Most developers interested in applying machine-learning tech-niques to develop intelligent, adaptive instruction products for theclassroom lack access to the large digital data sets needed to trainthe models.

I have identified three areas in which AI-based solutions haveshown promise for supporting teachers in challenging areas ofinstruction: adaptive instructional systems that allow teachers todifferentiate instruction at the student level for certain topic areasand skills; automated scoring of student writing assignments, whichsupports teachers’ ability to assign more writing in the classroom;and early warning systems, which alert administrators and teacherswhen students may need additional support to stay on track andprogress toward graduation.

To build a statistical model that can reliably predictoutcomes (e.g., whether an X-ray image contains a tumor) requiresaccess to existing data sets to train the system and to independentlyvalidate (or ground truth) the accuracy of the predicted outcomes.This process of training and validation is called supervised learning.