Skip To Content Skip To Left Navigation
NSF Logo Search GraphicGuide To Programs GraphicImage Library GraphicSite Map GraphicHelp GraphicPrivacy Policy Graphic
OLPA Header Graphic
 
     
 

Dr. Colwell's Remarks

 


"The Emerging Science of Learning"

Dr. Rita R. Colwell
Director
National Science Foundation
Institute for Human and Machine Cognition
University of West Florida
Pensacola, Florida

January 21, 2004

See also slide presentation.

If you're interested in reproducing any of the slides, please contact
The Office of Legislative and Public Affairs: (703) 292-8070.

Good evening, everyone. Thank you, Ken,1 for a wonderful introduction. I have been looking forward with great anticipation to my visit to the Institute of Human and Machine Cognition, and my visit today met—indeed, exceeded—my expectations.

[title slide on]
(Use "back" to return to the text.)

Tonight I plan to talk about "The Emerging Science of Learning." I'm especially pleased to discuss this topic because my visit today is to an institute breaking significant new ground in learning science. In many ways, the developing science of learning represents the broader trends of convergence in science and engineering—the intersection of research and education, and the explosion of interdisciplinary research, being two such trends. I would like to speak for a moment about the convergence of disciplines, and then consider what is the science of learning, what it could be—and why we need it. Then I will survey, very briefly, learning science across the scales, from the level of cell to classroom and beyond.

[South Pole aurora]
(Use "back" to return to the text.)

Diverse streams of discovery merge at the science of learning frontier. This wintertime image of the aurora australis, captured at South Pole Station, represents the National Science Foundation's willingness to go to the ends of the earth, when necessary, to invest in the frontiers of learning and discovery. Much of the excitement of discovery today in science and engineering ignites at the interfaces of disciplines.

[stylized fMRI brain]
(Use "back" to return to the text.)

A legitimate question might be why we need a science, the focus of which is of learning. I'm reminded of an observation by a physicist, Joe Redish of the University of Maryland, who recounted a conversation he overheard between two students in the hallway—one a senior and one a freshman. The older student commented, "Redish makes you think—that's his goal"...to which the younger student replied, "It's fine if I have to think, as long as I still get an "A"!

As Redish himself then observed, "Education is harder than physics!"

[slide not available]
(Use "back" to return to the text.)

"Imagine if we taught baseball the way we teach science." Alison Gopnik, writing in the New York Review of Books, makes that analogy, with humor but also with a grain of truth. Young children, she says, would read about baseball techniques, answer quizzes about baseball rules, and occasionally hear stories about baseball greats. Some coaches would argue that students should drill in fundamental baseball skills. "Undergraduates might be allowed," says Gopnik, "to reproduce famous historic baseball plays. But only in graduate school would they, at last, actually get to play a game."

[slide not available]
(Use "back" to return to the text.)

A century ago, school administrators organized classrooms along the model of mass production factories. The products—educated children—were produced through a top-down transfer of knowledge.2 Well, the world has changed, but many classrooms retain this antiquated structure.

[slide not available]
(Use "back" to return to the text.)

James Paul Gee, a University of Wisconsin professor of reading who watched his four-year-old learn, and enjoy, long and challenging video games, believes that such games embody some interesting principles of learning.3 Inspired to attempt a popular video game himself, he found it extraordinarily difficult.

"How in heaven's name do they sell so many of these games when they are so long and hard?" he asked himself, noting that subsequent games just get longer and more challenging. His conclusion: "...the theory of learning in good video games fits better with the modern high-tech global world that today's children and teenagers live in than do the theories (and practices) of learning...they see in school."

These are provocative words. Even while they do not take note of changing approaches in science and math education, they do underscore the need to develop the emerging science of learning—to develop a scientific basis for an educational system suited to the 21st century.

[Visual: quotation—"Unknown to many university faculty...is a large body of recent research from educators and cognitive scientists on how people learn."]
(Use "back" to return to the text.)

"Unknown to many university faculty in the natural sciences, particularly at large research institutions, is a large body of recent research from educators and cognitive scientists on how people learn." So wrote William Wood and James Gentile last month in Science Magazine.4 They noted that traditional lecture and laboratory courses are "relatively ineffective strategies" to impart concepts to undergraduates. Research suggests, said the authors, that "active, inquiry-based, and collaborative learning, assisted by information technology," trumps older teaching methods. Yet, the authors note, many researchers do not read the journals that report such results.

[Science of Learning]
(Use "back" to return to the text.)

