[ad_1]
Smart tutoring systems have been shown to be effective in helping to teach certain subjects, such as algebra or grammar, but creating these computerized systems is difficult and laborious. Now, Carnegie Mellon University researchers have shown that they can quickly build them by teaching the computer to teach.
Using a new method that employs artificial intelligence, a teacher can teach the computer by demonstrating various ways to solve problems on a topic, such as adding multiple columns, and correcting the computer if it responds incorrectly.
In particular, the computer system learns not only to solve problems in the way it was taught, but also to generalize to solve all other problems on the subject and to do it in ways that may differ from those of the teacher, said Daniel Weitekamp III, a PhD student at the CMU Human-Computer Interaction Institute (HCII).
“A student could learn a way to solve a problem and that would be enough,” Weitekamp explained. “But a mentoring system needs to learn all kinds of ways to solve a problem.” You need to learn how to teach to solve problems, not just how to solve them.
That challenge has been an ongoing problem for developers creating AI-based tutoring systems, said Ken Koedinger, professor of human-computer interaction and psychology. Smart tutoring systems are designed to continually track student progress, provide clues to the next step, and choose practice problems that help students learn new skills.
When Koedinger and others began building the first smart tutors, they programmed the production rules by hand, a process, he said, that took about 200 hours of development for each hour of tutored instruction. Later, they would develop a shortcut, in which they would try to demonstrate all possible ways to solve a problem. He noted that this reduces development time to 40 or 50 hours, but for many issues, it is virtually impossible to demonstrate all possible solution routes for all possible problems, reducing the applicability of the shortcut.
The new method can allow a teacher to create a 30-minute lesson in about 30 minutes, which Koedinger called “a great vision” among smart tutor developers.
“The only way to get to the full smart tutor so far has been to write these AI rules,” said Koedinger. “But now the system is writing those rules.”
A document describing the method, written by Weitekamp, Koedinger and HCII system scientist Erik Harpstead, was accepted by the Conference on Human Factors in Computer Systems (CHI 2020), which was scheduled for this month but canceled due to the COVID-19 pandemic. The document has now been published in the conference proceedings in the Digital Library of the Association of Computing Machinery.
The new method uses a machine learning program that simulates how students learn. Weitekamp developed a teaching interface for this machine learning engine that is easy to use and employs a “show and fix” process that is much easier than programming.
For the CHI article, the authors demonstrated their method on the topic of multi-column addition, but the underlying machine learning engine has been shown to work for a variety of topics, including solving equations, summing fractions, chemistry, English grammar, and environments for scientific experiments.
The method not only speeds up the development of smart tutors, but promises to enable teachers, rather than AI programmers, to develop their own computerized lessons. Some teachers, for example, have their own preferences about how addition is taught or what form of notation to use in chemistry. The new interface could increase adoption of smart tutors by allowing teachers to create the tasks they prefer for the AI tutor, Koedinger said.
Allowing teachers to create their own systems could also lead to a deeper understanding of learning, he added. The authoring process can help them recognize the trouble spots for students that, as experts, they themselves cannot find.
“The machine learning system often stumbles in the same places as the students,” said Koedinger. “While teaching the computer, we can imagine that a teacher can get new ideas about what is difficult to learn because the machine has trouble learning it.”
Reference
Weitekamp et al. (2020). An interaction design for teaching machines to develop artificial intelligence tutors. CHI ’20: Proceedings of the CHI 2020 Conference on Human Factors in Computer Systems. DOI: https://doi.org/10.1145/3313831.3376226
This article has been reissued from the following materials. Note: the material may have been edited for length and content. For more information, contact the source cited.
[ad_2]
