Expert Alternatives
The policy research team studies problems, evidence, reform alternatives, and constraints.
APPLICATION 01Research in Progress
Pension Reform Project
This is AIPOL's first use case, connecting policy experts' pension-reform alternatives, synthetic-citizen responses, AI-generated additional alternatives, and human review.
Research question
How can AI assist in developing complex policy alternatives?
Designing and validating the roles of humans and AI for pension reform, which intertwines intergenerational interests, long-term finances, retirement income, and system trust.
Synthetic Citizen Validation
Two full-journey rehearsals with 100 and 250 synthetic citizens validated the method and tooling.
Opinions and Issues
AI classifies participants' stated reasons into five issues and returns them as material for the next deliberation round.
Additional Alternatives
AI proposes based on opinions, while humans review, modify, and approve.
Development record
Research and tool development are conducted concurrently.
Methodology and execution paths are verified with synthetic data prior to real-world application, distinguishing results from actual public opinion.
First Synthetic Experiment · 100 citizens
100 synthetic citizens completed the full 13-stage journey. 97 finished, and 66% rated the AI-drafted alternative D as acceptable or conditionally acceptable.
First experiment record →Second Synthetic Experiment · 250 citizens
The final configuration—250 citizens, a D-only third round, and the confirmed policy-variable constraints—was executed end to end.
Second experiment record →First Field Application
Policy experiments will be conducted with a real audience at the Korean Association for Policy Studies flagship session.
Event Info →What will be shared
Public Assets to Leave Behind in the Case
Leave materials for reuse and critique by other research, rather than a single policy conclusion. Unpublished policy proposals and internal operations materials are not included.
Methodology
Policy development stages and human-AI roles.
Prompts and Conditions
Model, version, input, and review criteria.
Tools
Research engine and online/on-site participation tools.
Limitations
Framing, sample, errors, and unaddressed issues.