The science of learning is a broad canopy arching over diverse scientific streams. As noted in the National Academy of Sciences' study, How People Learn, "What is new...for a new science of learning is the convergence of evidence from a number of scientific fields." The social sciences investigate the nature of perception and memory, and the role of motivation and emotion in learning. Biosciences cover the gamut from molecular to behavioral foundations of learning. Cognitive neuroscience brings us insight into the neural basis of learning in humans and other species. The physical and information sciences and engineering are now creating machines that learn. Educational sciences cover pedagogy from schools to colleges to lifelong learning.

The convergence of insights from all these perspectives brings sharper resolution to our picture of intellectual development. Such research builds the basis for the classroom of the future, even the foundation for learning beyond the classroom, and for educating our future workforce.

However, we must encourage the two-way flow of discovery between researchers and educators. An insight gleaned in the cognitive neuroscience laboratory, for example, may or may not work in a K-12 classroom of 30 kids from multi-ethnic backgrounds.

The question of assessment—how do you test what students know?—is another critical area for progress. NSF sociologist Steven Breckler, who co-heads our Science of Learning Centers program, observes that the science of testing assessment has been "dormant" since not too long after World War II. Yet, fields such as statistics and psychology, with their rich traditions in the basic science of assessment, can help provide modern tools to assess learning.

[genes to learning and behavior]
(Use "back" to return to the text.)

Here's another argument for a science of learning—showing that components of learning operate at every scale from genetic to digital to societal. To truly investigate learning, we must integrate insights from every level through collaborations between biologists and engineers, or between psychologists and computer scientists. While those in closely related fields under the learning-science umbrella may collaborate now, those in more distant fields may not; a practitioner of cell biology will not talk often with, or use the same research jargon as, a researcher who studies how gender, race, or culture shape learning.

[National Science Foundation's Centers for the Science of Learning]
(Use "back" to return to the text.)

The National Science Foundation is now setting up Centers for the Science of Learning. One hallmark will be a given center's commitment to forging linkages across the scales of learning science. Our first competition brought in 48 proposals, and site visits have just gotten underway to 12 proposed centers. The top ones will be selected in March, and final evaluation by the National Science Board is expected later in the year.

[slide not available]
(Use "back" to return to the text.)

NSF has already made what we call catalyst or planning grants to a number of institutions to explore various models for learning science centers. I'll cite three examples that collectively suggest the breadth of learning science. My first example of a catalyst project focuses on human vision, perceptual learning, and brain plasticity. It is headed by Daniel Kersten at the University of Minnesota and co-principal investigators. Human vision is highly complex, with ten million retinal measurements sent to the brain each second, where some billion cortical neurons do the processing. As the team explains, a sea change has swept away the prevailing view of human vision as frozen in structure after a brief critical period. We are now beginning to regard vision as modifiable throughout the human lifespan. For example, the ability to process faces—such as identity recognition and emotions displayed—only matures in early adolescence.

In this image, we see a number of visual areas that have been mapped with functional Magnetic Resonance Imaging (fMRI). The question is how the functional organization of these areas changes with diverse visual experience.

[slide not available]
(Use "back" to return to the text.)

Functional MRI also helped to pinpoint these visual cortical areas in the brain. Different areas are more active with different kinds of visual stimuli—for example, V1 for edges and LOC for whole objects.

How does the brain's topographic organization and preference for different sensory stimuli change with experience? An example: "V1" can change to respond to tactile stimuli. Imaging can be used to map changes in perceptual function, documenting the brain's plasticity.

In any case, the center proposes to gather vision researchers, educators and engineers to ultimately improve the quality of life for the more than 3 million Americans who are visually impaired, not to mention 180 million similar people worldwide.

[Hari Narayanan: simple aquarium]
(Use "back" to return to the text.)

Another catalyst effort is attempting to unravel complex causal learning. Such learning is critical across science and engineering, notes the team lead by Hari Narayanan of Auburn University and colleagues. Yet the workings of complex systems may seem opaque because of the abstract thought required, which may challenge prevailing beliefs. Here we see an illustration: a concept map of an aquarium drawn by a child. It is quite simple.

[Hari Naryanan: complex aquarium]
(Use "back" to return to the text.)

Here is an expert's map of the interactions in an aquarium—actually quite complex. Among unanswered questions about how to promote such knowledge are: What is causal learning, and how do scientists and engineers understand and model causal systems? Also, how can we foster such learning from kindergarten on?

