Researchers at the University of Wisconsin at Madison say they are getting closer to designing a system to deliver the ideal lesson plan for each student, through a process they call “machine teaching.”
If the idea of machine learning, a popular area of artificial-intelligence research, is to let computers learn from data to detect patterns and better deal with large data sets, machine teaching looks for the best way to share particular information with a student, says Jerry Zhu, an associate professor of computer science at the university who is leading the project. Eventually, the approach could create the best lesson for a particular student, Mr. Zhu says.
“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,” he says.
One challenge in developing such models is that much of what is known about how students learn is qualitative, but to build an algorithm, the research team needs more quantitative data. To get it, they are working with researchers in psychology and educational psychology, as well as other computer scientists.
Initial applications for machine teaching include online tutoring systems and helping professors fine-tune a curriculum, according to Mr. Zhu.
“The hope is if we can quantify the student’s learning process,” he says, “then maybe we can come up with a more efficient curriculum or lesson.”
Of course, many professors have argued that good teaching defies the kind of bean counting inherent in a machine-learning approach. But Mr. Zhu argues that if his approach works, the proof will be in the results.