How did we get so smart?
Stephen Grossberg and an interdisciplinary BU team unlock the secrets of how we learn
Statues and masks from Africa and New Guinea perch sentinel-like around the office of Stephen Grossberg, chairman of the Department of Cognitive and Neural Systems (CNS) at the Graduate School of Arts and Sciences. Their carved abstract faces watch over a man who has devoted his life to unraveling the mysteries of the brain, to understanding how emotion and cognition interact to guide our judgments and form our identities.
Grossberg is one of the founders of the modern scientific movement to develop a model for how the brain works—how it sees, learns, recognizes, and keeps track of objects, memories, and complex situations. And now, supported by a $20.1 million grant from the National Science Foundation, he’ll lead a multidisciplinary, collaborative effort at Boston University and beyond to take that movement to the next level: to develop a new science of learning.
Grossberg heads the newly established Center of Excellence for Learning in Education, Science, and Technology (CELEST) at BU, which has a charter to study and model the behavioral and brain processes involved in learning, to develop new learning algorithms, and from there to develop new technologies and machine-based intelligences. Researchers will also create new science and mathematics curricula and other educational materials about how the brain learns, intending to translate their results into opportunities for improvements in teaching. Their agenda is ambitious.
CELESTial Dream Team
Sixty teams of scientists from leading universities were invited to compete last year for the funding to establish “science of learning centers,” but the NSF selected only BU and three other teams, lending important recognition to a program that has become a national leader in the cognitive sciences. Grossberg’s team includes “arguably the best group of brain and behavior modelers in the world,” he says, working in cooperation with “first-class experimentalists. Few other modeling groups link complex behaviors to their underlying brain mechanisms, or use those insights to develop intelligent, large-scale applications in learning and technology.”
The interdisciplinary CELEST group draws faculty from fields across BU’s cognitive science spectrum, as well as from institutions outside of BU. Grossberg, the center’s director, holds appointments in the Departments of Psychology, Mathematics, and Biomedical Engineering (as well as in CNS, where he holds the Wang Professorship). CELEST’s BU faculty also includes Daniel Bullock, Gail Carpenter, Robert Devaney, Howard Eichenbaum, Frank Guenther, Michael Hasselmo, Kathleen Kantak, Jacqueline Liederman, Ennio Mingolla, Barbara Shinn-Cunningham, Eugene Stanley, Chantal Stern, and Takeo Watanabe, who collectively represent five BU departments (CNS, Psychology, Physics, Mathematics, and Biomedical Engineering) and five BU research centers. The team is rounded out by cognitive neuroscientist Robert Sekuler of Brandeis University, neuroscientist Earl Miller of the Massachusetts Institute of Technology, psychologist Michael Kahana of the University of Pennsylvania, and researchers from Massachusetts General Hospital.
Our Minds, Our Selves
How do we learn? It seems purely intuitive, beyond explanation. And the answer is complicated, based on genetic factors, the brain’s own structure, and our social and physical environments. The processes of learning are just beginning to be understood, according to the National Science Foundation, and Joseph Bordogna, NSF deputy director, says, “This is the right time to make new investments in the science of learning, when scientists are revealing innovative ways to integrate research across many disciplines—biological, cognitive, computational, mathematical, neuro, physical, and social sciences; engineering; and education.” The agency hopes that a deeper understanding of the basic processes of learning will help scientists and educators devise methods of improving how people learn and develop machines that can work intelligently and independently.
 But the idea that the mind, the repository for who we are, can be modeled as a series of processes is an initially uncomfortable one. Our minds are the seat of our selfhood. Don’t our behaviors and feelings actually spring from something beyond physicality? Gail Carpenter, professor of cognitive and neural systems, knows that contemplating the brain as a physical object can make people uneasy. “But the more we know of how the brain functions, the more awe-inspiring it is,” she says. “Studying the details only increases the awe and spirituality about it.”
By understanding how brain mechanisms give rise to functions like vision, speech, and memory, CELEST researchers will establish a critical link between brain, behavior, and learning. Carpenter uses an analogy to explain how mechanism and function relate. If you open up the back of a television set without knowing what a television does—that it receives signals and transforms them into pictures—the electronic design will make no sense. But “if you do have the idea of what a television does,” she says, “then the components are working in the service of the function.”
Your brain, too, takes signals and turns them into pictures, but if you don’t know that vision takes place in the brain, you can’t deduce that just by examining the assembly of neurons. “When functions such as vision get adapted into different learning laws,” Carpenter says, “these connections and electrical impulses suddenly take on meaning within the context of the whole.”
Learning about Learning
Appropriately, for a program dedicated to the science of learning, CELEST has some potent educational goals that correlate to its research. One is to encourage the modeling of brain functions in the context of science and mathematics lessons. Another is to study the brain’s learning mechanisms to come up with better pedagogical methods, emphasizing outreach to diverse populations. And a related aim is to promote brain study in high schools and colleges, building on the curricular materials that Professors Eugene Stanley and Bob Devaney have already developed for Boston-area science and mathematics teachers. The educational thrust is meant “to translate ideas of current research into the high school curriculum,” says Stanley. “Every graduate student and professor associated with this project is encouraged to figure out how their current research can be used in a high school setting. We could turn on more high school children to the joy of doing science.”
