Loading...
Loading...
Smarter Elections Research
A public research program to test whether transparent, human-validated AI can make political and institutional information more traceable, comparable, and understandable without telling citizens what to think.
The core question
Smarter Elections starts from a simple premise: citizens need more than isolated facts. They need to understand why countries produce different outcomes, how institutions shape incentives, where evidence agrees or conflicts, and how confident we should be in an explanation.
The research challenge is therefore not merely to automate political information. It is to determine whether AI can help organize and explain public evidence while preserving provenance, uncertainty, human review, and political neutrality.
Why now
Public institutions publish enormous amounts of data, legislation, statistics, oversight records, party documents, and electoral information. The difficulty is not only access. It is connecting those materials into explanations that remain inspectable by citizens, journalists, researchers, and institutions.
Official information is spread across institutions, formats, languages, and update cycles.
Summaries often lose the chain between an assertion, its evidence, and the methodology that produced it.
AI can accelerate research, but it can also amplify unsupported synthesis unless provenance and review are designed in from the beginning.
Research design
Smarter Elections currently catalogues 31 countries while full publication remains deliberately limited as source review continues. Auditoría Cero turns that gap into a measurable research baseline.
Audit every catalogued country field-by-field: verified, outdated, contradictory, missing, or unsupported. The audit becomes a public baseline rather than an internal cleanup exercise.
Attach provenance to important claims and indicators: source, publication date, retrieval date, jurisdiction, field, review status, and revision history.
Use AI to structure public records, official statistics, electoral documents, and institutional material into a common schema while preserving the original source trail.
Treat AI output as a draft, not authority. High-impact claims require review, disagreements remain visible, and uncertain evidence is labeled rather than silently resolved.
Move from data to understanding: help users examine why outcomes differ across countries, which institutions and policies may matter, and what evidence supports competing explanations.
Measure accuracy, citation coverage, freshness, contradiction handling, correction speed, and whether explanations improve understanding without optimizing for political persuasion.
Neutrality safeguards
The system is designed to improve understanding, not optimize political behavior. These constraints are part of the research design and should be testable.
Public deliverables
Success is not defined by launching more pages. The program should create reusable research assets that independent reviewers can inspect, reproduce, criticize, and improve.
A structured pilot dataset with field-level provenance and review status.
Definitions, inclusion rules, field types, confidence states, and handling of contradictions.
A traceable record connecting claims and indicators to the evidence used.
The rules for collection, AI assistance, human validation, neutrality, updates, and corrections.
A public product layer that lets users explore evidence, comparisons, and explanatory context.
A transparent account of what worked, what failed, what remains uncertain, and what should change next.
Evaluation
The research should make failure visible. Metrics focus on data integrity, traceability, review quality, and understanding rather than engagement or ideological agreement.
Complete the Auditoría Cero framework, define the common schema, establish the source ledger, and publish baseline coverage gaps.
Validate Costa Rica first, then expand to a small set of contrasting Latin American democracies to test portability and source availability.
Test extraction and synthesis workflows, document failure modes, compare AI output with human-reviewed ground truth, and refine safeguards.
Publish the pilot dataset, codebook, methodology, evaluation results, limitations, and roadmap for broader country coverage.
Research context
This program is independently developed by Smarter Elections. It is not presented as an OpenAI-funded project, partnership, endorsement, or grant recipient.
We are tracking adjacent work because the research category is evolving. In 2026, OpenAI published an exploratory industrial-policy framework and later announced grants to independent projects, including a Yonsei University project testing transparent, human-validated AI methods for democratic accountability in South Korea.
Smarter Elections cites that work as external context only. Our methodology, datasets, conclusions, safeguards, and funding relationships must remain independently disclosed and publicly inspectable.
Collaboration
Research partners who can strengthen the methodology rather than simply amplify the project.
Public commitment
As the pilot progresses, this page should evolve from a research plan into a public record: what was tested, what changed, which assumptions failed, which sources were used, and where the evidence remains incomplete.
Review the current methodology