An interaction design project in which I conducted secondary research on Machine Learning trends and use cases, as well as primary research with stakeholders such as teachers and parents. The solution was a tablet-hosted Machine Learning system that employs passive inputs (eye-tracking) and manual inputs to measure and reflect learning style.

Our solution

A system that trains itself to individual students unique needs, while improving upon the grading and curriculum model. We introduce a new way of measuring understanding, followed by automated reporting to the teacher. Teachers can spend their time with students directly, instead of with their work.

Customizability could be indispensible in a learning setting, where each child has a unique way of processing information. The one-size-fits-all approach that makes of the majority of classrooms in America is an outdated system. 

However, despite automation, custom lesson plans may not be possible within the rigid structure of traditional schools. Furthermore, technology in classrooms was often cited as more of a distraction than a tool. Online resources were vague about specific curriculum and technology requirements, so I decided to conduct primary research. 

Insights

01. Grading is the the most labor intensive and least satisfying task.

02. Teachers are not opposed to technology.

03. Teacher-Student relationships are the most important and most rewarding part of the job

04. Curriculum structure can vary in flexibility per subject

Design Principles

01. Foster relationships

02. Alleviate grading for teachers

03. Do not create an isolating experience

We listed out the roles of teachers in two categories:

01. The formal roles required to educate,
02. and the less academic social demands.

I then marked each if they would be better filled by a human (*) or by a machine ([]). Some roles such as "keeping parents in the loop" could be completed by both.

Teaching the System: Inputs

Eye-tracking

Eye movements are indicative of concentration (Hafed and Clark, 2002; Mathot et. al, 2013), and tablet-mounted eye-trackers such as UMoove are currently on the market. Measurements of concentration would inform the system when a student might need a break. Furthermore, eye data indicates what kind of learner a student might be. For example, a student that spends time reading is probably a textual learner, while a student who looks at the picture more may be visual. 

Interactions

Other inputs would be responses on review questions, time information, and calls for assistance.

Teaching the Students: Outputs 

Variable Lessons

The system learns a child's learning style and outputs the appropriate medium. Lessons are pulled from a database and adjust the presentation of information based on each student. Lessons could include images, text, video, activities, audio, or crafts.

Review Questions

Questions appear as the student completes lessons. This is meant to measure comprehension and replace quizzes and homework, where feedback in the form of grades is given days after completion.

Group Activities

Students are prompted several times throughout the day to join other students for group work. Students are grouped based on differing ability to encourage teaching and learning.

Rapidly increasing abilities of Machine Learning, as well as advancing portable eye-tracking technology, will make this vision much less speculative in the coming years. Coupling this with the technology initiatives of schools could impact not only education, but the larger scale of the society those children will join.