The Happiness of Pursuit: What Neuroscience Can Teach Us About the Good Life (12 page)

Read The Happiness of Pursuit: What Neuroscience Can Teach Us About the Good Life Online

Authors: Shimon Edelman

BOOK: The Happiness of Pursuit: What Neuroscience Can Teach Us About the Good Life
10.32Mb size Format: txt, pdf, ePub
It is usually far easier and more effective to squeeze through the genomic bottleneck traits that facilitate the gaining of knowledge rather than ready-made knowledge as such. Imagine a neural circuit that solves a particular cognitive problem by making use of knowledge encoded in its pattern of synaptic weights, each of which is a numerical representation of the strength of the “kick” that the presynaptic neuron’s activity can deliver to the postsynaptic one through their connection. As I already noted, for this knowledge to be “innate,” the weights would have to be dictated by the genome, raising a host of technical challenges such as installing all the right numbers in the right places.
In comparison, a learning-based solution to the same problem would require that the genome specify merely the manner in which the weight of each synapse changes in response to the activities of the neurons it connects. This extremely powerful synaptic modification rule, which relies exclusively on information that’s available locally at the synapse, is actually much easier to specify in terms of an environmentally driven, genetically controlled sequence of biochemical events, making the evolution of learning from experience possible.
10
It seems, then, that the most valuable lesson that evolution offers, to all who would listen, is that the world is inconstant but learnable and that a good living can be made by those who can learn faster than it changes. (It may also seem that not everyone need listen: if you’re covered in impregnable scaly armor or are too poisonous even to look at, you may feel exempt from having to be also smart, while in truth you’re just one mass extinction event away from oblivion.) Accordingly, the most valuable present that an animal may receive upon being born, or hatched, or booted up from cold storage, is the ability to learn from its own experience and from that of its peers. Like the representation of the world on the Shield of Achilles, innate knowledge may be beautiful, yet it is wholly of the past: it depicts or replays the same old scenes. The quick-witted Ulysses, who argued so eloquently that his wisdom made him the most deserving of the gift of the shield, was also the one who needed it the least. Having won it in the debate with Ajax, Ulysses did not keep the Shield of Achilles, but gave it to Neoptolemus, son of Achilles, before setting sail for Ithaca.
Mirroring the World, Mustache and All, One Step at a Time
 
