Teaching computers to teach

Researchers at Pittsburgh’s Carnegie Mellon University have demonstrated how artificial intelligence can be used to teach computers how to teach.

Carnegie Mellon is at the forefront of education AI and researchers have been successful in overcoming the difficulties associated with creating intelligent tutoring systems for use in subjects including grammar and algebra.

The team said that the difficulty in teaching computers how to teach lies in the need for a tutoring system to learn (and therefore teach) every way of solving a problem, not just the actual solving of a problem. It must be taught problem-solving as a skill.

The research team designed a new method whereby the teacher demonstrates several ways to solve problems and can correct the computer if it responds incorrectly. The benefit is that the machine learns the way it was taught but additionally learns to generalise and solve other problems using methods that may differ from those demonstrated — it learns problem-solving skills.

Researchers say the new method not only promises to drastically cut the time required to develop intelligent tutors, it will also make it possible for teachers (rather than AI programmers) to build computerised lessons, meaning the uptake of intelligent tutor technology could significantly rise. The team sees the initiative as a valuable opportunity for educators as it provides insight into the learning process and helps identify hard-to-spot student ‘trouble spots’, which are common to those in AI learning.

For more information, click here.

Image credit: ©stock.adobe.com/au/phonlamaiphoto