Steve Grossberg has a habit he calls “taking a think walk.” When a person spends his life applying the rigor of math to the mysteries of the mind, as Grossberg has, sometimes he gets stuck. On these occasions, with the answer to a problem just out of reach, Grossberg walks. As a younger man, he strode through Boston’s Back Bay, sometimes stopping at a cafe to write down new ideas; but mostly he walked, enjoying the sparkling city at night, coaxing along the insights that always, finally, came. “Many scientists see or feel things before they fully know them,” says Grossberg, Boston University’s Wang Professor of Cognitive and Neural Systems. “Eventually things become luminous, glistening; everything becomes clear.”
Grossberg’s insights, spanning decades, have changed the way we understand the brain. One of the principal founders and current leaders in the field of neural networks, Grossberg has tried to answer two major questions: how does the brain control behavior, and how can technology emulate biological intelligence? His mathematical models touch almost every area of psychology and neuroscience, including learning, memory, vision, development, speech, language, attention, cognition, navigation, and even consciousness. The applications of his work have been equally vast, in fields from image processing to pattern recognition, manufacturing to medicine.
Grossberg’s scientific quest is also a spiritual one, sparked by early tragedy and propelled by a pursuit of transcendence. When accepting the 2015 Lifetime Achievement Award from the Society of Experimental Psychologists, he spoke movingly of his need “to grow toward something more enduring.” What Grossberg wants, simply, is to understand how the lump of gray tissue we call our brain gives rise to the creative, soulful presence we call our mind. “For me, science is a spiritual calling,” he says, “not just a profession.”
“Nature speaks in math,” says Les Kaufman, a professor of biology in BU’s marine studies program and a longtime friend of Grossberg’s. “Mathematicians seem, to a lot of people, completely spaced out. But in reality they’re tapped into the heartbeat of the universe.”
So unexpectedly beautiful
Grossberg was born in 1939 and grew up in Jackson Heights, Queens. He describes his mother Elsie, a schoolteacher, as a remarkable woman. “She got her PhD equivalency at a time when it was hard for a woman to even go to college,” he recalled in an interview for Talking Nets: An Oral History of Neural Networks. “She very much influenced my religious attitude toward learning.” Grossberg’s father died of Hodgkin’s lymphoma when Steve was one year old. The early memory of his father’s death gave Grossberg a sense of urgency, a knowledge of life’s fragility, and a desire to pursue eternal truths. “Even though I could only be here for a very short time,” he says, “I wanted to somehow touch the enduring beauty of the world.”
As a boy, Grossberg excelled at art and music, able to play Gershwin’s Rhapsody in Blue after less than a year of studying piano. But he was most intrigued with people. “There are three things in New York: buildings, pets, and humans,” says Grossberg. “Of the three, the humans interested me most.” Although painfully shy—or perhaps because of it—he yearned to understand human behavior, and thought he might eventually become a psychiatrist. He attended Manhattan’s highly competitive Stuyvesant High School, graduating first in his class, and accepted an Alfred P. Sloan Scholarship to Dartmouth College in 1957.
Like many Dartmouth freshmen, Grossberg took a class in introductory psychology. There, he became intrigued by data about how humans learn and remember. In particular, he was entranced by how people learn lists, like the alphabet or a poem. For example, when people learn a short list, like AB, they also remember the reverse list, BA. And when learning a long list of items, like the entire alphabet, we tend to remember the beginning and the end best, while the middle gets muddled. Why? While most people greet this information with a shrug, the teenage Grossberg saw philosophical paradoxes, and his mind exploded with questions.
“There are three things in New York: buildings, pets, and humans,” says Grossberg. “Of the three, the humans interested me most.”
He went home for winter vacation his freshman year, pondering. Visiting with his childhood friend Dick Samuel, he described his ideas and ambitions. He told Samuel that he had decided on his life’s work: to write equations that would explain how a brain gives rise to a mind. Returning to Dartmouth, Grossberg turned to differential equations—a type of mathematics that describes how things change over time—hoping they would help him explain the dynamic human brain. Grossberg wanted to use them to understand profound questions. Like, how do brains adapt so quickly to rapidly changing events? How does short-term memory become long-term memory? At Dartmouth, he began taking long walks to explore his ideas; or, when it got too cold, sinking into a favorite soft chair at the library. In those quiet places, the insights snapped in his brain like firecrackers. The feeling was ecstatic and intense, but also frightening. “I was exploding with ideas, but as with any great passion, not totally in control,” says Grossberg. “There was no field for this, no one who could advise me. That made my life both exciting and dangerous. It was the creative process of a young person lurching around trying to find his way.”