The most literal manner in which the brain may attempt to anticipate the future is by learning representations whose unfolding over time reflects the dynamics of the events that they stand for. Such mirroring of one dynamical system by another is not a trivial matter. Unless the two systems are in every detail identical and are identically connected to the rest of the world, their trajectories will sooner or later diverge, even if initially they unfold in lockstep. This representational falling out receives an exemplary treatment in the mirror scene in the Marx Brothers’ 1933 movie
Duck Soup
.
As any card-carrying Marxist will tell you, in this celebrated scene Chico sets out to impersonate Groucho, with the aid of a painted mustache, black-rimmed round glasses, a fake nose, and an unlit cigar. He then encounters an identically dressed real Groucho, who decides that he is seeing himself in a mirror. (Both Groucho and the alleged reflection are wearing long white nightgowns, nightcaps, and socks.) Chico plays along by trying to mirror Groucho’s every posture and move. The two make faces, wiggle their behinds, go on all fours, and do a couple of silly walks that may have inspired the much later opus by Monty Python. The spell is broken when Harpo, also made-up and dressed as Groucho, joins them, with predictable consequences.
The two Grouchos’ aping of one another is funny because it is sustained for several minutes, because it is occasionally imperfect, but most of all because while it lasts it appears improbably well coordinated. They begin each silly walk while separated by a wall; they then amble, skip, and sashay in near-perfect unison across an open space where they are in plain view of each other; then another wall comes between them and the increasingly suspicious real Groucho starts plotting his next move, designed to expose his “reflection” as a fake. The fake Groucho’s success in mirroring the real one’s moves makes us laugh because intuitively we know just how unlikely it is that two people could execute a dance in lockstep without orchestrated timing and a thorough rehearsal.
11
Rehearsal on the part of the system that aims at representing the world helps it shape its dynamics by giving
experience
an opportunity to make its imprint. How this happens is best understood using the no. 4 conceptual tool of representation space, from Chapter 3. Under conditions that favor learning, experiencing a series of stimuli—activating in quick succession a series of points in a representation space—causes them to become associated with one another, in the order of their activation. Repeatedly traversing such a learned trajectory consolidates it into a memory trace that is both a record of past experience and a basis for prediction. If a later event has just caused the first and then the second element in a sequence of representations to be activated, chances are that whatever it is that causes the third element to become excited will come along soon; the system now has some idea what to expect next.
Computationally, the knowledge of ordered sequential dependencies among representations has the form of conditional probabilities. This means that its acquisition and use obey the Bayes Theorem—the Promethean gift of probability theory to cognition. Once learned (through accumulation of experience, subjected to statistical inference), a pattern of sequential dependencies can be used to predict where a sequence is likely to go, given where it comes from in the space of possibilities. Like a quad on a college campus or a pedestrian square in a city after a heavy snowfall, this initially pristine representation space becomes covered with a skein of forking paths, some deeper and wider than others, which grow in response to experience.
Our encounters with the world come in fits and starts, one event at a time, with not much happening in between. Not all possible situations that could in principle be represented given the brain’s resources get to be experienced, and those that do are not experienced all at once. This is what imparts to a possibility space the characteristic structure of a crisscrossing network of paths that run through an otherwise untrodden territory. The paths are punctuated with occasional stops that correspond to distinctive, hence memorable, events. At each of these stops, there is a cache of information. Because different tasks require different kinds of possibility spaces, the contents of those representational caches, as well as the pattern of paths, vary from one task to another.
The least abstract possibility spaces are those that represent aspects of actual physical space and time. To learn how to aim, or aim at, a flying object, I need to represent its possible locations and velocities. If the object is ballistic and therefore travels along a parabola, like a catapulted cow would, learning and subsequently predicting its trajectory is a simple matter of estimating a few parameters. If the object is self-propelled and has a mind of its own, like a hummingbird that flits here and there and then brakes in midair in its approach to a hibiscus flower, I need to learn and represent something of its mind.
If the object to be tracked is self-propelled as well as articulated (not the same as articulate, an attribute I’ll get to in due time), and if it is acting willfully, as Groucho Marx habitually would, the problem of anticipating its moves—that is, estimating the relative probabilities of various likely future moves, given the past ones—is considerably more complex. The observer working on populating the possibility space with data must in this case learn the likely changes in the object’s location (relatively easy), bodily configuration (tricky, especially if a silly walk is in the works), and mind state (very tricky, but not out of the question if you are at least as clever and observant of your object of attention as Chico Marx appeared to be of his brother).
All these learned patterns of change are represented as paths through possibility space, which I shall call PaThS (an almost-acronym that is easier to pronounce than the cryptic “PTPS” and more transparent than the pedantic “PaThPoSp”). The idea of a path through a representation space has been with us since the previous chapter, where I invoked it to explain how brains counter the effects of the changes in the visual appearance of objects, such as faces, that are brought about by changes in vantage point. As you will recall, the brain makes sense of a vaguely familiar face seen for the first time from a new angle by first populating a face representation space with tracks that record how familiar faces change appearance with vantage.
By now, it should be clear to us that the dimensions of the possibility space can stand for anything at all. Suppose that Chico undertakes to represent Groucho’s state of mind by focusing on just two of its qualities: affect and arousal. For this purpose, Chico needs to allocate two dimensions of his possibility representation space—the mood plane, as it were. Within it, various conceivable mood shifts of Groucho’s would be represented by PaThS of appropriate shape and placement. For example, a transition from being quietly happy (low arousal, positive affect) to being violently miserable (high arousal, negative affect) would correspond to a diagonal move from one corner of the mood plane to the opposite one.
Although most of the things and events that populate the possibility space are quite abstract, the physical environment not only gets represented within it but serves as a scaffolding that supports and structures the rest of the possibility space. The representation of the environment is privileged in this manner for a simple reason: it anchors in place episodes and actions and imposes order on sequences of events, which unfold as the learner experiences his or her or its corner of the world. As the learner gets older, avoiding being buried under a pile of indiscriminate memories becomes more and more of a challenge. If time is nature’s way of keeping everything from happening at once, then space—or rather, the memory of space—is nature’s corkboard, where everything that happens has its place.
Where Was I?
 
