Using an artificial intelligence tool can be as accurate as human assessors in level one screening for a systematic review
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
BACKGROUND: Artificial intelligence (AI) offers a promising solution to expedite various phases of the systematic review process such as screening. OBJECTIVE: We aimed to assess the accuracy of an AI tool in identifying eligible references for a systematic review compared to identification by human assessors. METHODS: For the case study (a systematic review of knowledge translation interventions), we used a diagnostic accuracy design and independently assessed for eligibility a set of articles (n = 300) using human raters and the AI system DistillerAI (Evidence Partners, Ottawa, Canada). We analysed a series of 64 possible confidence levels for the AI's decisions and calculated several standard parameters of diagnostic accuracy for each. RESULTS: When set to a lower AI confidence threshold of 0.1 or greater and an upper threshold of 0.9 or lower, DistillerAI made article selection decisions very similarly to human assessors. Within this range, DistillerAI made a decision on the majority of articles (93-100%), with a sensitivity of 1.0 and specificity ranging from 0.9 to 1.0. CONCLUSION: DistillerAI appears to be accurate in its assessment of articles in a case study of 300 articles. Further experimentation with DistillerAI will establish its performance among other subject areas.
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.
Direct model labels (unvalidated)
Per-model category and study-design labels from the labeling rounds. They are machine output, unvalidated, and the disagreement between models ships as data. No study design here is MEDLINE-validated yet.
| Model arm | Categories | Study design | Confidence |
|---|---|---|---|
| gemma | Metaresearch Domain: Methods · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Observational | high |
| gpt | Metaresearch Domain: Methods · Genre: Empirical About the Canadian research system: no · About a Canadian topic: no | Other design | high |
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.143 | 0.065 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.018 | 0.003 |
| Bibliometrics | 0.002 | 0.004 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.013 | 0.007 |
| Open science | 0.003 | 0.000 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.000 |
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 it