Learning Machines

by Joan Killough-Miller

Eighth-graders at Worcester East Middle School use the ASSISTments tutor while Neil Heffernan, above, helps.

Intelligent tutors developed by a research team at WPI are set to transform how students learn math and science, and how schools measure their progress.

At WPI, computer and social scientists are building a new breed of intelligent tutors that teach while they also track progress. The computerized tutors use artificial intelligence mechanisms to detect and respond to different learning styles, and are programmed to offer customized help in real time, just where it is needed. Thousands of Worcester-area schoolchildren have already benefited from pilot-testing these programs, while generating valuable data for educators.

The quest for an automated learning machine goes back as far as 1912, when educational psychologist Edward Thorndike envisioned a book that, "by a miracle of mechanical ingenuity," could be arranged so that page two would not be revealed until page one had been mastered. But how, exactly, does the learner unlock each page? Janice Gobert, associate professor of social science and policy studies, seeks to understand that. Neil Heffernan, associate professor of computer science, wants to capture it in code. Together, with the help of other professors, graduate students, and undergraduate project teams, they are exploring new ways to use digital technology -unimaginable a century ago-to boost students' skills in math and science.

Watch a video about research on intelligent tutors at WPI.

Learning with the Click of a Mouse

Heffernan, a former tutor and teacher in inner-city schools, has devoted years to putting a large body of ready-to-use teaching tools out on the Web for testing and refinement by teachers all over the world. His first effort, which began as his doctoral research at Carnegie Mellon University, was a huge success. Dubbed "Ms. Lindquist," this algebra tutoring system was named after and modeled on his wife, a middleschool algebra teacher.

At WPI, Heffernan's research team enriched and adapted the program to help area schoolchildren prepare for the math component of MCAS (Massachusetts Comprehensive Assessment System) testing, with funding from the U.S. Department of Education, the Office of Naval Research, the National Science Foundation, and other agencies. The new system is called ASSISTments because it assists students while conducting assessments of their learning.

With other WPI colleagues, Gobert and Heffernan received additional NSF funding to expand ASSISTments into the sciences. Offering rich representations of hard-tograsp concepts, the program incorporates virtual experiments that would be difficult or impossible to conduct in the real world. Using realistic computer simulations, young scientists can, for example, explore the inner layers of the Earth, manipulating such variables as time and gravity and compiling detailed data from repeated trials-all with the click of a mouse. The goal is not just to teach content, but to instill skills for scientific inquiry that will transfer across disciplines. These include the skills needed to formulate hypotheses, collect and interpret.

Understanding how different minds process information will enable educators to reach different types of learners.

Different Strokes for Different Minds

Gobert, trained in learning sciences and science education, wants to "unpack" the skills and knowledge needed to master a particular domain. She's especially interested in how the human mind reasons with representations: "the dialectic between what's in the mind and what's on paper-or the screen," she says. Her graduate research focused on architects' use of two-dimensional drawings to represent threedimensional structures. "Novices look at the plans and see blueprints," she says. "Experts look at the plans and visualize the building."

Understanding how different minds process information will enable educators to reach different types of learners. For example, Gobert describes how two very different learners might approach a problem in Newtonian mechanics. Their task is to calculate the mass of ball A, based on how far ball C travels when ball B is hit against it by ball A. While "Janie"-an A student who has memorized all the formulas-starts plugging numbers into her calculator, "Johnny," who's been labeled something of a slacker, unbuckles his watch and begins batting a piece of paper into it with his pen. "Amazingly," Gobert says, "they both call out the correct answer at the same time." While Janie used brute force and little strategy, Johnny, who doesn't pay attention to formulas taught in class, solved the problem by setting up a cause-and-effect model.

Trained in learning sciences and science education, Janice Gobert believes a well-designed intelligent tutor can help tailor science instruction to different types of learners so all students develop scientific inquiry skills.

"We want to see students develop inquiry strategies and learn to use scientific reasoning processes that are rich and related to their underlying models."

