The Forgetting Machine
Page 1
PRAISE FOR THE FORGETTING MACHINE
“Quian Quiroga, himself an imaginative pioneer in the brain sciences, combines state-of-the-art knowledge of the human mind, admirable cultural literacy, and an enticing presentation style. If you wish to take a fascinating and memorable journey into the riddles of human perception and memory, The Forgetting Machine is the gate to enter.”
—YADIN DUDAI, PROFESSOR, WEIZMANN INSTITUTE
OF SCIENCE AND NEW YORK UNIVERSITY
“Rodrigo Quian Quiroga is one of those rare computational neuroscientists who really knows how to bring complex and abstract concepts to a popular audience. This charming and informative book explains current understanding of how memories are encoded in the brain in elegant prose that reflects Quian Quiroga’s engagement with philosophy and the arts as well as hard-core science.”
—ALISON ABBOTT, NATURE MAGAZINE
“The author, a noted brain scientist, takes the reader on an exciting whirlwind tour of vision and memory. His take-home message is that our brains don’t faithfully record the pixels making up any one scene nor do they recall anything but a minute fraction of our life events. Most of what we do, see, and remember is filtered, interpreted, and inferred.”
—CHRISTOF KOCH, CHIEF SCIENTIST AND PRESIDENT,
ALLEN INSTITUTE FOR BRAIN SCIENCE, SEATTLE
“Rodrigo Quian Quiroga highlights for the reader one of the grand challenges of brain science—and, indeed, science as a whole: the quest to understand the mysterious properties of human memory. He does so while providing an eloquently intelligible primer on the processes that underlie our recollections. He leads the reader from an explanation of basic sensory perception through a description of how the brain processes abstract concepts, invoking along the way insights from Aristotle, Plato, and Borges. In an era of terabyte thumb drives, the author emphasizes repeatedly that memory thrives as a uniquely human trait. Analogies to digital recording devices are off base. Human memory distinguishes itself from a mere digital storage device by an ability to continually extract meaning from raw information.”
—GARY STIX, SENIOR EDITOR, SCIENTIFIC AMERICAN
Copyright © 2017 by Rodrigo Quian Quiroga
All rights reserved. No part of this book may be used or reproduced in any manner whatsoever without written permission except in the case of brief quotations embodied in critical articles or reviews.
Originally published in Spanish as Qué es la memoria in 2014 by Editorial Paidós. Copyright © 2014 by Rodrigo Quian Quiroga.
Translation by Juan Pablo Fernández
BenBella Books, Inc.
10440 N. Central Expressway, Suite 800
Dallas, TX 75231
www.benbellabooks.com
Send feedback to feedback@benbellabooks.com
“First E-Book Edition: October 2017”.
Library of Congress Cataloging-in-Publication Data
Names: Quian Quiroga, Rodrigo, author.
Title: The forgetting machine : memory, perception, and the “Jennifer Aniston neuron” / Rodrigo Quian Quiroga.
Other titles: Que es la memoria. English
Description: Dallas, TX : BenBella Books, Inc., [2017] | Translation of: Que es la memoria / Rodrigo Quian Quiroga. Editorial Paidos, 2014. | Includes bibliographical references and index.
Identifiers: LCCN 2017025074 (print) | LCCN 2017025866 (ebook) | ISBN 9781944648558 (electronic) | ISBN 9781944648541 (trade paper : alk. paper)
Subjects: | MESH: Memory—physiology | Perception | Brain—physiology | Nervous System Physiological Phenomena
Classification: LCC QP406 (ebook) | LCC QP406 (print) | NLM WL 337 | DDC 612.8/23312— dc23
LC record available at https://lccn.loc.gov/2017025074
Editing by Alexa Stevenson
Copyediting by Scott Calamar
Proofreading by Rachel Phares and Karen Wise
Indexing by Amy Murphy Indexing & Editorial
Text design and composition by Aaron Edmiston
Front cover by Pete Garceau
Full cover by Sarah Avinger
Printed by Lake Book Manufacturing
Distributed by Perseus Distribution
www.perseusdistribution.com
To place orders through Perseus Distribution:
Tel: (800) 343-4499
Fax: (800) 351-5073
E-mail: orderentry@perseusbooks.com
Special discounts for bulk sales (minimum of 25 copies) are available.
