Evaluating firearm examiner testimony using large language models: a comparison of standard and knowledge-enhanced AI systems
Bibliographic record
Abstract
This study evaluated the decision-making of Large Language Models (LLMs) in interpreting firearm examiner testimony by comparing a standard LLM to one enhanced with forensic science knowledge. The present study is a replication study. We assessed whether LLMs mirrored human decision patterns and if specialised knowledge led to more critical evaluations of forensic claims. We employed a 2 × 2 × 7 between-subjects design with three independent variables: LLM configuration (standard vs. knowledge-enhanced), cross-examination presence (yes vs. no), and conclusion language (seven variations). Each model condition performed 200 repetitions per scenario. This yielded a total of 5,600 measures of binary verdicts, guilt probability ratings, and credibility assessments. LLMs showed low conviction rates (9.4%) across conditions, with logical variations as a function of the way in which the firearm expert’s conclusion was formulated. Cross-examination produced lower guilt assessments and scientific credibility ratings. Importantly, knowledge-enhanced LLMs demonstrated significantly more conservative evaluations of firearm evidence across all match conditions compared to standard LLMs. LLMs, particularly when enhanced with domain-specific knowledge, showed advantages in evaluating complex scientific evidence compared to human jurors in Garrett et al. (2020), suggesting potential applications for AI systems in supporting legal decision-making.
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 | 0.020 | 0.116 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 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".