[slide not available]
(Use "back" to return to the text.)

Here is another catalyst grant, but this one takes a cultural approach. Carol Lee of Northwestern University co-heads an exploration of the diverse cultural pathways to learning by different human groups. According to Lee and her colleagues, core learning and development fields have not, in the main, embraced cultural diversity; yet, by the year 2015, more than 40% of schoolchildren in 15 U.S. states will come from racial and ethnic minorities. Needed is an approach to learning that understands an individual within "nested social contexts" from family to classroom to neighborhood. Such studies help nourish the new field of cultural biology, which asks how the brain and its social environment interact.

[NSF Science of Learning Centers: repeat slide]
(Use "back" to return to the text.)

From brain plasticity to complex causal learning to cultural contexts is a wide span, yet this is just a sampling of the science of learning. I will turn now to a survey across the disciplines and scales of learning research, from cellular to social to digital. Overlaps between these disciplines make them more than the sum of their individual parts, producing knowledge not attainable with the resources and outlook of just one discipline. As NSF cognitive neuroscientist Greg Solomon points out, over the history of science, when the harder and softer sciences have come together, the "big" questions may often have been generated by the "softer" science.

Let's begin this journey at the cellular level, where recent research has begun to identify some biochemical signaling pathways used in learning. Microarray technology—looking at the action of many genes at once—has also opened the way to identifying gene expression related to learning.

[John Anderson: computer algebra tutor screen capture]
(Use "back" to return to the text.)

At the level of neuroscience, we know that "in the past twenty years scientists have learned more about the brain than they have in the previous two hundred," as author and neuroscientist Richard Restak puts it. He also notes that "over 90 percent of all the...scientists [studying the brain], who have ever lived, are still living today."

The application of brain-imaging techniques—such as functional MRI—to learning has really just begun. A promising example is the work of Carnegie Mellon University's John Anderson and his colleagues. They have applied brain imaging to improve the assessment of a computer algebra tutor used in middle and high schools. Here we see the algebra tutor interface—a complex screen with multiple regions. Critical aspects of mathematics learning have brain correlates that can be identified with functional MRI.

[John Anderson: three brain regions tracked]
(Use "back" to return to the text.)

The group used fMRI to identify three key brain regions shown here, which were discovered in middle-school-age students learning algebra. Combined with other techniques such as eye-tracking, the researchers believe they have reached a new threshold of "mind reading"—and, importantly, have begun to show how instruction can be improved by sensitivity to individual students' uniqueness.

[fMRI brains with dyslexia]
(Use "back" to return to the text.)

fMRI has also revealed new insights about dyslexia, the common reading disability that has afflicted many talented people. This powerful imaging technique—which arose from obscure research on the energy state of an atom's nucleus, not from cognitive science--has now revealed that regions of the brain operate differently in dyslexics, as seen in these images by Guinevere Eden of Georgetown University. These insights have spawned new educational strategies to teach people with dyslexia to map symbols to corresponding sounds. Intervention early in children's lives seems promising.

[word slide: bullets/multiple intelligences/importance of individuality]
(Use "back" to return to the text.)

Moving to another level in the science of learning, new theories take a broader look at what we term "intelligence," thereby opening the door to consider the unique learning capabilities and challenges of each student. Howard Gardner's theory posits eight intelligences: linguistic, logical, musical, spatial, bodily kinesthetic, interpersonal, intrapersonal, and naturalistic. Schools mostly value only two: linguistic and logical.

Appreciation of the diversity of our mind's capabilities leads naturally to the idea that there are "all kinds of minds," in the catchphrase of University of North Carolina researcher, Mel Levine. He discards labels such as attention deficit disorder and focuses instead on each student's own "breakdown points" in learning. Interventions can include improving such abilities as a child's time management.

[slide not available]
(Use "back" to return to the text.)

One of the multiple intelligences, spatial learning, is little understood. I would like to focus on it a bit because it underlies much of science, from visualizing molecules to the structure of the earth to a surgical procedure to a rocket engine. Much research has been done about how we learn to read, yet we do not know how people acquire—or how to teach--these important spatial skills.

David Holtzman at the State University of New York is studying the brains of snakes for clues to spatial learning. Adult snakes experience ongoing neurogenesis faster and throughout more of the brain than do mammals, so activities affecting neurogenesis can be identified more readily in snakes. This snake has been trained to find an escape hole by following either visual cues or a mouse odor. Both of these activities are examples of spatial learning. The inset picture here, showing a snake's brain, illustrates cells produced after a spatial learning experience.

