Six - A BEND IN THE ROAD

Six

A BEND IN THE ROAD

HOW LOVE CHANGES WHO WE ARE

AND WHO WE CAN BECOME

It is 7:15 and twilight is gathering around the hospital, when a man enters the emergency room. He is fiftyish, with a paunch to his belly and a grayish tint to his face. He has chest pain, he tells them at the front desk, a gnawing, persistent discomfort behind his breastbone. The triage nurse charts his vital signs: heart rate, blood pressure, and respiratory rate, all elevated. He dons the obligatory paper gown and takes his place on a gurney as an intern readies the cardiac monitor.

If a wayward clot or a shard of crystalline cholesterol has blocked a cardiac artery, part of his heart will die, and he will need admission to the hospital, where the staff and their mechanical aides will try to save his life. But as is so often the case in medicine, his symptoms do not point in one direction: they fit with dozens of ailments, most comparatively harmless—an ulcer, an anxiety attack, a pulled muscle, an undigested bit of beef. If everyone with his chest pain were admitted, our health care system would reach bankruptcy sooner than it will on its present course. No diagnostic test is definitive here; the situation calls for expert judgment applied to stubbornly indeterminate facts. Thousands of times a day, armed with less information than can lead to certainty, someone decides: is this person having a heart attack?

Physicians have scrutinized this question for centuries, gathering the slender hints and occult signs whereby an ailing heart betrays its presence. They understand what a heart attack is, how and where and why it occurs, whom the risk factors prey upon most avidly. Still, medical skill at detecting myocardial infarction is far from perfect. William Baxt in the Department of Emergency Medicine at the University of California, San Diego, hoped to augment human diagnostic proficiency by enlisting a computer assistant capable of learning to distinguish heart attacks from the varied conditions they resemble. Baxt fed the program details from the case histories of 356 patients. In the next 320 chest pain cases, physicians made the right call four fifths of the time. The computer scored 97 percent.

The de facto division of labor between man and machine usually splits along pleasing lines. Computers typically excel at the tireless repetition of algorithms—the menial mental labor that human beings have little inclination to undertake. Supercomputing inroads into championship chess rely on mechanical combinatorial might, not clever strategy. Prodigious calculators they may be, but our smartest machines can’t make sense of a simile, summarize a sitcom plot, or take the dog for a walk. One wonders, then, how a few hundred lines of programming code, with no ability to understand anything about cardiac physiology, so rapidly outpaced human physicians in arriving at a medical diagnosis.

A human being has dual hearts—the first, a pulsating fist of muscle in the chest; the second, a precious cabal of communicating neurons that create feeling, longing, and love. The two hearts intersect for a moment here because the program that so brilliantly assessed cardiac endangerment is a neural network—poetically named, for it is neither neural nor network, but a series of mathematical statements that model the brain’s own computing agility.

The modus operandi of the neural network is unique. Standard computer software runs on the expertise, routines, and contingent responses that human masterminds script in advance. Such a program cannot handle any situation that the writer cannot foresee; once written, the program itself does not change. The meticulously crafted core of a neural network, on the other hand, learns from experience and transforms itself—a capacity created by small-scale software modeling of the communication that occurs among organic, brain-dwelling cells. Before a neural network can compute a solution, it first absorbs information from intensive training sessions. That learning phase gradually alters the program’s innards. Neural networks (also called parallel-distributed processing or connectionist models) excel at gleaning subtle patterns that hundreds of variables jointly determine. The best neural networks are more astute at diagnosis than doctors, better at forecasting weather than meteorologists, and more profitable stock pickers than mutual fund managers.

A neural network is machine intuition. After a connectionist program delivers its answer, one cannot obtain meaningful knowledge about the processing details—why it says Patient A suffered a heart attack but Patient B did not. Examining the network yields minimal information about the basis for its conclusions. Because a neural network taps into the brain’s own data-processing mechanism, it arrives at sophisticated, unanalyzable inferences—as does humanity’s emotional heart. Understand how a neural network functions, and you will know the innermost secrets of the intuitions that guide us in love.

