MIRA Lab

Platforms / Algebra, in development

MIRA-Math

An Algebra I environment for linear functions in context. Students work through each task in five steps, and their teacher decides how much MIRA, the AI tutor, helps at each step.

Domain
Linear functions in Algebra I
Format
Individual tasks with an AI tutor, set up by the teacher
Status
In development
Theoretical basis
Productive friction, placed by the teacher

Why it exists

In our other platforms, the design team decides where learners must reason without AI help. Teachers know better where their own students go wrong. MIRA-Math moves this decision to the teacher. For each step of a task, the teacher chooses how much MIRA helps and writes one line on why.

Each task follows five steps. In Learn, students set the starting level and the rate of a filling tank and watch the water level and the line on the graph change together. In Explain, they put the relationship into their own words. In Solve, they write the equations and find where the lines cross. In Defend, they judge a claim from an AI tool, such as "Tank B is always lower, because its rate is lower," and support their judgment with the tanks and the graph. In Extend, they predict how a change will move the answer, rate how sure they are, and then run the simulation to check.

The question is how teachers decide where AI should support students and where it should challenge them, and what these decisions mean for what students can explain and defend on their own. Teachers will help design MIRA-Math and will set the friction for their own classes.

What students and teachers see, and what the platform records

Design mockups. The platform is in development, and the final design may change.

Design mockup of the Learn step with one water tank, sliders for the starting level and the rise each minute, and a matching line graph

Learn (design mockup). Students set the starting level and the rise each minute for Tank A and run it. The water level and the line on the graph move together. MIRA’s help is open at this step.

Learn (design mockup)

Students set the starting level and the rise each minute for Tank A and run it. The water level and the line on the graph move together. MIRA’s help is open at this step.

Planned records: each setting the student tries and each run of the simulation.

Design mockup of the Defend step with two water tanks, a claim from an AI tool, agree and disagree buttons, a graph of both lines, and the student’s written reason

Defend (design mockup). Students judge a claim from an AI tool and explain their judgment with the tanks and the graph. MIRA’s help is off at this step, so the reasoning stays with the student.

Defend (design mockup)

Students judge a claim from an AI tool and explain their judgment with the tanks and the graph. MIRA’s help is off at this step, so the reasoning stays with the student.

Planned records: the student’s judgment of the claim and the written reason.

Design mockup of the Extend step with a prediction about the crossing point, a confidence rating, and the graph after the change

Extend (design mockup). Students predict how the crossing point will move, rate how sure they are, and then run the change to check their prediction. MIRA gives hints only.

Extend (design mockup)

Students predict how the crossing point will move, rate how sure they are, and then run the change to check their prediction. MIRA gives hints only.

Planned records: each prediction, its confidence rating, the result after the run, and requests for help.

Design mockup of the teacher view with the task goal, likely student errors, and a help setting and reason for each step

Teacher friction settings (design mockup). The teacher states the goal of the task, marks where students are likely to go wrong, and chooses how much MIRA helps at each step, with a one-line reason. Confidence ratings, a final word problem, and a unit quiz stay on for every class.

Teacher friction settings (design mockup)

The teacher states the goal of the task, marks where students are likely to go wrong, and chooses how much MIRA helps at each step, with a one-line reason. Confidence ratings, a final word problem, and a unit quiz stay on for every class.

Planned records: each friction setting and the teacher’s reason.

What it measures

Planned measures. The analysis connects teachers’ decisions to what students can do.

Teacher friction decisions

Where each teacher turns MIRA’s help up or down, and the reasons teachers give. The analysis compares these decisions with the errors teachers expected.

Claim evaluation

Whether students judge an AI claim correctly and support their judgment with evidence from the tanks and the graph.

Prediction and confidence

Whether students’ predictions match what happens when they run the change, and whether their confidence matches their accuracy. A gap between confidence and accuracy is a sign of the fluency trap.

End-of-unit performance

Results on the final word problem and the unit quiz, which every class completes. These show what students can do at the end of the unit.

Designed and developed by Ji Hyun Yu, MIRA Lab, with colleagues in Data Science and Mathematics at the University of North Texas.