Grossberg describes himself as a “pantheistic, Einsteinian Jew,” who hoped that his theoretical quest could bring him a little closer to God. “But I also worried when I was very young: what if I actually succeed? Will the mind lose its mystery? No. The reverence only grows, because how things work is so harmonious and beautiful, so unexpectedly beautiful.”
After months of struggle, Grossberg derived his first model of the brain as a network of interconnected neurons. He created a system of nonlinear differential equations that defined this “neural network” and mathematically explained properties of short-term, medium-term, and long-term memory at a time when these distinctions were still new. Most modelers in computational neuroscience still use these equations today.
It turned out, happily, that Grossberg’s equations could also explain and predict much of the psychological data that had puzzled and intrigued him about list learning. Over time, he developed a theoretical method embedding simpler models into increasingly realistic models—a kind of model “evolution”—leading to an ever more detailed and comprehensive understanding of the brain.
A latter-day Pythagoras
Fast-forward thirty years to Christmas Eve, 1986, near closing time at the Newton Highlands post office. The holiday rush was over, so Grossberg, his wife Gail Carpenter, and their five-year-old daughter had the place to themselves. Their mission: seal, stamp, and send nearly a thousand brochures to anyone who might want to attend the first international neural network conference, scheduled for the following summer in San Diego. The electrical engineering society IEEE was sponsoring the conference, with Grossberg as general chairman, and Grossberg and Carpenter were doing much of the planning, organizing, and advertising on a shoestring budget. “It was the first time anything like this had been done,” says Carpenter, a BU professor of mathematics, “and so we were trying to get it going. Steve and I stood in that post office licking these stamps for I don’t know how long. It was sort of crazy, but it was also fun. It’s the kind of thing we were willing to do. We both passionately believed in these things.”
Grossberg and Carpenter met at the Massachusetts Institute of Technology (MIT) in 1974. He calls her the love of his life and his closest collaborator, and their friend Kaufman describes the couple as both “joyous” and “intellectually ferocious.” “Anything they do, they really do to perfection,” he says. After graduating from Dartmouth as its first joint major in mathematics and psychology, Grossberg earned his PhD from Rockefeller University by proving novel theorems about his learning models, which landed him a job at MIT. Carpenter joined the MIT faculty several years later, after writing a brilliant PhD thesis on the Hodgkin-Huxley equations, which describe how a nerve impulse travels down an axon. The two scientists, standouts in the small community interested in math and mind, became friends, then lovers, and married—in the BU Castle—five years later. They became colleagues as well, with their first joint paper published in 1981. “Meeting Gail changed my life completely,” says Grossberg in Talking Nets. “From feeling isolated and not having someone to talk to, I suddenly had the best person in the world to talk to.”
Grossberg was productive at MIT, but his work included new intuitions bridging psychology and neuroscience, and new mathematical models to make these intuitions precise. Many scientists found his work and its mathematical language impenetrable, and he had trouble publishing papers in psychology or neuroscience journals. Daniel Levine, a professor of psychology at the University of Texas at Arlington who was Grossberg’s student at MIT, recalls his mentor’s struggle: “He wanted to build bridges that people on either side didn’t want built.”
Fortunately for Grossberg, then-BU president John Silber was hunting for talent. He saw that Grossberg’s game-changing ideas came with strong recommendations from some of the world’s top scientists and mathematicians. “I was fascinated by his interests,” recalled Silber at a conference celebrating Grossberg’s 65th birthday. The late Silber, a former philosophy professor, saw the young mathematician as a “latter-day Pythagoras,” referring to the Greek philosopher-scientist. “He believes that all things are numbers,” said Silber of Grossberg. “That is, that you can offer a scientific account of things as ineffable as the human mind.”
Grossberg became a full professor at Boston University in 1975. “The appointment of Stephen Grossberg was actually a transforming event in the history of Boston University,” said Silber in the birthday speech. And it was. Having spent his early career straddling math, neuroscience, and psychology, and feeling misunderstood and academically homeless as a result, Grossberg resolved to build a neural network community so that others wouldn’t struggle as he had. With Carpenter, he founded BU’s department of cognitive and neural systems (CNS), the Center for Adaptive Systems, the International Neural Network Society, and the journal Neural Networks. They and their CNS colleagues won upwards of $85 million in grants from many sources, including the National Science Foundation, the Alfred P. Sloan Foundation, the Office of Naval Research, the Air Force Office of Scientific Research, and the Defense Advanced Research Projects Agency. BU’s pioneering focus on the computational modeling of mind and brain attracted gifted students and faculty from around the world. Grossberg says he personally trained more than a hundred PhD students and postdocs, including many of BU’s current leaders in computational neuroscience.
