Researchers from Carnegie Mellon University have created a way to rapidly build and teach intelligent tutoring systems. The new method uses artificial intelligence (AI) to allow teachers to teach an AI tutor by demonstrating several ways to solve problems and correct the computer if it is wrong.
The new method allows teachers to create a 30 minute lesson in 30 minutes using a machine learning program that simulates student learning. The team created a teaching interface for the machine learning engine that is user friendly and employs a process that is easier than programming.
The team demonstrated the new method with multi-column addition, although it can work with a variety of subjects. The method speeds up the development of intelligent tutors and makes it possible for teachers to build their own computerized lessons, eliminating the need for AI programmers. Teachers can adjust the system to their teaching preferences, leading to deeper insights into learning. The system will stumble in the same places that students will, showing teachers difficulties that they may not have seen otherwise.
Typically, computers can generalize to solve all problems in a given topic in a way that differs from the teacher. Intelligent tutoring systems are designed to track student progress, provide hints and pick practice problems. But these learning systems need to learn all the of ways to solve a problem, which has been a challenge to their development.
To overcome this, the team programmed production rules by hand, which took 200 hours of development for each hour of tutored instruction. Later, they created a shortcut that attempted to demonstrate all of the possible ways of problem-solving, which cut development time to 40 or 50 hours. But it is impossible to demonstrate all of the possible solution paths for all of the potential problems for many topics, which initially reduced the shortcut applicability.
A paper on this technology was published in the conference proceedings in the Association for Computing Machinery’s Digital Library.