A broader question is whether the environment—especially cognitive activities in the environment--can increase the number of neurons in a snake's brain, or even whether neurogenesis could be sped up in the brains of mammals, including humans. This research begins to lay the groundwork to explore such questions.

[slide not available]
(Use "back" to return to the text.)

How spatial skills develop in the child's brain is the focus of research by Nora Newcombe at Temple University. Her data seem to indicate a "fundamental transformation of spatial thought between 18 and 24 to 30 months," she says. 18-month-olds, for example, can find an object in a sandbox but can't remember its location very well after being distracted. Insights about why a child finds a peek-a-boo game such fun actually have relevance to robotics—as in teaching a robot to navigate a new environment such as a collapsed building, or perhaps rough terrain on another planet.

[slide not available]
(Use "back" to return to the text.)

One of the great insights of learning science is the discovery that where people start from—the knowledge they have and assumptions they hold—has a huge impact on how well they learn. "All learning involves transfer from previous experiences," stresses the NAS study How People Learn. Students' understanding of new material must begin with an understanding of the point of view they already hold. The children's book shown here—Fish is Fish—whimsically depicts this truth. It tells the story of a fish who learns about life on dry land from a tadpole-turned-frog. While the frog describes birds, the fish envisions fish with wings; when cows are described, the fish imagines fish with udders. This perspective on the need to know "where learners are coming from" really applies to all ages.

[three girls with math blackboard in background]
(Use "back" to return to the text.)

Moving up to the social level, it has been shown that stereotypes matter in learning, for poor expectations can lower test performance. For example, if women and blacks are subtly reminded of gender or racial differences in performance, they may score lower on tests.

Caroline Dweck and Catherine Good at Columbia University lead a team that studied calculus students over time. It is well known that students who view intelligence as a fixed trait may do less well when faced with difficulties, compared with students who see intellectual ability as a potential to develop. An academic atmosphere that portrays intelligence as a fixed trait may reinforce students' negative beliefs when they are already prone to believing race or gender stereotypes.

[Dweck graph: "sense of belonging"]
(Use "back" to return to the text.)

The project's results show that females who view math intelligence as a fixed trait—"There's really not much I can do to change how well I can learn calculus"—do not develop as strong a sense of belonging to their calculus class. As you can see from the graph, students who perceive their intelligence as fixed have a lower sense of belonging in their calculus class. When a negative stereotype also came into play, their sense of belonging became even more fragile.

[virtual handshake]
(Use "back" to return to the text.)

I'll move now to my final level of learning, touching upon various ways that information technology and artificial intelligence are really just beginning to help us to explore and enhance human learning.

The science of learning certainly feeds into the concept of human-centered computing—that computer systems should dovetail with and expand human capabilities instead of forcing humans to adapt to machines.

The virtual handshake depicted here is still a fantasy, yet being able to touch and smell may be important to virtual learning and other tasks. Medical instructors say that surgeons need to be able to smell—part of perceiving when they have nicked a vessel, for example.

[comparison of old and new airplane cockpits]
(Use "back" to return to the text.)

Here is just one example of the mind-machine connection from here at the Institute of Human and Machine Cognition. Pilots, especially in combat aircraft, must process complex information and make rapid-fire decisions, yet the traditional cockpit display at left is a cumbersome way to convey information fast. This type of format, bristling with dials, was actually inherited from the steam engine, because the Wright brothers and their contemporaries had no other models available. The contrasting, more intuitive display at right engages a pilot's central and peripheral vision in a comprehensive way, enabling a pilot to perceive a plane's status in a fraction of a second, instead of taking several seconds as required with the old display.

With the new field called machine learning, it's almost like being able to explore learning in another dimension. Computers can process huge databases much better than we can, and give us insight into our own learning in the process. Machine learning has been applied to counting insect populations, such as the stoneflies we see here. The goal of a project at Oregon State University and the University of Washington is to build a device that rapidly and automatically identifies insects to family, genus or species. Machines can recognize patterns in huge databases much better than we can. Fast and cheap counts of insect populations would be an important tool for the environmental sciences.

Information technology also helps us tap into a deep, natural motivation for learning—that is, to be able to teach others, a motivation for all ages. John Bransford of the University of Washington and others have developed the concept of "teachable agents" (really virtual students, listed here by name), who real students can teach. These are "social agents" who need explicit instruction to succeed.

