{"id":"W4390537026","doi":"10.1007/s43681-023-00397-z","title":"AITA: AI trustworthiness assessment","year":2024,"lang":"en","type":"article","venue":"AI and Ethics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":false,"ca_institutions":"Polytechnique Montréal","funders":"","keywords":"Trustworthiness; Computer science; Psychology; Computer security","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009600389,0.001232449,0.0008019659,0.005549954,0.001228178,0.006324608,0.001531957,0.001450385,0.0258849],"category_scores_gemma":[0.0886412,0.0006863526,0.001220079,0.002679214,0.00095674,0.006153252,0.003954158,0.003139648,0.01086956],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001286615,"about_ca_system_score_gemma":0.00407363,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002558031,"about_ca_topic_score_gemma":0.002191057,"domain_scores_codex":[0.9867361,0.004885417,0.0012238,0.001045721,0.005756873,0.0003520764],"domain_scores_gemma":[0.9541078,0.02295638,0.002969416,0.005684347,0.01278914,0.001493011],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0006756666,0.000434427,0.01308617,0.001525889,0.0004896448,0.0003437227,0.001397313,0.01314697,0.003411903,0.1604545,0.2107642,0.5942696],"study_design_scores_gemma":[0.0002609171,0.0003624124,0.007482859,0.0008888489,0.0003586889,0.0009851306,0.0009426683,0.306672,0.01076265,0.3751495,0.2959087,0.0002256909],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.01829472,0.001172909,0.8364435,0.002392785,0.0009866409,0.001632935,0.008525464,0.05688957,0.07366141],"genre_scores_gemma":[0.2511089,0.0008505963,0.699822,0.0006868154,0.0004327275,0.002284935,0.01468624,0.005858386,0.02426937],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.0258849,"threshold_uncertainty_score":0.08659363,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2466775578676759,"score_gpt":0.5558212560243472,"score_spread":0.3091436981566713,"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."}}