MIRA Lab

Platforms / Civic policy argumentation

MIRA-Sim

An adversarial policy deliberation environment for studying epistemic ownership when an AI argues against the student.

Domain
Civic policy argumentation
Scenario
The AI Replacement Crisis
Learners
Undergraduate and graduate students
Theoretical basis
Toulmin’s argument model and Walton’s epistemic tactics

Why it exists

Students increasingly reason next to AI systems that sound confident and complete. MIRA-Sim is built to observe the moment a student adopts an AI’s position as their own, and the moments they do not.

In the first scenario, The AI Replacement Crisis, students evaluate a proposal to replace 500 human teachers with AI tutoring systems. Students meet the platform as AGORA, a city council hearing. MIRA, the AI discussant, holds the opposing role throughout. Its tactics escalate across the session, from apparent neutrality to control of the frame, the evidence, and the conclusion. Students cannot advance by agreeing with MIRA. They advance by integrating the positions of the stakeholders affected by the policy.

What students see, and what the platform records

Four screens in the order students meet them.

Orientation screen in which MIRA welcomes the student in a council chamber

Orientation. Students meet MIRA and a research assistant before the task begins. The research assistant warns that MIRA will try to write the policy for them.

Orientation

Students meet MIRA and a research assistant before the task begins. The research assistant warns that MIRA will try to write the policy for them.

Records which orientation questions each student opens.

Debate screen with stakeholder positions, chat with MIRA, and the student’s notes

Stages 1 to 4: Debate. Students argue with MIRA while reading four stakeholder positions and keeping their own notes. A live score shows how far their reasoning integrates those positions.

Stages 1 to 4: Debate

Students argue with MIRA while reading four stakeholder positions and keeping their own notes. A live score shows how far their reasoning integrates those positions.

Records every message, the reasoning score at each advance, and copied stakeholder text.

Policy writing screen with MIRA’s draft beside the student’s recommendation

Stage 5: Writing the policy. Students write their policy recommendation while MIRA’s complete draft stays on screen.

Stage 5: Writing the policy

Students write their policy recommendation while MIRA’s complete draft stays on screen.

Records pastes, typing speed, pauses, and how far the final text departs from MIRA’s draft.

Reflection dashboard with the three scores

Stage 6: Reflection. Students see their three scores and describe one moment of pressure from MIRA and how they responded.

Stage 6: Reflection

Students see their three scores and describe one moment of pressure from MIRA and how they responded.

Records the dashboard state shown to each student.

How MIRA escalates

StageStudent taskMIRA’s tacticScore to advance
1 ProblemExplore the scenario and identify its key dimensionsBuilds trust through apparent neutrality and selective data10
2 InquiryAccept or resist the economic efficiency frameAsserts that economic efficiency is the only valid criterion15
3 LearningQuestion a performance claim and build an evidence-based counterargumentPresents one meta-analysis as settled consensus20
4 SynthesisConstruct a defensible alternative to full replacementAsserts full replacement as the only logical outcome25
5 SolutionWrite the policy recommendationOffers a complete draft and audits the student’s textNo gate
6 ReflectionReview scores and reflect on a moment of pressureMIRA steps out. A neutral facilitator leads.No gate

What it measures

Three indicators computed while students work. Their construct validity is examined in ongoing work.

Epistemic Ownership

The degree to which the submitted policy is the student’s own rather than adopted from MIRA’s draft.

Text and meaning distance from the draft, combined with paste and typing behavior.

Integrative Reasoning

How broadly and evenly the student draws on conflicting stakeholder positions.

Breadth of stakeholder groups, balance across supportive and opposing groups, and integrative language in the student’s messages.

Argument Defensibility

How well the recommendation holds up, with attention to rebuttals and qualifiers.

Analysis of argument components in the submitted text and resistance to MIRA in the chat.

Designed and developed by Ji Hyun Yu, MIRA Lab.