In the previous chapter, we looked at memory from a macrocosmic perspective—how a person remembers and learns. Neural network theory begins at the opposite, miniature pole: mathematicians translate the brain’s memory-making steps into equations, implement them on computers, and strive at elevating their apprentices into better learners. They have not only produced a program wise in the ways of heart attacks, but they have also radically altered how scientists think about living mechanisms of learning. And the quest has come full circle: the thriving new field of computational neuroscience applies the heady mathematics of connectionism back to the original problem of divining how brains—and thus human lives— operate the way they do. These discoveries reveal a last, luminous power of the limbic domain: love alters the structure of our brains.

MEMORIES ARE MADE OF... WHAT?

Under the right conditions, a collection of neurons can learn. A rat that runs a maze, a German shepherd that sits on command, and a child who recites the Gettysburg Address reflect the ability of a nervous system to record information and hold it in abeyance. Years may elapse before the rat, dog, or child uses that preserved data to influence the muscular contractions we know as behavior. Everyone who attended an American elementary school can remember, for instance, how the Gettysburg Address begins, the initial syllables rising to the surface of the mind and the tip of the tongue as readily as a magician might draw a chain of knotted scarves from Lincoln’s stovepipe hat. Recalling this minute datum is a masterful conjuration in physiology: that single improbable sequence is somehow inscribed within a group of living cells. It is one mote among billions, encrypted in suspended animation, and the organic function of those cells whisks it into consciousness on a moment’s notice. Lincoln’s apparently unforgettable phrase remains embodied in the marvelous entanglement of the brain’s cellular components for as long as they live. But how?

Any system that aims to warehouse data forms a material record. The King James Version Bible stores its truths in scattered dots on a page, “The Mona Lisa” in pigments on a canvas, a compact disc in pits on a glittering plastic platter. The mechanisms may vary, but the end is unswerving: a durable physical representation of knowledge.

The brain has no dots or paints or pits, only neurons. Every mental activity—contemplating a theorem, savoring a hot fudge sundae, dreaming of the boy next door—consists of neurons firing in a certain sequence. When neurons cache Lincoln’s remarks at Gettysburg, the specific neural sparkle defining that string of words must be made to last. Millions of neuronal flash patterns course through the brain every minute. How does any one attain permanence?

MEMORY’S METAMORPHOSIS

Consider a simplified network model for storing and retrieving a sensory input. Suppose that in this array, weak tendrils join each neuron to every other:

This network is naïve; it stands ready to receive experience but has recorded none.

A sensory input arrives:

A pianist transforms the cryptic swirling of dots and lines on the music in front of him into the fluttering of his fingers, and then into beauty. Here the reverse occurs: the richness of a sensory experience is translated into the brain’s peculiar notation system— no staffs or semiquavers; instead, neurons fire.

One needn’t linger over the details of that conversion—why this neuron and not that one? Somewhere, an array of neurons depicting sensory data about the stimulus will flicker into brief life. A brain needs millions of neurons to portray such a symbol, but this example will proceed more comfortably on a smaller scale—in this neural network, registering those crisscrossed lines necessitates just sixteen. To make a memory, the network immortalizes the association of that particular group by strengthening their previously faint linkages.

When the figure passes from view and the cells quiet down, the skeletal remnants remain:

The fortified conjunctions permit these neurons to fire together again. When a few go off (A), they trigger their quiescent fellows along the slightly worn paths between them. Like a string of dominoes, they race to tip each other to a common fate. The old pattern is rejuvenated (B), and with it, a recapitulation of the original character.

This storage scheme, the brainchild of psychologist Donald Hebb, is a powerhouse. Hebb proposed the mechanism a few years after World War II drew to a close. Only within the past fifteen years, however, did researchers explore its mathematical premises and build large-scale computer models of Hebbian learning. Both endeavors—the mathematical insights and their implementation in computer simulations—have illuminated more than a few of the mysteries about why people think and feel the way they do.

