{"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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001982316,0.00006168408,0.0001401757,0.0001167826,0.0001248887,0.00003255802,0.00007067923,0.00008462206,0.0001297207],"category_scores_gemma":[0.0002773516,0.0000454453,0.00001981061,0.0002352788,0.00006280548,0.0001229425,0.0000298482,0.000213992,0.000002646201],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001163167,"about_ca_system_score_gemma":0.0000776709,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002567742,"about_ca_topic_score_gemma":0.01471454,"domain_scores_codex":[0.999145,0.00005555026,0.000124725,0.0001773688,0.0002365001,0.0002608844],"domain_scores_gemma":[0.9996741,0.0001332825,0.00001907619,0.0000649811,0.00001822576,0.00009030163],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004155174,0.00008576192,0.0003389138,0.0006216562,0.00002966247,0.00004348767,0.3197867,0.00008131362,0.00042716,0.2596048,0.1443745,0.2745645],"study_design_scores_gemma":[0.002017947,0.0002301529,0.002868951,0.00138994,0.00003342705,0.000001809271,0.09263804,0.0208226,0.00002714416,0.009907363,0.8697971,0.0002655763],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5506691,0.0962896,0.01426032,0.254728,0.003487136,0.002712399,0.0001196951,0.0006144504,0.07711928],"genre_scores_gemma":[0.9914133,0.001665367,0.0002327489,0.003861039,0.0003458118,0.00001445439,0.00001972756,0.000006372069,0.002441166],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7254226,"threshold_uncertainty_score":0.821106,"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."}}