different photographs of the actress and to no other picture we displayed—including eighty of celebrities like Kobe Bryant, Julia Roberts, Oprah Winfrey, and Pamela Anderson, as well as photos of ordinary people, places, and animals. In the same experiment, with the same test subject, I found a neuron that responded only to photographs of the Leaning Tower of Pisa, one that fired when shown the Sydney Opera House, one that responded to pictures of Kobe Bryant, one that preferred photos of Pamela Anderson, and so on.4 I had chosen these photographs because those particular people and places were very familiar to the patient, and I assumed—correctly, as it turned out—that well-known things are represented by more neurons (since they have more memories and associations related to them) and would be more likely to elicit responses.5
Figure 8.2: The Jennifer Aniston neuron
Responses of a neuron in the hippocampus that fired in response to different photos of Jennifer Aniston and did not fire in response to images of other people, places, or animals. (To save space, we show only four of the seven photos of Jennifer Aniston, and only eight of the eighty other pictures that were used.) The thick lines show the average response to six presentations of each picture. Each image was shown to subjects starting at time zero.
In another patient, we recorded a neuron that responded only to photos of actress Halle Berry, including those of her dressed as Catwoman, a character she played in the movie of the same name. What is notable about this latter response is that Berry’s face was almost completely obscured, yet the patient knew it was her, and so the neuron responded accordingly. Even more interesting was the fact that this neuron responded to Berry’s name written on the screen, proving beyond doubt that it was reacting to the concept and not to particular visual features in the pictures we used. As in the preceding case, the neuron did not respond to photos of other people, places, or animals, or to any other written name.
Figure 8.3: The Halle Berry neuron
Responses of a neuron in the hippocampus to different pictures of Halle Berry, to Halle Berry in costume as Catwoman, and to her name spelled out on the computer screen
A third example worth highlighting is of a neuron that responded to different photographs of me and to my name, whether written on the screen or voiced by a computer. This result (and many more)6 shows clearly that the responses of these neurons can be prompted by different types of sensory stimuli. Logically, this makes sense: Seeing a photograph of a person, or reading or hearing that person’s name, all give rise to the same concept. However, the processing that takes place in the brain is completely different in all three instances—involving visual areas in the case of photos and written names, and auditory areas in the case of names voiced by the computer—and yet all of these stimuli end by eliciting similar responses in single hippocampal neurons. Another interesting point is that, a couple of days before we carried out the experiment, the patient in question had not met me, had never seen my face, and did not know my name. This means that the encoding of concepts by neurons in the hippocampus is relatively fast; it may take a couple of days, maybe a few hours, or perhaps just seconds.
Thus it seems that we have found a neural basis for the abstractions contemplated by Aristotle and Aquinas. From the first responses in the retina, through the processing of information along the ventral visual pathway, we arrive at last at an encoding of concepts, the meaning that we extract from stimuli. But why are these neurons doing this? What do we gain by encoding concepts in the hippocampus? The answer is given by the responses of the neuron in the next example, and by revisiting the evidence provided by the case of H.M.
A neuron in the entorhinal cortex (an area adjacent to the hippocampus) responded to several photographs of Luke Skywalker (as played by Mark Hamill in the Star Wars movies) and to the name Luke Skywalker, both written on the computer screen and spoken by a synthesized voice. Nothing new so far. However, this same neuron also responded to Yoda, another Star Wars character closely related to Luke Skywalker.
Figure 8.4: A neuron that responded to my pictures and my name
Responses of a neuron in the hippocampus to different pictures of me and to my name, Rodrigo, both spelled out and spoken by a computer (the latter in the lower right corner). This neuron had similar responses when presented with the photographs and names of three colleagues who performed studies with this particular patient.
Why is it interesting that the Luke Skywalker neuron also responded to Yoda? This example, among many others,7 shows that these neurons can respond to related concepts. In other words, they encode the connections that we keep in our memory. And, indeed, it is these associations between concepts that form the very core of memory itself.
Figure 8.5: The Luke Skywalker neuron
Responses of a neuron to three pictures of Luke Skywalker and to his name, written and spoken (lower right panels). The neuron also responded to Yoda, another character from Star Wars.
Let us examine this piece by piece. Given the undeniable evidence provided by H.M.’s case and others like it, we know that the hippocampus and its surrounding areas are involved in the formation of declarative memory, the memory of events and concepts: H.M. was unable to generate new memories from the moment his hippocampus was removed. It is no coincidence that, precisely in this area, we have neurons that encode concepts, since, as we’ve seen, we tend to remember abstractions and forget details. As I write these lines, I am conscious of multiple circumstances: what it is that I want to say, what words I will use, what I am wearing, the details of a trip I will take tomorrow to Seville for a conference, etc. However, in a few months, or even a few days, I will remember, if I’m lucky, only a few general ideas (perhaps the fact that I was writing about concept neurons the day before my trip to Seville); the details will have been lost. (W
hile revising the text of this English translation, a few years after writing these lines for the original Spanish version of the book, I notice that I don’t remember at all my whereabouts when writing this paragraph or even this chapter. It all fused together as “the memory of writing the book.” I do remember, however, some of the circumstances of my trip to Seville, which was a significant departure from my daily routine. I remember giving a talk—I don’t remember exactly about what, but I assume it was about concept cells—and going for a long walk with my friend and colleague Gonzalo Alarcón, who impressed me with his knowledge of the architecture of Seville Cathedral. I also remember meeting a former student in the lobby of the hotel, with whom I discussed a paper, and I remember meeting Miguel Ángel Gea, the magician I talked about in previous chapters, who took me to a bar that charged only €0.40 for a bottled beer—so cheap that I still remember the price—and then to a steakhouse for dinner. That’s basically it. Of something so seemingly memorable as a multiple-day trip to Seville, I can remember barely a handful of concepts that I have linked together in my memory. All the rest faded into oblivion, or are details—like those of the content of my talk—that I guess based on reasonable assumptions.)8
It further makes sense that the concepts we find these neurons responding to tend to be familiar ones, since these are the exact concepts we are most likely to keep in memory. (I am certain to remember my mother if I see her walking down the street, but it is unlikely that I will remember seeing someone I don’t know.) Moreover, it is no coincidence that these neurons encode associations, since associations constitute the basis of memory: on my trip to Seville, I generated episodic memories supported by the associations between the concepts involved (Gonzalo, Gea, my student, the cathedral, the €0.40 beer, and so on).
The Forgetting Machine Page 9