Hebb’s central proposal remained theory until the advent of experimental techniques for taking electrical measurements from individual brain cells. In a refreshing physical affirmation of mathematical abstraction, the data demonstrate that neurons in the living brain behave as Hebb predicted. The brain makes memories by enhancing the couplings between concurrently firing neurons.

A printed page in this book may survive for hundreds of years. CDs have a shelf life of one or two decades and the average Etch A Sketch scrawl lives minutes at most. A neuron’s “on” state lasts for a thousandth of a second. The ephemeral span of its signals forces the brain to render present and past data differently. At any instant, the precise configuration of firing neurons specifies what a brain is representing now. But the past lies dormant within the network’s structure, formed by accumulated links of varying potency. Each constellation of mute connections embodies the potential for a previous ensemble to be reanimated and remembered. This separation of past and present makes the network into a living and eccentric time machine.

Every bit of life impinging on the brain changes some of its links, although any individual datum affects only a minuscule fraction of the innumerable totality. As subtle changes accrue, experience rewires the microscopic structure of the brain—transforming us from who we were into who we are. At a Lilliputian level, the brain is an elaborate transducer that changes a stream of incoming sensation into silently evolving neural structures. Minor events exert only a transitory alteration in a few far-flung neuronal ties, while formative experiences lay down resilient patterns that prevail for a lifetime.

And the limbic brain, wherein some of those cryptographic patterns reside, can reach beyond the frail borders of one mind and into another—a fact with far-reaching consequences.

THE ECHO’S RING

Construct a neural network and set it in motion, and its odd memory mechanism begins to work divisive magic. Neurons that fire together once tend to do so again as they become bound to one another with increasingly close ties. Cells that are never on simultaneously start to suppress one another. The once homogeneous network spontaneously fractures into squabbling cliques. Team members spark one another to fire en masse. Opposing squadrons fight for the chance to be active. At any moment, a single neuron receives stimulation from compatriots and inhibitory signals from enemies. As the cells trade their flurry of pros and cons, each finds at its own activity level. The network as a whole then settles into a certain conformation of active and inactive units. A network is most stable when a set of allies is firing—when one team wins.

In the brain, some neurons receive input from ten thousand others and may deliver outgoing messages to ten thousand more. With such extensive dissemination of signals, a simplifying assumption of total commingling is not far off the mark. Just as team member cells goad one another into excitability, so do compatible networks motivate one another in the brain. Likewise, dissonant networks compete and drive one another down. The unforeseeable result of this catholic discourse is that emotional memory skirts the linear flow of time.

When a network turns on, it immediately dispenses electrical encouragement to every other concordant network. And the secondaries light up in direct proportion to their shared affinity. If network A fires, and B is highly compatible, then A will galvanize B. This companionable coactivation echoes on: with B aroused, its allies will awaken, and so forth. Like ripples blooming from a central splash, memory networks spread out along lines of similarity, bringing the most mutually congruent to life, fading in influence as correspondence drops off.

Think the word dog, and the circuits encoding for German shepherd and golden retriever warm up in your mind, and those for walk and bone and flea a little less so. The strong activation of dog leaves mutual fund in hibernation (except in the unlikely event of an idiosyncratic bridge—today Fido ate your brokerage statements, say). Computer-based neural networks operate this way, and so do human beings. Showing a person the word dog actually makes him respond more quickly to words like bone and flea, while reaction time to mutual fund remains unchanged.

For those animals with a limbic brain, emotionality forms a principal dimension of that associative network. In place of dog retrieving bone and walk and flea, a particular emotion revives all memories of its prior instantiations. Every feeling (after the first) is a multilayered experience, only partly reflecting the present, sensory world.

In chapter 3, we saw that the evanescence of emotions, their pulse-and-fade propensity, is nearly musical. Now the metaphor draws closer. A musical tone makes physical objects vibrate at its frequency, the phenomenon of sympathetic reverberation. A soprano breaks a wineglass with the right note as she makes unbending glass quiver along with her voice. Emotional tones in the brain establish a living harmony with the past in a similar way. The brain is not composed of string, and there are no oscillating fibers within the cranium. But in the nervous system, information echoes down the filaments that join harmonious neural networks. When an emotional chord is struck, it stirs to life past memories of the same feeling.

