‘Machine Teaching’ Research Promotes Personalized Learning


Researchers at the University of Wisconsin at Madison are hoping that ‘machine teaching,’ a new system that harnesses the power of artificial intelligence, will be able to deliver ideal lesson for  each student based on their behavior and learning patterns. The concept of machine teaching is the reverse engineering of machine learning.

The project rests on the idea that machines can collect and interpret data and track learning patterns that could help identify the best way to share knowledge with learners.

Jerry Zhu, an associate professor of computer science at the University of Wisconsin-Madison, who is leading the project, says this AI system could generate an ideal, tailored lesson plan for students:

“We are really building upon what the cognitive community knows about learning, and in particular what we need is a cognitive model that’s computational and individualized,” Zhu said.

A current limitation confronting researchers is a lack of quantitative data. In order for them to build the algorithm, relying on qualitative student learning data doesn’t suffice. That’s why the project brings together scientists from computer science, psychology and educational psychology.

Machine teaching requires a good model of how learners behave and how that behavior adapts to different kinds of learning experiences and activities, Timothy T. Rogers, a professor of cognitive psychology and a colleague of Zhu’s, explains.

Some of the initial implementations of machine teaching will be curriculum personalization and the building of online tutoring ecosystems, Zhu says. If their AI project is successful, they say it will shake up education:

“My hope is that machine teaching has an impact on the educational world. It’s quite different from how people usually think about education,” says Zhu. “It will give us optimal, personalized lessons for real, human students.”

Using complex mathematics, researchers are modeling human students in order to design the best possible lessons. One example of the process is identifying the minimum amount of activities needed to teach an idea. Zhu asks, “Can five really good questions teach the material, rather than 20?

The project was funded by a two-year seed grant from the University of Wisconsin-Madison, and Zhu’s team is seeking additional funding from outside sources As Zhu explains, the research is still at an abstract level, “heavy on math and theory,” and currently the researchers:

“[A]re taking baby steps to apply the methodology to ‘simple’ educational tasks, such as single-digit addition for young kids.”

To answer critics of machine teaching’s limitations, Zhu says that if his machine teaching project works, the results will convince those who doubt it.