{"id":"W4416529406","doi":"10.1186/s41077-025-00396-6","title":"Ability of AI detection tools and humans to accurately identify different forms of AI-generated written content","year":2025,"lang":"en","type":"article","venue":"Advances in Simulation","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; Centre Hospitalier Universitaire de Sherbrooke; Alberta Children's Hospital","funders":"","keywords":"Reliability (semiconductor); Content (measure theory); Content analysis; Health services research","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":[0.0001933601,0.00007927813,0.0002242335,0.0001672353,0.00004005085,0.000009465215,0.00003407569,0.0000704249,0.00001157049],"category_scores_gemma":[0.000500952,0.00006725958,0.00003040232,0.000298905,0.00004546242,0.0003390344,0.00001634461,0.000112418,9.463552e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001081235,"about_ca_system_score_gemma":0.00004491182,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002032815,"about_ca_topic_score_gemma":0.0009069731,"domain_scores_codex":[0.9989744,0.00003909322,0.0005770476,0.0001792441,0.000122942,0.0001073113],"domain_scores_gemma":[0.999103,0.0002230456,0.0001027245,0.0001550628,0.0003794934,0.00003673288],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000563462,0.0001661266,0.5003901,0.0004294447,0.00001259299,2.787344e-7,0.001055866,0.02777314,0.08054656,0.0006885302,0.000002795411,0.3883711],"study_design_scores_gemma":[0.0001584511,0.0002793688,0.6433201,0.000264254,0.00002452868,2.489636e-7,0.0007925595,0.01551921,0.3307849,0.008635323,0.0001549159,0.00006617187],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9720364,0.0003787797,0.02607116,0.0005356767,0.0002837886,0.0006145631,0.00000553803,0.00001249492,0.00006166117],"genre_scores_gemma":[0.9994783,0.00009938854,0.00006993885,0.0002390729,0.00003127109,0.00003023162,0.00002349197,0.000004312401,0.00002397631],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.388305,"threshold_uncertainty_score":0.2742766,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2063734400272602,"score_gpt":0.500212117337602,"score_spread":0.2938386773103417,"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."}}