{"id":"W4414155525","doi":"10.1002/cesm.70045","title":"Artificial Intelligence Search Tools for Evidence Synthesis: Comparative Analysis and Implementation Recommendations","year":2025,"lang":"en","type":"article","venue":"Cochrane Evidence Synthesis and Methods","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Public Health Agency of Canada; Canadian Agency for Drugs and Technologies in Health","funders":"","keywords":"Leverage (statistics); Agency (philosophy); Automation; Information access; Information system; Focus (optics)","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5434167,0.004466051,0.01451882,0.05755179,0.00294019,0.02132914,0.01037959,0.01065427,0.01701196],"category_scores_gemma":[0.7944948,0.005166974,0.02221752,0.04252778,0.005646571,0.02322231,0.01058562,0.008826071,0.002310247],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.03289887,"about_ca_system_score_gemma":0.06701157,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02234984,"about_ca_topic_score_gemma":0.03843734,"domain_scores_codex":[0.3601515,0.4203703,0.1667434,0.007715577,0.04156524,0.003453903],"domain_scores_gemma":[0.189787,0.653183,0.05497453,0.02500495,0.07388275,0.003167769],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.002051485,0.000324798,0.003360971,0.6282439,0.02176848,0.000168485,0.003357016,0.001541025,0.0004222436,0.01561072,0.02193579,0.3012151],"study_design_scores_gemma":[0.002427694,0.0005484923,0.002824289,0.90628,0.0312859,0.0001149646,0.002935458,0.002381747,0.0006463244,0.01795589,0.03235086,0.0002484158],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.00842589,0.6815293,0.06590431,0.1491202,0.009555088,0.0653085,0.006724999,0.001072805,0.01235889],"genre_scores_gemma":[0.09288192,0.2389354,0.5339437,0.03338864,0.00153065,0.09318811,0.004200036,0.0004722761,0.001459405],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4565833,"threshold_uncertainty_score":0.5630486,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4332274836753609,"score_gpt":0.6187690955254639,"score_spread":0.185541611850103,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}