One manifestation of these orchestral evocations is the immediate selectivity of emotional memory. Gleeful people automatically remember happy times, while a depressed person effortlessly recalls incidents of loss, desertion, and despair. Anxious people dwell on past threats; paranoia instills a retrospective preoccupation with situations of persecution. If an emotion is sufficiently powerful, it can quash opposing networks so completely that their content becomes inaccessible—blotting out discordant sections of the past. Within the confines of that person’s virtuality, those events didn’t happen. To an outside observer, he seems oblivious to the whole of his own history. Severely depressed people can “forget” their former, happier lives, and may vigorously deny them when prompted by well-meaning guardians of historical verity. Rage affords hatred an upper hand that is likewise obtuse, sometimes allowing a person to attack with internal impunity those he has forgotten he loves.

The consequences of emotional reverberation in the brain’s networks reach beyond selective amnesia during a dominant mood. A childhood replete with suffering lingers in the mind as bitter, encoded traces of pain. Even a tangential reminder of that suffering can spur the outbreak of unpleasant thoughts, feelings, anticipations. As if he had bumped a sleeping guard dog, the adult who was an abused child may feel the fearsome jaws of memory close after he glimpses a mere intimation of his former circumstance. In a sad empirical confirmation, maltreated children flipping through pictures of faces exhibit a hugely amplified brain wave when they encounter an angry expression.

Other people are troubled by emotional-memory networks that are simply too ready to pass around the signals that comprise negative feelings. Such a person finds he can’t shake an unpleasant emotion once it gets going. Rather than dwindling within minutes as they should, an emotion and its associated repercussions may drown out the rest of his mind for days. That kind of limbic sensitivity makes the thousand natural shocks the flesh is heir to well-nigh unbearable.

Remedies do exist for those whose networks engage in excessive emotional reverberation: some psychopharmacologic agents act as the damper pedal on a piano does, applying a gentle, restraining influence on buzzing strands. Why they possess this property is not yet known, but the impact of these medications on emotional virtuality is what one might expect: emotional chords are quieter and fade sooner. For those whose limbic networks are high-strung, the relief can be lifesaving.

One person, for instance, related that every minor setback made her ruminate for days. “I know it doesn’t make any sense,” she said, “but my boss corrected my spelling on a report the other day, and my mind wouldn’t stop: he thinks I’m incompetent, my work isn’t good enough, I’ll lose my job. All of that is ridiculous, I know. I’m the best manager he’s got. But when anything goes wrong, I just can’t shake this awful feeling.” So prominent was her sensitivity to emotional slights that she retreated from intimacy. No matter how cautious, her partner was bound to do or say something that hurt her feelings, and then she felt terrible for weeks. Being in a relationship, she said, was like trying to dance barefoot—eventually her toes would get bruised and she would flee.

A touch of the right medication diminished her emotional twanging to a normative range. For the first time in her life, she was able to feel a minor pang. She could be upset for half an hour or so, and then get on with her day. “Is this what life is like for everyone else?” she asked. “No wonder they can be in relationships.” With her vulnerability reduced to livable levels, she was readier for love. As she said, she was now dancing with shoes on.

THE ROAD CURVES

As a neural network sees more of the world, its ensuing quirks and kinks confound and complicate the human experience of love.

Here is the network after encountering the first sensory input.

When a second, similar sight presents itself—

—the network again disassembles those features and transmutes them into an assortment of winking neurons. Because Instance Two resembles Instance One, this neuronal rendition overlaps considerably with the last.

Hebbian machinery reinforces these connections:

If we show it a third and then a fourth similar item,

the process repeats: these items, too, utilize many of the same neurons, and the links between them grow ever stronger.

When the network rests after storing the four information sets, the curious aspects of neural memory reveal themselves.

