{"id":"W4409258158","doi":"10.1613/jair.1.16665","title":"Detecting AI-Generated Text: Factors Influencing Detectability with Current Methods","year":2025,"lang":"en","type":"article","venue":"Journal of Artificial Intelligence Research","topic":"Topic Modeling","field":"Computer Science","cited_by":40,"is_retracted":false,"has_abstract":true,"ca_institutions":"National Research Council Canada","funders":"Ministère de la Défense Nationale","keywords":"Current (fluid); Computer science; Artificial intelligence; Physics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.008975294,0.0001781402,0.0003485642,0.0009583941,0.0004507625,0.0005179428,0.001529566,0.00009549037,0.0000194739],"category_scores_gemma":[0.002589159,0.0001306269,0.0001243203,0.002561848,0.0001803074,0.0007245133,0.0004093162,0.001757587,0.00000745213],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003955972,"about_ca_system_score_gemma":0.001145746,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00013104,"about_ca_topic_score_gemma":0.000102032,"domain_scores_codex":[0.9957563,0.001130927,0.001035487,0.0004250958,0.001013752,0.0006384068],"domain_scores_gemma":[0.9952238,0.001424943,0.0002601029,0.0006019783,0.002282967,0.000206182],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00004944531,0.00009655602,0.003148183,0.00005027706,0.00003906231,0.00001640292,0.001930785,0.01041003,0.0443195,0.007041025,0.00000938407,0.9328893],"study_design_scores_gemma":[0.00004002972,0.0003056207,0.0005420648,0.0002673373,0.00001032639,0.00001921993,0.001589063,0.2578182,0.7080408,0.03087657,0.0003180967,0.0001727026],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3859554,0.0003538092,0.6127303,0.0003515363,0.0004148271,0.0001182336,1.942271e-7,0.00002309378,0.00005262137],"genre_scores_gemma":[0.8724082,0.00002741606,0.1274147,0.00002253703,0.0001037128,0.000004198163,6.919904e-8,0.000008009612,0.00001125076],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9327167,"threshold_uncertainty_score":0.763594,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2192959255969502,"score_gpt":0.498691993893544,"score_spread":0.2793960682965937,"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."}}