Platforms / Data science problem solving, in development
HyCoReg
A scenario-based environment for problem-based learning in data science. Each student works as a data analyst at a fictional company and completes a full analysis of real learning data with four AI agents and a team of classmates. It builds on MIRA-Agent, so code still runs only after the student explains it.
- Domain
- Problem-based learning in data science
- Format
- Individual and team stages with four AI agents
- Status
- In development
- Theoretical basis
- Co-regulation and productive friction
- Funding
- NSF IUSE Award #2142327 (PI: Xun Ge)
Why it exists
HyCoReg stands for Hybrid Cognition for Co-Regulated Problem Solving. In hybrid cognition, the regulation of learning is shared among learners and AI agents. Generative AI agents can now plan, check, and evaluate work for a learner. Research on collaborative learning treats support from others as a step toward the learner’s own regulation. When an agent keeps doing this work, support becomes delegation. HyCoReg asks whether learners take up the regulatory work an agent offers, or hand it back to the agent.
The project moves through five rooms on one office floor, and each room is one stage of problem-based learning. In the Meeting Room, the student hears the problem from a client and writes it in their own words. At Your Desk, the student lists what they know and what they need to learn, and plans the analysis. In the Analysis Lab, the student carries out the analysis in Python. In the Team Room, three or four students compare their findings, question each other’s decisions, and agree on one recommendation. In the Boardroom, the client changes a condition and the student reruns the analysis.
Four AI agents, each disclosed as AI, have one role each. Jordan (Senior Analyst) supports planning and monitoring. Sam (Teammate) supports motivation and goal setting. Riley (QA Reviewer) supports peer review. Morgan (Project Lead) coordinates the team. Every offer gives the student two equal choices, to do the work or to let the agent do it. The agents never write the student’s code, analysis, or report.
What students see, and what the platform records
Design mockups from the storyboard. The platform is in development, and the final design may change.
Project map (design mockup)
The five stages of the project are five rooms. A room opens when the student finishes the one before it, and earlier rooms stay open for revisiting. The four AI agents and their roles are always visible.
An agent offers help (design mockup)
At the planning board, Jordan offers to suggest what the student needs to learn. The student can add the items alone or let Jordan suggest some. Suggested items stay editable, and the student decides what to keep.
Defend: the client challenges a decision (design mockup)
Dana, the client, questions one of the student’s real analysis decisions by email. The student keeps or revises the decision and writes a reason.
Defend with a teammate (design mockup)
In the Team Room, students review each other’s findings cards and question or challenge decisions. Riley, the QA reviewer, comments only when the student asks.
Extend: a changed condition (design mockup)
In the Boardroom, Dana changes a condition, for example four outcomes instead of two. The student reruns the analysis and explains what changed and why.
How the work was shared (design mockup)
After the final task, each student sees, room by room, how often they did the planning, checking, or reviewing themselves and how often they handed it to an agent.
What it measures
Planned measures. The analysis compares the same student across individual and team stages.
Take-up and hand-over
For each agent offer, whether the student did the work or returned it to the agent, with the agent’s role and the stage.
Individual and team stages
Each student’s hand-over rate when working alone and when working with a team.
Defend and Extend responses
Whether students keep or revise a challenged decision, the reasons they give, and how they adapt their analysis to a changed condition.
Performance without agents
At the end, the agents switch off, and each student explains, defends, and extends their analysis alone. The analysis asks whether students who handed over more work do less well on this task.
Designed and developed by Ji Hyun Yu, MIRA Lab, for the NSF IUSE project led by Xun Ge at the University of North Texas.