The evolutionary roots of the spatially indexed
episodic memory
system lie in a very common existential need. How can an animal that must leave its burrow to make a living improve the odds of getting back home at the end of the working day, preferably with provender for the family? By keeping track of what happened where on its last few forays into the wild, with an eye toward drawing generalizations about what usually, or at least sometimes, happens where (which would enable the itinerant animal to anticipate what will happen where the next time around).
Put yourself in the place of a desert bighorn sheep. You live near Borrego Springs, California. It is early August, and the heat at noon can be deadly. The neighborhood is patrolled by a puma, which, unlike yourself, is not a herbivore. Your only navigation device is your brain. If you have no idea where you are or how to get back to that water hole you drank from the other day, taking in local attractions would be the last thing on your mind, even if you love the desert as much as I do. And yet, it is the same set of cognitive computational skills that makes you good at sightseeing (and remembering the sights) and at getting back from point B to point A.
Reliable and effective way-finding, orienteering, and hoarding of personal episodic experience all depend on a representation of the layout of the environment. How does the brain do it? An effective representation supports navigation not by indicating where its owner is on some kind of mental map—if it did, bighorns would be in need of map-reading instruction and training as badly as most army cadets are. As any number of street-corner tourist maps in big cities will tell you, the answer to the question “Where am I?” is “You are here,” which is true, but not very helpful. One would hope that the brain can do better than that.
Indeed, instead of being like tourist maps, which require interpretation, representations of space in the brain directly assist way-finding behavior by explicitly encoding various useful cues, such as up-to-date direction information to certain key locations. While I am confident that you know precisely where you are at this moment (
there
), I believe also that you know more than that. In particular, I bet that you can point in the direction of the closest source of potable water or food, and not just because I asked you earlier to imagine life as a bighorn. Speaking for myself, a decisive demonstration of my knowledge of where food is relative to where I am now would be for me to get up and head
directly
to the fridge, which is not visible from where I am writing these lines. (This task is not entirely out of the question: the paper-and-plaster walls of my house are no match for this grizzled specimen of
H. sapiens.
)
When the ability to take shortcuts through new territory between previously visited locations was first discovered in the rat, in the 1940s, scientists interpreted it as evidence for the existence in the rat brain of cognitive maps. Recording from the hippocampus, an area of the brain that had been implicated in navigation, they eventually discovered neurons that fired at a high rate only when the rat visited particular locations. These “place cells” serve as the foundation for representing space—and, it turns out, much more.
Location cues derived through dead reckoning from wandering about are enough to form the hippocampal representation of space: much of rats’ way-finding is done in the dark of the night, and even blind mole rats (such as Molly from Chapter 2) start taking shortcuts across the open space in the middle of an enclosure after exploring its boundaries. At the same time, any additional cues that happen to be available get tacked right onto the basic spatial scaffolding. Thus, the so-called place cells respond selectively not just to particular places but to the combination of sights, sounds, smells, and textures encountered in those locations—in other words, to episodes that the animal experienced there.
12

Other books

Jacob by Jacquelyn Frank
When Cicadas Cry by Laura Miller
Resolved by Robert K. Tanenbaum
Eleven Twenty-Three by Jason Hornsby
The Man From Saigon by Marti Leimbach