For starters, memory within the brain is fluid. Like a squirrel dispersing nuts to multiple hiding places to thwart theft, the brain scatters its memory treasures across a number of individual connections. The boon of distribution is security; the disadvantage, infidelity. A brain can lose a neuron here and there and the stored data suffers relatively little, a property of neural network memory termed graceful degradation. But as new facts rain down upon and trickle through the network, some older links are dissolved. Dissimilar information patterns can cohabitate in the brain without much mutual disturbance, because they rely on largely different sets of neurons. But like patterns jostle and coincide and overlie, and cannot avoid erasing and emending each other’s contours.

The brain’s past is not carved into the solid rock we would like (and imagine) memory to be. Instead, the story of each life is traced on a sand dune that the winds of time and experience gradually sculpt from one shape to another. After their first instant, our memory traces start shifting away from what they were, and we can never retrieve the pristine data they once encoded.

One inconvenient outcome of this mnemonic drift is the miserable fallibility of eyewitness testimony. Although most people do not realize it, they are incapable of remembering events as they happened. In study after study, they incorporate fragments from earlier or later incidents, general expectations and implications gleaned from leading questions into their memories of what they saw and heard. The resulting ragout of fact, fancy, suggestion, and innuendo feels as convincing as a more accurate memory would. Psychologist Ulric Neisser interviewed forty-four students the morning after the 1986 explosion of the space shuttle Challenger, asking them where they had been when they first heard of the disaster. He repeated the questions two and a half years later. None of the later accounts correlated entirely with the originals, and fully a third of them were “wildly inaccurate.” The students were cheerfully ignorant of the fabrication that passed in their minds for a memory; many insisted that the recent, wrong versions were actually right. “As far as we can tell,” Neisser observed, “the original memories are just gone.”

In a neural network, new experiences blur the outlines of older ones. The reverse is also true: the neural past interferes with the present. Experience methodically rewires the brain, and the nature of what it has seen dictates what it can see.

In our network, a group of neurons stands out by virtue of joint ties: they share connections that the network strengthened three or four times.

This group of cumulative links stores the elements that the four inputs have in common. Blend the quartet together, and their content is manifest:

This composite preserves shared features—dual vertical pillars united by a single horizontal crossbar. Individual elements, such as serifs and swirls, register faintly and tend to wash out. While each of the four contributors is a single, serial data point, the network encodes most strongly their synthetic summary, the trend they exemplify—a roman H. Simply by encountering a series of calligraphic instances, the neural network has automatically extracted and emphasized the H-ness latent within all of them.

Prototype extraction—the distillation of pure, intoxicating principles from the muddle of diverse experience—is the natural, inescapable outcome of neural memory. We have distinct linguistic labels for “memory” and “condensation,” but within the brain they are one. The structure of human thought is shaped accordingly.

Here is the network with the ingrained prototype:

Show it a fifth figure:

This one is seductively similar to the prototype. And when the neural network tries to process the latest input, time spins backward and veracity breaks down.

The fifth figure triggers several prototype neurons, but in a flash, their mutual chain reaction ignites. Like a bank of stadium lights, the prototype team blazes into phosphorescent glory.

The A-team easily outshines the neurons that for one mini-moment encoded the true appearance of the last character. With the resurrection of the A-neuron band, the network departs reality. What it now portrays is a cross between factual present and prototype past:

As the network views it, the last input is an H—the listing twin towers and drooping crossbar have almost assumed the customary crisp relationship that the prototype displays. These interpretive changes to the actual, ambiguous sensory input are made courtesy of the peculiar memory mechanism that a living brain is obliged to employ.

The prototype team constitutes an Attractor—a coterie of ingrained links that can overwhelm weaker information patterns. If incoming sensory data provoke a quorum of the Attractor’s units, they will trigger their teammates, who flare to brilliant life. An Attractor can overpower other units so thoroughly that the network registers chiefly the incandescence of the Attractor, even though the fading, firefly traces of another pattern initially glimmered there. A network then registers novel sensory information as if it conformed to past experience. In much the same way, our sun’s blinding glare washes countless dimmer stars from the midday sky.