Please contact Aida Herrera at aida@benbellabooks.com.
To my parents,
Hugo and Marisa
CONTENTS
CHAPTER 1
How Do We Store Memories?
CHAPTER 2
How Much Do We See?
CHAPTER 3
Does the Eye Really See?
CHAPTER 4
How Much Do We Remember?
CHAPTER 5
Can We Remember More?
CHAPTER 6
Could We Become More Intelligent?
CHAPTER 7
Types of Memory
CHAPTER 8
How Does the Brain Represent Concepts?
CHAPTER 9
Can Androids Feel?
NOTES
INDEX
ACKNOWLEDGMENTS
ABOUT THE AUTHOR
Chapter 1
HOW DO WE STORE MEMORIES?
In which we discuss the importance of memory, the activity of neurons and their connections, the encoding of memories in the brain, the mechanisms of neural plasticity, and memory storage capacity
The pursuit ends, under torrential rain, on the roof of an abandoned building in a postapocalyptic Los Angeles. Rick Deckard (Harrison Ford), the android hunter, can barely crawl backward as he tries to escape his fate at the hands of Roy Batty (Rutger Hauer), a Nexus-6 android and leader of the “replicants.” Seconds before, Batty hauled a falling Deckard, his enemy, to safety, and now he stands over him while Deckard looks up, confused, afraid, defiant. The replicant observes a vanquished Deckard still fighting for his life and, on the brink of death himself—a death timed and preordained by his manufacturer—he takes a dove between his hands, sits in front of Deckard, and says:
I’ve seen things you people wouldn’t believe. Attack ships on fire off the shoulder of Orion. I watched C-beams glitter in the darkness at Tannhäuser Gate. All those moments will be lost in time, like tears in rain. Time to die …
I begin this book with the final scene of Blade Runner1 because Roy Batty’s words perfectly illustrate how memory relates to questions about who we are—about what it means to be human, and what makes up our identity. Roy Batty’s memories are indeed what distinguish him from other replicants. These memories are what make him feel like a person despite not being human, and justify his urge to cling to and prolong his short life. He may be an android, but Batty’s lament is one that feels familiar to all of us, when we wonder if all the memories that constitute our selves and feel so enduring are in fact ephemeral, and may be lost—like tears in the rain—when we die and our brain perishes.
The Encyclopaedia Britannica defines memory as “the encoding, storing, and retrieval in the human mind of past experiences.” As such definitions go, the one given by Britannica is somewhat narrow and provides but a minimal glimpse of the scope of the problem—on the other hand, these dry words are quite interesting because they raise a myriad of questions. For example, this definition refers to “the human mind.” Blade Runner is a science fiction classic itself based on another classic of the genre, Philip K. Dick’s Do Androids Dream of Electric Sheep? in which the hero at one point says, “The electric things have their lives too. Paltry as those lives are.” But do they? Aside fro
m the fictional Roy Batty, could androids one day have inner lives and memories, like we do? What about other animals, or a computer? Do memories make them conscious of their own existence? How can we know if they are? As we delve deeper into the Britannica definition, we can also ask: What is the mind? Is it simply a working brain? Is it merely the activity of billions of neurons, or something more than that? And, if the former is the case, how do these neurons store and retrieve so much information about our lives?
