{"id":"W4392350413","doi":"10.52609/jmlph.v4i1.113","title":"Evaluating the Precision of ChatGPT Artificial Intelligence in Emergency Differential Diagnosis","year":2024,"lang":"en","type":"article","venue":"The Journal of Medicine Law & Public Health","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medical diagnosis; Triage; Artificial intelligence; Differential diagnosis; Computer science; Differential (mechanical device); Human intelligence; Machine learning; Medical emergency; Medicine; Engineering; Pathology","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02915905,0.000353442,0.0006682433,0.003862743,0.0006907778,0.00182443,0.001056029,0.001046979,0.001547751],"category_scores_gemma":[0.2201605,0.0002534022,0.0006295434,0.002300688,0.001016683,0.001756192,0.002473857,0.0008247434,0.0006671631],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001460127,"about_ca_system_score_gemma":0.001499678,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004190054,"about_ca_topic_score_gemma":0.004179207,"domain_scores_codex":[0.9705704,0.01857671,0.004000362,0.001844905,0.004426313,0.0005813932],"domain_scores_gemma":[0.6709272,0.2705913,0.01739966,0.00881347,0.02925318,0.003015115],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.004775933,0.000963587,0.7438846,0.001795432,0.0002369484,0.0008679676,0.01914978,0.00394325,0.001800493,0.0007076873,0.004220447,0.2176538],"study_design_scores_gemma":[0.0003960053,0.008099286,0.9106768,0.001326495,0.0005351726,0.003325161,0.0207628,0.03195602,0.007778796,0.001264284,0.01368158,0.000197672],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9886265,0.001196466,0.002948553,0.000714026,0.0001128115,0.0005434596,0.0007366179,0.0001714208,0.004950048],"genre_scores_gemma":[0.9930937,0.0003261882,0.005320008,0.0001415372,0.00004996856,0.0001830476,0.0004406767,0.00001371002,0.0004311683],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02915905,"threshold_uncertainty_score":0.1542097,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4049069743792399,"score_gpt":0.5374027853009822,"score_spread":0.1324958109217422,"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."}}