“They’re builders,” says Jacqueline Liederman, a professor of brain, behavior, and cognition in BU’s psychological and brain sciences department, speaking of her longtime friends Grossberg and Carpenter. “They build paths. They build gardens. They’re constantly having a project together. They’re just constantly conquering something.”
By the 1980s, interest in neural networks was surging, propelled by engineers using neural models for applications like image processing, pattern recognition, and robotics. “When Gail or I lectured in the ’80s, technologists would run up to the podium asking for the code,” recalls Grossberg. One early adopter was Boeing, which created an automated inventory system to search their massive stock of airplane parts quickly and efficiently. The 1987 neural network conference—the result of Grossberg and Carpenter’s stamp-licking adventure at the Newton Highlands post office—was a triumph, attracting around 2,000 scientists and engineers from around the world and establishing that the field had arrived.
“Everybody who had any interest in the brain sciences and artificial intelligence was there—engineers, mathematicians, psychologists,” says Sucharita Gopal, a BU professor of earth and environment, who attended the conference as a graduate student. “It was magical, because things like that happen once in your lifetime. It was the birth of that field.”
It was at BU that Grossberg introduced—and with Carpenter, developed—probably his widest-reaching contribution to psychology and neuroscience: Adaptive Resonance Theory, or ART. ART tackles a thorny question: how can people quickly learn new facts throughout life without just as quickly forgetting old ones? Formally this problem is called the stability-plasticity dilemma: how can the brain be plastic enough to learn new things, but also stable enough to remember useful old ones? Our brains have the remarkable ability to sort through a massive amount of rapidly changing data, take what we need, match it with stored knowledge, use it immediately, and sometimes remember unique experiences for a lifetime. For decades, scientists and engineers have been trying to design machines that do the same.
Grossberg’s major insight was that networks of neurons within the human brain are able to bridge knowledge and emotion through interactions he calls resonances. These resonances trigger fast learning and rapid adaptations—hence adaptive resonance—and allow people to focus attention and learn about important events. As an example, say you open the refrigerator on the day after Thanksgiving, hoping to make a turkey sandwich. You already know what turkey leftovers look like, so you scan the fridge trying to match the expected picture in your mind. Seeing the Tupperware of turkey—a match!—triggers resonating signals between the brain cells that make the observation and those that carry the expectation of seeing turkey. These resonating signals support your conscious visual awareness of the turkey and an emotional feeling of satisfaction. If, instead, you scan the fridge and find no match, you feel a wave of frustration. The frustration is part of a cognitive and emotional “reset” that enables you to search elsewhere for lunch, rather than staring helplessly at the empty fridge for hours.
This kind of experience illustrates the ART model of human learning. ART explains how humans cope with a huge amount of information—by using expectations and attention to identify, recognize, and predict important objects and events, and to combine this knowledge with emotions that help us make and execute an action plan. In this case, ordering a pizza for lunch.
“Today when I come to a new problem, I come in like a mature artist,” says Grossberg. “I’ve practiced my scales, I have my repertoire, so I’m ready to play new music and do my best to reveal its beauty.”
Grossberg first introduced these ideas in 1976, and Carpenter played a critical role in transforming them into mathematical algorithms that were eagerly snapped up by engineers and technologists. Since the 1980s, Carpenter and Grossberg have developed and refined these algorithms, and the reach of the ART models has been profound. “ART led to an explosion of results by literally thousands of practitioners, with many more to come,” says Donald Wunsch, a leading neural modeler and professor of electrical and computer engineering at Missouri University of Science & Technology. “The field of applying ART to interesting problems in brain research and engineering has not yet played itself out.”
Engineers have used ART to help four-legged robots navigate—a useful technology for bomb detection—and in experimental software to help elderly drivers detect nearby cars. ART-based software helps doctors classify cardiac arrhythmias, diagnose ovarian cancer, and detect sleep apnea. “We won’t see computers that are smarter than people,” says Wunsch. “But we’re going to see more and more things done by neural networks that used to be done by people.” ART is also helping to clarify one of the greatest mysteries in science: our own consciousness. Grossberg predicts that “all conscious states are resonant states” and has classified the brain resonances that are activated when we see, hear, know, and feel.
Grossberg, now 75, is gratified to see the long reach of his work and continues active programs of research. His creative process is different now than when he was a shy teenager at Dartmouth, but the quest remains the same: to catch a glimpse of the truth, to touch, albeit briefly, the eternal. “Today when I come to a new problem, I come in like a mature artist. I’ve practiced my scales, I have my repertoire, so I’m ready to play new music and do my best to reveal its beauty,” he says.