{"id":"W4380575774","doi":"10.2196/46939","title":"Putting ChatGPT’s Medical Advice to the (Turing) Test: Survey Study","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Education","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":188,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"National Science Foundation","keywords":"Chatbot; Advice (programming); Test (biology); Turing; Turing test; Psychology; Computer science; Medicine; World Wide Web; Artificial intelligence","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":[0.0119336,0.0002820083,0.0004613019,0.001475207,0.0009457163,0.001322573,0.0006519715,0.001202243,0.002838821],"category_scores_gemma":[0.0353862,0.0006181257,0.000508755,0.001340476,0.001199933,0.002046939,0.001572706,0.001400454,0.001931227],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00114595,"about_ca_system_score_gemma":0.001193225,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002898324,"about_ca_topic_score_gemma":0.00263556,"domain_scores_codex":[0.9904444,0.0059068,0.0008253406,0.0006343707,0.001320683,0.000868402],"domain_scores_gemma":[0.9763589,0.01132325,0.003696817,0.001420575,0.004443396,0.002757097],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000483011,0.00225148,0.8705977,0.0004472143,0.00009890729,0.0006726343,0.06280661,0.0002608115,0.0008313981,0.0005475056,0.01336881,0.04763385],"study_design_scores_gemma":[0.0001114017,0.004111656,0.9229431,0.0002098081,0.00004704729,0.001226112,0.0507862,0.00211665,0.0007199986,0.0002660384,0.01737331,0.00008854643],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961875,0.00009829948,0.0004906936,0.0005522901,0.0000134179,0.000325823,0.0004598142,0.00002490233,0.001847196],"genre_scores_gemma":[0.9966629,0.0001944329,0.000635734,0.0008803461,0.00002154246,0.0006672246,0.0003573147,0.00001837876,0.0005620932],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0119336,"threshold_uncertainty_score":0.06311166,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1311681862863586,"score_gpt":0.5029894972674881,"score_spread":0.3718213109811295,"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."}}