The End of the Policy Analyst? Testing the Capability of Artificial Intelligence to Generate Plausible, Persuasive, and Useful Policy Analysis
Bibliographic record
Abstract
Policy advising in government centers on the analysis of public problems and the developing of recommendations for dealing with them. In carrying out this work, policy analysts consult a variety of sources and work to synthesize that body of evidence into useful decision support documents commonly called briefing notes. Advances in natural language processing (NLP) have led to the continuing development of tools that can undertake a similar task. Given a brief prompt, a large language model (LLM) can synthesize information in content databases. This article documents the findings from an experiment that tested whether contemporary NLP technology is capable of producing public policy relevant briefing notes that expert evaluators judge to be useful. The research involved two stages. First, briefing notes were created using three models: NLP generated; human generated; and NLP generated/human edited. Next, two panels of retired senior public servants (with only one panel informed of the use of NLP in the experiment) were asked to judge the briefing notes using a heuristic evaluation rubric. The findings indicate that contemporary NLP tools were not able to, on their own, generate useful policy briefings. However, the feedback from the expert evaluators indicates that automatically generated briefing notes might serve as a useful supplement to the work of human policy analysts. And the speed with which the capabilities of NLP tools are developing, supplemented with access to a larger corpus of previously prepared policy briefings and other policy-relevant material, suggests that the quality of automatically generated briefings may improve significantly in the coming years. The article concludes with reflections on what such improvements might mean for the future practice of policy analysis.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.081 | 0.291 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.002 | 0.005 |
| Scholarly communication | 0.008 | 0.013 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.007 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".