{"id":"W2402471406","doi":"","title":"Fostering Diagnostic Accuracy in a Medical Intelligent Tutoring System.","year":2013,"lang":"en","type":"article","venue":"AIED Workshops","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Intelligent tutoring system; Artificial intelligence; Human–computer interaction","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0005619175,0.0002033454,0.0004852501,0.0001275958,0.00004555473,0.00005915508,0.0001913205,0.0002540601,0.0008617553],"category_scores_gemma":[0.1526303,0.0001671764,0.0001223275,0.0003580135,0.00007039096,0.00009559831,0.000341766,0.0005798668,0.0009807105],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001766799,"about_ca_system_score_gemma":0.0001441011,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004279146,"about_ca_topic_score_gemma":0.00003054838,"domain_scores_codex":[0.9978156,0.00006975757,0.0006832823,0.0003894079,0.0005394297,0.0005025213],"domain_scores_gemma":[0.9568671,0.04171242,0.0001096747,0.0005228309,0.00009953253,0.0006884166],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0000779974,0.0005012885,0.1116928,0.0001950872,0.00007487711,0.001390583,0.0006447766,0.00009533183,0.00006172358,0.001030152,0.003899679,0.8803357],"study_design_scores_gemma":[0.00969081,0.0007504979,0.8466035,0.1076836,0.0003423639,0.001332246,0.004678754,0.02011057,0.001161689,0.001561652,0.004597227,0.00148717],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9786177,0.001139478,0.00123901,0.004369874,0.001086883,0.0009342952,0.000001207865,0.0002615152,0.01235004],"genre_scores_gemma":[0.9973158,0.0002866786,0.0003987894,0.0009720887,0.0004692129,0.0002288397,0.00001430953,0.00003407861,0.0002802574],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8788486,"threshold_uncertainty_score":0.9997972,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04059755092363861,"score_gpt":0.34766565679485,"score_spread":0.3070681058712114,"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."}}