Memory plays a leading role as we pose ourselves these questions. Not only does it underlie our ability to think at all, it defines the content of our experiences and how we preserve them for years to come. Memory makes us who we are. If I were to lose my ability to hear and begin using a cochlear implant, I would no doubt continue to be the same person. If I were to suffer from heart failure and depend upon an artificial heart, I would be no less myself. If I lost an arm in an accident and had it replaced with a bionic limb, I would still be essentially me. To take this argument to its conclusion: as long as my mind and memories remain intact, I will continue to be the same person, no matter which part of my body (other than the brain) is replaced.2 On the other hand, when someone suffers from advanced Alzheimer’s disease and his memories are obliterated, people often say that he “is not himself anymore,” or that it is as if the person “is no longer there,” though his body remains unchanged. Thus we see the importance of memory to arguments about who we are, about what constitutes our being and distinguishes us from other animals, robots, or computers.
Science is born from questions. These questions are what nourish and inspire scientists and drive their obsessive quest for answers. And science is the quest itself, not just the end of it. For a scientist, arriving at a final answer is no more important than triggering the cascade of questions and experiencing the rush of fascination that propels the exploration of them. If reaching the final answer were all that mattered, science would be extremely frustrating, because the truth is that many questions will remain unanswered, perhaps forever. In the last few decades, neuroscience has advanced far more than in all the previous history of mankind, and yet many of the most profound queries, perhaps those that fascinate us the most, are still there, beckoning. To make matters even more interesting, these questions transcend the domain of science. As we try to understand how the activity of neurons encodes the remembrance of our experiences, we are inevitably led to ask about self-awareness, about the thing that makes us feel that we are a person. As we ponder the distinction between mind and matter, we find ourselves discussing topics considered by Plato, Aristotle, Descartes, and many others, topics that are endlessly revisited by the philosophers of the twenty-first century and that recur again and again in literature. These are the topics of artificial intelligence and neuroscience conferences but also science fiction movies—topics that touch everything from ethics and religion to education and our relationship with technology.
If I could choose only one of this book’s messages to impart—just one—it would be the vastness of the problem, the fascination born from exploring the workings of our memory and attempting to understand how our brain achieves such momentous feats as reconstructing details from the last scene of Blade Runner, the bars of a Beethoven symphony, or fleeting moments from our childhood.
Many think of the brain like a black box—a complex, mysterious organ that generates both mind and thoughts and is able to treasure memories that can be retrieved to consciousness on command. This is enough for some, but for others (neuroscientists among them) the mystery is not the end but the beginning. Like a child who listens to a radio and must remove the screws to see what is inside—and, once the radio’s innards are exposed, turn its dial and push its buttons to see what they do—the initial question gives rise to more questions, and to the inevitable realization that we still understand almost nothing.
When it comes to the brain, though, some things we do understand. So let us begin by discussing the basics—with neurons. In the same way that transistors are the basis of electronic circuits, neurons are the basis of brain function, arranged in groups, connected with one another in networks, producing with their activity our ability to see, listen, feel, and remember. But how do neurons generate the different functions of the brain? How does their activity result in our ability to write, run, or be aware of our existence? This is the question that, in its various nuances and facets, we neuroscientists ask ourselves every day, and despite the fact that we are so far unable to answer it fully, there are some elementary principles we’ve come to rely on that are relatively easy to grasp.
Figure 1.1: Network of neurons
Image adapted from an original drawing by Santiago Ramón y Cajal
Neurons have basically two states: they are either at rest or producing what we call action potentials, i.e., being active or “firing.”3 Just as a transistor transmits current to other parts of a circuit, neurons transmit their firing to other neurons (through the axons) and receive the firing of other neurons (through the dendrites). But the analogy with an electronic circuit ends there, because the contact between neurons is not electrical but chemical. When activated, neurons generate electrical discharges at the ends (or terminals) of their axons, which release chemical compounds called neurotransmitters. Through a process called synapsis, these neurotransmitters are received by receptors on the dendrites of other neurons, in turn generating small electrical discharges in them. The workings of many drugs rely upon this chemical interface: painkillers, tranquilizers, and hallucinogens do nothing more than alter the balance of neurotransmitters in the brain and the capacity of neurons to receive and transmit information. It is also key to understanding certain cognitive processes—for example, reward mechanisms that result from discharges of the neurotransmitter dopamine. More importantly for our purposes, neurotransmitters like glutamate play an important role in strengthening or weakening connections between neurons, which is precisely how memories are formed.