Neural Attractors allow human beings to decipher handwriting, wherein the misshapen letters in the sensory stimulus deviate markedly from the straight and true strokes that first-graders strive to imitate and adults have long since abandoned. Translating a handwritten note into English letters is the work of a few milliseconds for a neural network but arduous labor for a computing devicse that processes reality as it is, without first subjecting it to the distorting and defining influence of Attractors. Conversely, Attractors make the act of proofreading exceptionally difficult for neural processors. Reading “taht” in the middle of a sentence, for instance, often activates the brain’s heavily engrained Attractor for “that.” Most of the time, the mind’s inner eye will see only the autocorrected “that,” thus circumventing the opportunity to correct a textual error that consciousness never experienced.

Verify for yourself that reality-distorting Attractors dwell inside your mind.

This is the Kanisza triangle, but the apparent three-point polygon is pure figment: a trio of Pac-Men, six segments, no triangle. Yet the mind’s conviction of triangularity is irresistible, because these suggestive impostors conspire to activate the brain’s hardwired neuronal shortcuts leading to the perception of lines and edges. Try to see only what is really there, and you will find the truth disappearing behind the simpler, less accurate world your brain is determined to deliver.

Or read the following:

The central “letters” in both words are identical, ambiguous twins—duplicates of the fifth figure displayed to our neural network. In the first word, the brain’s learned Attractor for “THE” forces the equivocal figure into an unsteady “H,” while in the second, a different Attractor makes the same lines merge into an incomplete “A.” Although every bit as valid from the perspective of realism, not many people will see “TAE CHT.” Nobody will see the truth: “T?E C?T.”

Einstein’s relativity theory proved that a local concentration of mass warps space, angling the arrow flight of nearby objects, even bending the trajectory of light. The fabric of space, he said, is not a rigid plane like a billiard table or a bowling lane, impervious to the presence of the bodies traveling its surface. Instead, space is like a taut sheet of rubber indented by matter—dimpled lightly by the pea-size mass of a planet, a deep concavity stretching out from the enormous density of a sun.

The brain’s habit of concentrating experience into Attractors likewise makes the mind a pliable Einsteinian fabric strewn with incurvations. At the bottom of each force field well is an Attractor, convoluting the plain, true line of thought and standing ready to exert its influence on information patterns venturing close enough to be twisted or trapped. Even time flows differently in the neighborhood of a mass, as it does in the vicinity of a strong Attractor.

Wallace Stevens on time, heart, and mind:

It is time that beats in the breast and it is time That batters against the mind, silent and proud, The mind that knows it is destroyed by time.

Time is a horse that runs in the heart, a horse Without a rider on a road at night. The mind sits listening and hears it pass.

With time as a stallion loose in the heart, each Attractor compels a bend in the road that horse follows. While sifting through the sensory present, the brain triggers prior knowledge patterns, whose suddenly reanimated vigor ricochets throughout the network. Old information comes alive, and a person then knows what he used to know. Because people are neural beings, the past is potentially vibrant within them.

The limbic brain contains its emotional Attractors, encoded early in life. Primal bias then forms an integral part of the neural systems that view the emotional world and conduct relationships. If the early experience of a limbic network exemplifies healthy emotional interaction, its Attractors will serve as reliable guides to the world of workable relationships. If a diseased love presents itself to a child, his Attractors will encode it and force his adult relationships into that Procrustean bed. Because his mind comes outfitted with Hebbian memory and limbic Attractors, a person’s emotional experience of the world may not budge, even if the world around him changes dramatically. He may remain trapped, as many are, within a virtuality constructed decades ago—and, as Mark Twain observed, a person cannot depend on the eyes when imagination is out of focus.

Limbic Attractors spawn a vexing and fascinating aspect of emotional life—“transference,” Freud’s term for the universal human tendency to respond emotionally to certain others as if they were figures from one’s past. Freud thought transference living proof that a banished memory can escape confinement and hover before a loved one’s features, overshadowing a present angel with a past devil or vice versa.

