{"id":"W4398183239","doi":"10.1038/s41746-024-01127-3","title":"Generalization—a key challenge for responsible AI in patient-facing clinical applications","year":2024,"lang":"en","type":"article","venue":"npj Digital Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":103,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"GlaxoSmithKline foundation; Cystic Fibrosis Trust","keywords":"Generalization; Key (lock); Computer science; Artificial intelligence; Bioethics; Machine learning; Data science; Health care; Computer security; Political science; Epistemology","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":[0.0004180981,0.0001235168,0.0002736076,0.0002775491,0.00006222273,0.0000380271,0.00006267276,0.0001339693,0.00007198113],"category_scores_gemma":[0.0009926913,0.0001002415,0.00007167167,0.0004958448,0.0001115482,0.0002219736,0.00001734811,0.000230569,0.0001082738],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000111271,"about_ca_system_score_gemma":0.000378762,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007030158,"about_ca_topic_score_gemma":0.00005601386,"domain_scores_codex":[0.9983455,0.00002670567,0.0008018284,0.0003631002,0.0002135267,0.0002493466],"domain_scores_gemma":[0.9986919,0.0006032221,0.00005161566,0.0002535022,0.0002164052,0.0001833865],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002171291,0.0003032053,0.006020563,0.000491964,0.00003325416,0.0000248313,0.003164357,0.00001731978,0.0003521287,0.02049654,0.01630992,0.9525688],"study_design_scores_gemma":[0.0003595961,0.002500319,0.001840137,0.001702297,0.00009228316,0.00005903432,0.00309942,0.007995756,0.001103809,0.03677939,0.9441839,0.0002840641],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6166777,0.01331982,0.07934923,0.2415406,0.005159179,0.00669138,0.0001003521,0.0006874455,0.0364743],"genre_scores_gemma":[0.9936781,0.00038854,0.0004282342,0.002117738,0.001652435,0.0003835561,0.0002828206,0.00003035868,0.001038235],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9522848,"threshold_uncertainty_score":0.408773,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2320941903222258,"score_gpt":0.5085691543340696,"score_spread":0.2764749640118437,"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."}}