When does a neuron fire? When the activity it receives from other neurons exceeds a certain threshold. This mechanism gives rise to a variety of firing patterns determined, among other things, by the connections between neurons. For example, a given neuron, N, may fire due to the activity of the neurons that connect to it, and then transmit this activation to several other neurons. In turn, one of these latter neurons may transmit a discharge back to neuron N, prompting it to fire again. The variety of potential behaviors of a network of neurons is further enriched by the fact that activation patterns depend on the types of neurons involved. Excitatory neurons discharge neurotransmitters like dopamine and glutamate, which (in general) stimulate activity, while inhibitory neurons discharge neurotransmitters like gamma-aminobutyric acid (or GABA), which suppress it.
Figure 1.2: Synapsis
The electrical discharge of a neuron is transmitted (from the terminals of its axon to the terminals of the dendrites of the neurons that connect with it) by the release of neurotransmitters, in a process called synapsis.
Among neuroscientists, there are a surprising number of physicists (myself among them) who at some point in their scientific careers decided to take the plunge and dedicate themselves full-time instead to the study of the brain. Investigating the activity of neurons and neural networks—and how this activity gives rise to different firing patterns and replicates cerebral functions—is one of the favorite pursuits of many physicists-turned-neuroscientists, the practitioners of a discipline known as computational neuroscience. One of the field’s pioneers is John Hopfield, an American physicist at Princeton University, who described what we now call Hopfield networks.4 Basically, Hopfield networks provide a model for how the chaotic activity of a neural network can organize itself into stable configurations that represent different memories. Let us imagine a network of interconnected neurons, each either firing or silent. “Memory A” corresponds to a particular configuration of the network—for example, silent, firing, firing, silent, silent, . . . (or, in binary langu
age, 0, 1, 1, 0, 0, . . .); another memory, “Memory B,” corresponds to a different configuration—for instance, silent, silent, firing, firing, firing, . . . (0, 0, 1, 1, 1, . . .); and so on.
Figure 1.3: Neural representation of two different memories
Memory A corresponds to the neurons highlighted in gray and Memory B to those highlighted in black.
From a given initial state, the network converges to the closest memory. For example, the configuration 1, 1, 1, 0, 0, . . . is closer to Memory A than Memory B, and so the network evolves until it reaches the pattern of Memory A; on the other hand, the configuration 0, 0, 1, 1, 0, . . . resembles Memory B, and so the network converges to that result. The process by which the Hopfield network, starting from a given initial configuration, converges to that of the closest memory unfolds via methods imported from physics. Leaving details aside, as illustrated in Figure 1.4, the general idea is that from a network’s configuration, one can define a total network energy and create an energy landscape—each point in the landscape corresponds to a different configuration—assigning memories to energy minima determined by the connectivity patterns across the neurons. Then, starting from an initial configuration, the network evolves as a ball going downhill, progressively reducing its energy (changing the configuration at each step) until it reaches the minimum that corresponds to the closest memory. The initial configuration that provides the starting point for evolution on this landscape might be the result of spontaneous variations, as when we retrieve a memory seemingly from nothing, or of activations triggered by a particular stimulus—like, for example, looking at Rick Deckard’s face while watching Blade Runner. The image of Deckard activates a specific group of neurons, which in turn activate others, and on and on, until we retrieve a representation that resembles our memory of him. Since our vision of Deckard changes (we may see him straight on or in profile, shaved, wearing different clothes, etc.), the initial representation is not exactly the same as the one we store in memory, but, as long as it is similar, the network of neurons in our brain evolves until it arrives at a configuration that corresponds to our recollection of Rick Deckard, android hunter. Sometimes we may struggle to recognize an acquaintance, for example, because he’s changed his hairstyle, or shaved after long sporting a beard, or simply because many years have passed. This increased difficulty in recognizing someone results from the difference between the pattern of activation generated by seeing the person and the one we have used to “store” this person in our memory.