CELEST will also create Web-based and hands-on training tools and classroom activities for teachers and students. One such educational project is an online game that uses machine learning algorithms. “Even young kids can play it and get intuitions about advanced mathematics,” Carpenter says.
At a higher level, the researchers intend to make CELEST a virtual conference room for interdisciplinary collaboration and training in the physiological, psychological, and pedagogical aspects of learning. The center will sponsor frequent seminars, workshops, colloquia, and publications. The impact of its inquiries is expected to be broad, touching on the fields of computer science, engineering, mathematics, technology development, psychiatry and medicine, and, of course, education.
Pioneering the Brain-Mind Connection
Stephen Grossberg’s interest in the brain started in 1957, when he was seventeen and a freshman psychology student at Dartmouth College. He saw the need for a quantitative basis for expressing psychological insights, so he took up mathematics. The emerging field of computing was not advanced enough to do the calculations needed.
He was inspired by his mathematics professor, John Kemeny, who co-wrote the influential programming language BASIC. (Kemeny was also Albert Einstein’s last mathematical assistant and later became president of Dartmouth.) Grossberg did the first joint major in mathematics and social sciences at Dartmouth in order to develop the math tools to understand how “moment by moment, individuals can adjust to a changing world and learn on their own.”
He did his graduate work at Stanford University and the Rockefeller University, and over the years has theorized and developed many of the fundamental principles that explain how our minds work. He helped to found the fields of computational neuroscience and theoretical cognitive science, disciplines that make use of mathematical models to describe brain functions. With Carpenter, he also pioneered the field of neuromorphic technology, which involves building biologically based computer intelligence.
The scientific revolution that Grossberg helped start—the quest to understand how the mind and the brain work—is challenging, both because of the complexity of the subject matter and because “new intuitions and new mathematics” need to be introduced to make progress. “My own early work was devoted to discovering these intuitions and developing the mathematical framework through which they could be expressed,” he says. That work established a new paradigm in science—“to introduce mathematical laws describing the brain as a highly non-linear system (the whole is greater than the sum of its parts), a non-stationary system (development and learning go on throughout life), and a non-local feedback system (widespread interactions among brain cells provide contextual meaning and stability) to clarify how the brain successfully adapts to a changing world.”
Grossberg developed his “adaptive resonance” theory to propose an explanation of how we learn quickly without forgetting what we already know. The ability is linked to our ability to expect events and pay attention to those that interest us. “Expectations are often automatically activated,” he says. “It is not always a question of will. You are automatically focusing attention during every wakeful moment. You are doing selective processing. Expectation and attention help us learn quickly without catastrophic forgetting.”
A computer memory that learns accumulatively can be overloaded and crash. But our brains do not suddenly lose old information in this way, as we sift what is important to retain based on prior experience. The brain is constantly making judgments based on the sum total of past experiences, and it weights decisions emotionally. “Cognition and emotion interact intimately,” says Grossberg.
Smarter Computers
When it comes to applying CELEST’s findings to technological innovations, Carpenter says that the goal is “emulating the functions that humans do well that computers don’t.” The researchers hope eventually to develop next-generation artificial intelligence based on the way our brains process information and adapt to change.
Classical artificial intelligence was often based on mathematical logic. It did not incorporate emotions. Our human brains, however, rely constantly on emotions in making judgments. Grossberg envisions a new generation of machine-based intelligence that will simulate emotions as well as perceptual and cognitive functions. Autonomous robots and systems built with this new generation of AI algorithms could better cope with change, just as human brains do. A goal-oriented algorithm in computer software, for instance, could successfully sift through medical records that include contradictory and incomplete information and draw conclusions that take those contradictions into account.
 The research driving these hopes is as interdisciplinary as the CELEST team itself. Professor of Psychology Michael Hasselmo, for instance, is looking for links between physiology and behavior by creating a “virtual rat,” modeling data that researchers have collected over decades of experiments on rat brains. “We have thousands of publications on neuroscience data but few neuroscientists have consolidated the information,” says Hasselmo. His virtual rat project can be described as a sort of rarefied computer game with a rat protagonist.
Hasselmo’s group also explores the brain’s ability to guide behavior with episodic memory, which is memory acquired and contingent upon a particular place in time, such as remembering a weekend spent in Paris. An analogous experiment by Professor Howard Eichenbaum, director of BU’s Center for Memory and Brain, compares the model to the behavior of a live rat. His group tests episodic memory by measuring how a rat encounters different odors in a sequence of cups of sand.
Such studies allow researchers to “understand the data we get on the structural components of cells,” Hasselmo says. “You can’t understand how a car drives on the highway by looking at the chemical composition of gasoline. By understanding the functional mechanics of the engine, you know how it powers the car.”
— Wendy Wolfson
The Boston University Department of Cognitive and Neural Systems will hold an anniversary conference celebrating its fifteenth year and Stephen Grossberg's sixty-fifth birthday on September 16-17, 2005. Please visit http://cns.bu.edu/events for information. And, for additional information on CELEST, see http://cns.bu.edu/celest.
This article was originally published in the Spring 2005 edition of Arts & Sciences
Top photo : Kalman Zabarsky, Boston University Photo Services
Other Photos: Boston University Photo Services
|