If students teach agents—such as Betty here—well, the agents can solve problems and pass quizzes. Betty does well or poorly depending on how she has been taught. The students have a chance to refine their teaching in the process, and they develop a deeper understanding of a subject than do those who prepare for a test. "These environments are very motivating," Bransford says. Students are more willing to stick with teaching Betty until she gets it, than to correct their own work.

Here are some results. Students were assessed on how well they created concept maps of the nitrogen cycle. The groups in the middle and at right learned by teaching, and the self-regulated group—on the right—did the best.

Another quick snapshot showing the value of learning by teaching: students were assessed on their accuracy at reasoning tasks, specifically, to induce, imply or translate. In all cases, those who learned by teaching—the green bars—did best.

Another approach using new media employs web-based video to lend greater depth and richer context to students' science investigations. A problem is set up—say to convey the concept of invasive species—using video. In between video segments, students carry out related experiments. I would like to give the flavor of one such project led by Bob Sherwood at Vanderbilt University in which students are learning about invasive species in their areas. Ultimately the hope is for the project to link classrooms communicating about their investigations in different parts of the country.

Here is a short video excerpt.

[slide not available]
(Use "back" to return to the text.)

Moving out of the classroom and into the realm of robotics, Daniela Rus at the Massachusetts Institute of Technology works on self-organizing robots. These consist of multiple, autonomous components that make local decisions, which create coordinated behavior for the whole system. In the video we will see a distributed learning system in action, where each module gets negative feedback if it disconnects and or moves backward, and positive feedback otherwise. The video, please...

[Play Rus "inch-ten-step" video: "the crystal robot"] video not available
(Use "back" to return to the text.)

I'll show you an animation now, again from Rus's work. The animation demonstrates how self-organizing robots can move along by conforming to the changing shape of the terrain, whether moving through a tunnel or up the side of a building. They could also distribute themselves to form a sensor network for monitoring. Such robots could work well in hazardous or remote environments. Let's watch the video now...

[Rus tunneling video] video not available
(Use "back" to return to the text.)

[Manuela Veloso's Robocup soccer dogs]
(Use "back" to return to the text.)

Here are robots from another realm—the world of sports. The robot dog teams of Manuela Veloso of Carnegie-Mellon University and her colleagues compete internationally in the sport. Lest it all look only like fun-and-games, Manuela points out that Nobel economist Herb Simon ranked robot soccer right at the top of robot achievement, because it creates the ultimate environment for research on communication, movement and learning. The dogs learn in two ways: offline, like athletes practicing skills, they learn which function they should apply in given conditions. But they also learn during a game.

As Veloso says, "Our robots learn by themselves which play works best. They adapt to their opponents, and they become smarter players by the end of a game." Let's watch a short clip of a match. Keep your eye on the red team.

[Video of robosoccer match: 51 seconds]
(Use "back" to return to the text.)

[GK-12 sites]
(Use "back" to return to the text.)

I will conclude now by leaving the world of robotics and returning to emphasize the synthetic character of our new ideas about learning. Seen here are what NSF calls our GK-12 sites across the country—a program that implicitly acknowledges that the best learning is not one-way. In this program to place graduate students in classrooms, the rivulets of learning flow in all directions, among the graduate student, the classroom teacher, the K-12 students, and everywhere in between. A Harvard fellow commented, "The program helped make me realize that you can equipartition your energy into teaching and research and be effective at both." In the meantime, making such connections among the various levels of learning will help underscore that early learning is as important as a higher degree.

[concluding montage]
(Use "back" to return to the text.)

The saying goes that "A little learning is a dangerous thing." But putting many minds together—from cellular biologists to neuroscientists to cultural anthropologists and beyond--we have learned much more than a little. We have already learned that learning reorganizes the brain; that different parts of the brain may be ready to learn at different times; that cultural and social settings influence learning—and myriad other insights. These are just the beginning of what we may expect from the emerging science of learning.

Thank you.


1 Kenneth Ford, Founder and Director, Institute for Human & Machine Cognition
Return to speech.

2 How People Learn, NAS study.
Return to speech.

3 Article in The Chronicle of Higher Education, 6-20-03, "From Video Games, Learning About Learning," by James Paul Gee.
Return to speech.

4 Science, 12/19/03.
Return to speech.

 

 
 
     
 

 
National Science Foundation
Office of Legislative and Public Affairs
4201 Wilson Boulevard
Arlington, Virginia 22230, USA
Tel: 703-292-8070
FIRS: 800-877-8339 | TDD: 703-292-5090
 

NSF Logo Graphic