- Feature Stories
- News Flash
- Making Noise
- The Fence
- Beyond the Cut
- Inspire Innovation
The young boy gazes from his front yard to the end of the quiet cul-de-sac, two blocks from Carnegie Mellon. His parents moved to the street after his father, Joel Tarr, accepted a joint faculty position in the university’s history department in 1967. Each morning, five-year-old Michael is mesmerized by the parade of professors and students walking along Forbes Avenue to campus. Often among them is a serious-looking gentleman with thick, dark glasses. It’s Herb Simon, the father of artificial intelligence, who is making fundamental contributions to the field of cognitive psychology and would go on to receive a Nobel Prize in economics. Tarr’s father points out the influential scientist and one day introduces his son to the future Nobel laureate. On occasion, the boy and professor chat.
When the youngster reaches school age, he rides in a carpool with the children of neighbor Raj Reddy, a new faculty member at Carnegie Mellon who would become the founding director of The Robotics Institute and a dean of the School of Computer Science. Reddy would also later win a Turing Award, often dubbed the Nobel of computer science, for his work in artificial intelligence systems.
This is quite a neighborhood, and it’s one where Michael Tarr seems to fit in nicely. His father, now 77, amassed his own academic accomplishments as a world-renowned author and researcher regarding the environmental history of cities and technologies. Maybe that’s why he wasn’t impressed when his son, still in elementary school, could outperform him in mathematics. "But I knew it was all over when Michael began to beat me at ping-pong," the elder Tarr recalls, laughing.
Michael's mother, a developmental psychologist, passed away suddenly when he was six, leaving his father to raise him and his younger sister. With his dad working hard to secure tenure, the Carnegie Mellon campus becomes a second home to the young boy. He spends hour after hour playing on the Cut and often attends Tartan football games.
When he sprouts into a tall, lanky teenager, the terminal room in Science Hall (now Wean Hall) becomes one of his favorite hangouts. "I was bored with high school, plus they had better computers," he says. It's also where he first overhears discussion about advances in artificial intelligence. A science fiction devotee, he is captivated by the idea that human cognitive skills could be automated. During his senior year of high school, he takes an introductory class in cognitive psychology at Carnegie Mellon and another course in creativity in engineering. He also works as a computer programmer for several economists and psychologists.
The African proverb says it takes a village to raise a child, and for the younger Tarr, there's no doubt that village is Carnegie Mellon. "You really had a sense that you were in a university community, and Michael took advantage of it," his father says. There comes a time, though, when every child must leave the nest; Tarr's college years are spent at Cornell where he earns an undergraduate degree in psychology. Then he attends MIT graduate school in brain and cognitive sciences. His advisor there, Steven Pinker, is studying human visual cognition, or what the brain does to translate the raw image data collected by the eye into the objects and scenes we actually perceive. It seems to Tarr like interesting research; vision is our dominant sense, and roughly one-third of the cortex is used for visual processing—a vast amount of neural real estate.
He becomes absorbed in the question of how our visual system accomplishes the task of recognizing objects. Every day, we routinely identify hundreds of familiar and novel objects. It may seem simple to find a friend in a crowd or to identify your car in a lot, but how the brain performs this function is quite complicated—and largely unknown. "We don't have good answers for how the brain works in these cases," Tarr says. If we did, perhaps we could build better artificial vision systems or develop new therapies for people with dyslexia, autism, and other conditions that result in trouble recognizing visual patterns.
So, in graduate school, Tarr explores how people can visually recognize that 3D objects are the same even when seen under different viewing conditions. For example, you can identify a sailboat as a sailboat whether it's night or day, whether the vessel is moving toward you or tacking away. Prevailing wisdom held that we reconstruct a singular 3D model in our minds of an object and call up that same model every time we see the object under different viewing conditions. Tarr and Pinker test this notion through experiments that measure how long it takes people to recognize new and familiar shapes in various orientations. They are astonished by their results, finding that we actually encode a unique image of an object each time we see it. In other words, our representation of an object isn't a single 3D model, but rather a collection of such snapshots, in visual memory that we draw upon to recognize objects from different vantage points.
"Well-known people in the field were completely against the idea, and it was a very pitched battle for about 10 years," Tarr recounts. "I was having a hard time getting grants and publishing papers. But I think most people are convinced now that our approach was a different way to solve that problem." In 1998, the American Psychological Association called the discovery a "critical insight" while awarding Tarr a Distinguished Scientific Award for Early Career Contribution to Psychology. They cited his "ingenious and rigorous experimental tests" that yielded "surprising results that challenge long-standing assumptions." Today, discussion of those studies can be found in nearly any textbook on cognitive psychology.
Tarr's groundbreaking work earns him a PhD at age 27, along with an assistant professorship at Yale. His academic reputation prompts graduate student Isabel Gauthier to approach him about studying how the brain recognizes a different kind of object—the human face. Many scientists have long believed that our face-recognition system is functionally and anatomically separate from the part of the brain that recognizes other objects. It's a reasonable theory. Face recognition is one of the hardest visual discrimination tasks we perform, as we must detect slight differences in a large number of features across a wide range of facial expressions. Furthermore, our ability to recognize individuals on the basis of their faces alone is vital to us as social beings. That becomes devastatingly clear in people with autism, who have difficulty identifying faces, which experts say may contribute to the social impairment at the core of this disorder.(Continued …)