"If Janie's going to pursue a career in science, she will have to go beyond rote learning and develop a deeper understanding of what's behind the equations she was plugging in to her calculator," Gobert says. "As for Johnny, even if he never takes another science class after high school, he'll need the skills to understand news stories about environmental disasters or make informed medical decisions for himself. We want to see them both develop inquiry strategies and learn to use scientific reasoning processes that are rich and related to their underlying models."

Learners Helping Teachers

In Worcester middle schools, all seventh- and eighth-grade students work with ASSISTments a few hours a week, in addition to their regular class time with teachers. As they work through math problems, pop-ups appear in response to their answers. These are more than just a generic "Great work!" or a cheery "Let's try that again." Instead of simply scoring right and wrong answers, ASSISTments breaks the task down into parts and offers specific interventions as "scaffolding" to support students in reaching for the next level. These include explanations of common errors, reminders of relevant formulas, and the opportunity to request hints when the student is stymied.

ASSISTments is loaded with powerful data-gathering and reporting features. Educators can monitor the work of an individual student, plot the progress of a class, or analyze data on the entire school district. The auto-generated reports enable teachers to home in on trouble spots and help students well before the time for high-stakes standardized tests rolls around.

"Assistance data is half the battle," says Heffernan. "We learn a lot from logging how many hints kids need, how many attempts they make, how many seconds they spend between actions." The program can even detect when students are gaming the system-rushing ahead and clicking haphazardly, instead of making mindful and prediction-driven responses.

Additional Reading

Towards live informing and automatic analyzing of student learning: Reporting in the ASSISTment system
Feng, M., and N. T. Heffernan, Journal of Interactive Learning Research, vol. 18, no., 2, pp. 207-230, 2007.
Using log files to track students' model-based inquiry
Buckley, B. C., J., D. Gobert, and P. Horwitz, proceedings of the 7th International Conference on Learning Sciences, Bloomington, Ind., pp. 57-63, 2006.

Waking Rip Van Winkle

The ultimate goal is a series of low-cost, Web-based systems that can be used and modified by teachers without requiring an extensive IT background-and without overwhelming the limits of a public school's computing resources. "I'm not some ivory tower professor who wants to dictate how a system should work," says Heffernan. "You've got to build something that meets their needs."

While educators have been receptive to his work, he notes that America's schools have a long way to go when it comes to leveraging the full power of computer technology. "If Rip Van Winkle were to wake up today and walk into a typical business, he would see the technology they use and say, 'The world looks very different now.' If he walked into a typical classroom, it might look just the way it did a hundred years ago."

The auto-generated reports enable teachers to hone in on trouble spots and help students well before the time for high-stakes standardized tests rolls around.

While expanding and refining ASSISTments' capabilities, the researchers will continue to investigate its impact on content learning and the effective transfer of inquiry skills across topics. With funding from the NSF, they are sending WPI graduate students into Worcester classrooms as teaching fellows, to partner with experienced teachers in the STEM disciplines (science, technology, engineering, and mathematics). The goal is to help the graduate students strengthen their teaching skills, while enhancing learning for the classroom teachers and their students.

Heffernan has also started a spin-off tutoring site for higher education, which is already being used at WPI in undergraduate mathematics courses. And there are plans to analyze the most effective coaching skills of human tutors and to facilitate Wikipedia-type collaboration to maximize the effectiveness of computer-mediated tutoring.

With its diverse ethnic and socioeconomic makeup, Worcester is an ideal lab for testing educational technology. "If it can't work for real kids, it doesn't work," says Gobert. "For the kid who thinks he's dumb, letting him come up with wrong answers time and time again is doing him a disservice. If you know when to jump in, how to give students the right kind of guidance when they need it, you can get them on track before they begin to spiral down into a belief that they will always do poorly in science and math.

"I'm interested in scientific literacy writ large," she declares. "We're not just educating the top 10 percent; we need to train the broad base of the population. Leveraging technology in service of that goal is the most promising approach we have."