{"id":"W4402748869","doi":"10.1038/s41591-024-03258-2","title":"A toolbox for surfacing health equity harms and biases in large language models","year":2024,"lang":"en","type":"article","venue":"Nature Medicine","topic":"Risk Perception and Management","field":"Social Sciences","cited_by":81,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"National Institute of Biomedical Imaging and Bioengineering; Google; Fogarty International Center; National Institutes of Health; National Science Foundation","keywords":"Toolbox; Equity (law); Health equity; Computer science; Business; Data science; Actuarial science; Economics; Political science; Health care; Programming language; Economic growth","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04958851,0.001822072,0.0008638984,0.004396148,0.001624298,0.006561044,0.003063844,0.003092259,0.01199422],"category_scores_gemma":[0.2053649,0.000968112,0.001993019,0.001838284,0.00317036,0.008082793,0.01211799,0.004870016,0.002692249],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002629315,"about_ca_system_score_gemma":0.00399294,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003548497,"about_ca_topic_score_gemma":0.009496817,"domain_scores_codex":[0.9588252,0.03255075,0.00209821,0.002364721,0.003828096,0.00033308],"domain_scores_gemma":[0.8526655,0.1139453,0.00499704,0.02001184,0.006977877,0.00140243],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0006978051,0.000580689,0.02193952,0.002508585,0.0005741154,0.0006861688,0.01155369,0.08372994,0.008808151,0.2741272,0.07161985,0.5231743],"study_design_scores_gemma":[0.0001862406,0.0002213759,0.00217913,0.0009548353,0.00008221542,0.0003442689,0.001902629,0.3894475,0.006599646,0.5165958,0.08130445,0.000181892],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01426653,0.001013302,0.9534582,0.006754437,0.0001906721,0.0008029753,0.004350927,0.009449854,0.009713111],"genre_scores_gemma":[0.1399678,0.0003984318,0.8495782,0.001507901,0.0001139711,0.001537153,0.003810754,0.0009715388,0.002114205],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.9504115,"threshold_uncertainty_score":0.2622522,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07475341040796825,"score_gpt":0.4734156035758205,"score_spread":0.3986621931678522,"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."}}