ARIA is presenting at the Khan Lab School AI in Education SummitOctober 24 · Mountain View
Teacher activity

About

About ARIA

Mission

ARIA studies when a math tutor should speak up, and when it should stay quiet.

A student-led research initiative at Marvin Ridge High School. We measure the timing decision on public data first, build the tutor around it, and talk to teachers before anything reaches a classroom.

  • Research

    Offline evaluation on public data. A written-down protocol, a locked test set, and the negative results published next to the positive ones.

  • Tool

    The ARIA tutor and its teacher-facing settings. Each step is checked symbolically; Quiet, Balanced and Supportive are cost settings for the timing rule.

  • Engagement

    Teachers, the Khan Lab School summit, and a planned reviewed pilot. Criticism of the study design is the most useful thing anyone can give us right now.

What ARIA is

Getting the answer is not the same as learning to recover.

Origin

The question came from wondering why tutors interrupt students who are mid-thought. Help that arrives too fast can remove the chance to plan, work through confusion, and recover when a strategy fails. That motivates the study; it is not evidence for it.

The gap

Most tutoring software responds to every step. The learning-sciences literature has long asked when help should be given and when it should be withheld, but few tools treat staying quiet as a decision in its own right.

The research question

Can a tutor that verifies each step and chooses when to intervene reduce unnecessary interruptions without leaving stuck students unhelped, and does that timing change how students persist? ARIA is the instrument for asking it.

Scope

Why math?

Four reasons, all specific to mathematics. They explain a scope choice. They do not promise an outcome.

  1. 01

    A step can be checked.

    Algebra and calculus steps can be verified symbolically with SymPy. A wrong step is a checkable fact, not a guess about how the student feels, so ARIA only considers speaking after a verified error.

  2. 02

    Productive struggle is well studied in math.

    The assistance dilemma (Koedinger and Aleven, 2007) and productive failure (Kapur, 2008) both come out of math learning. There is a real literature for this work to be measured against.

  3. 03

    Wheel-spinning was defined on math practice.

    Beck and Gong (2013) described students who keep attempting a skill without mastering it, using math practice data. It is the outcome ARIA-Pace predicts.

  4. 04

    The logs we have are math.

    The public ASSISTments 2009–10 skill-builder logs are math problem sets. Our replay says something about those logs only, and nothing about other subjects.

This is a scope choice, not a hidden limitation. ARIA is not a general homework assistant, and we make no claims about subjects we have not studied.

Roadmap

Five phases, in order. We are early.

  1. Phase 1Synthetic think-aloud dataComplete
  2. Phase 2Timing model on real student logsCurrent
  3. Phase 3Student interviewsPlanned
  4. Phase 4Formative prototype testingPlanned
  5. Phase 5Classroom and teacher-informed evaluationPlanned
Read the full roadmap

Team

Two student researchers.

Naren Saravanan and Karthik Malireddy, Marvin Ridge High School, Waxhaw, North Carolina.

Meet the team

Get involved

Where we need help.

  • Teachers

    Try the 8 moments and help plan a reviewed classroom pilot.

    Get in touch

  • Researchers

    Review our methods, replicate the replay, or collaborate.

    Get in touch

  • Students

    Tell us how a hint should feel when you are stuck.

    Get in touch

  • Advisors and mentors

    We are looking for an adult advisor for ethics review and study design.

    Get in touch