Science has a way of supplanting myths with no less fantastic truths: transference exists because the brain remembers with neurons. Any system that undertakes Hebbian processing carries out the same distortion, whether the system is living or machine. Computer-based neural network programs are structures whose memory mechanism reduces experience into compact, occasionally fallacious expectancy. So, too, are we.

Because human beings remember with neurons, we are disposed to see more of what we have already seen, hear anew what we have heard most often, think just what we have always thought. Our minds are burdened by an informational inertia whose headlong course is not easy to slow. As a life lengthens, momentum gathers. A wistful aside from two neuroscience researchers:

WHEN WORLDS COLLIDE

No individual can think his way around his own Attractors, since they are embedded in the structure of thought. And in human beings, an Attractor’s influence is not confined to its mind of origin. The limbic brain sends an Attractor’s sphere of influence exploding outward with the exuberance of a nova’s gassy shell. Because limbic resonance and regulation join human minds together in a continuous exchange of influential signals, every brain is part of a local network that shares information—including Attractors.

Limbic Attractors thus exert a distorting force not only within the brain that produces them, but also on the limbic networks of others— calling forth compatible memories, emotional states, and styles of relatedness in them. Through the limbic transmission of an Attractor’s influence, one person can lure others into his emotional virtuality. All of us, when we engage in relatedness, fall under the gravitational influence of another’s emotional world, at the same time that we are bending his emotional mind with ours. Each relationship is a binary star, a burning flux of exchanged force fields, the deep and ancient influences emanating and felt, felt and emanating.

Rachel Naomi Remen, in her book Kitchen Table Wisdom, describes her own brush with another mind’s defining authority. As an adolescent she was ungainly, and her relationship with an older cousin centered on the woman’s acceptance of her embarrassing clumsiness. When Dr. Remen matured to elegant womanhood, she could not escape her cousin’s encapsulated conviction of that outgrown identity. In her cousin’s presence, she reverted to tripping on curbs, dribbling food on her clothes, spilling the contents of her purse across the floor of a restaurant. The inner conception we carry of others “may be reflected back to them in our presence and may affect them in ways we do not fully understand,” she writes. “Over the years, I have come to wonder if it may even be communicated more directly, by the sharing of a private image in a mysterious yet tangible way, as my cousin did with me.”

The limbic transmission of Attractors renders personal identity partially malleable—the specific people to whom we are attached provoke a portion of our everyday neural activity. In the vistas of imagination, the self is a proud ship of state—subject to the winds and tides of circumstance, certainly, but bristling with masts and spars and beams, fairly bursting with solidity. We would scarcely imagine that identity could be as fluid as the seas that supposed self rides upon.

E. E. Cummings paints a lover’s power to render identity in this way:

your homecoming will be my homecoming—

my selves go with you, only i remain; a shadow phantom effigy or seeming

(an almost someone always who’s noone)

a noone who, till their and your returning, spends the forever of his loneliness dreaming their eyes have opened to your morning

feeling their stars have risen through your skies . . .

The reach of limbic Attractors stretches beyond the moment. The sine qua non of a neural network is its penchant for strengthening neuronal patterns in direct proportion to their use. The more often you do or think or imagine a thing, the more probable it is that your mind will revisit its prior stopping point. When the circuits are sufficiently well worn such that thoughts fly down them with little friction or resistance, that mental path has become a part of you—it is now a habit of speech, thought, action, attitude. Ongoing exposure to one person’s Attractors does not merely activate neural patterns in another—it also strengthens them. Long-standing togetherness writes permanent changes into a brain’s open book.

In a relationship, one mind revises another; one heart changes its partner. This astounding legacy of our combined status as mammals and neural beings is limbic revision: the power to remodel the emotional parts of the people we love, as our Attractors activate certain limbic pathways, and the brain’s inexorable memory mechanism reinforces them.

Who we are and who we become depends, in part, on whom we love.