{"id":"W2316009442","doi":"10.1177/154193120304700147","title":"Calibration of Confidence in Situation Awareness Queries","year":2003,"lang":"en","type":"article","venue":"Proceedings of the Human Factors and Ergonomics Society Annual Meeting","topic":"Human-Automation Interaction and Safety","field":"Psychology","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo; Defence Research and Development Canada; Canadian Armed Forces","funders":"","keywords":"Overconfidence effect; Air traffic control; Confidence interval; Perception; Workload; Aviation; Task (project management); Computer science; Psychology; Calibration; Simulation; Social psychology; Statistics; Engineering; Mathematics","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":[0.002889308,0.0005280788,0.0004202154,0.0005148904,0.0001753711,0.001111241,0.0004748059,0.0007683471,0.001056499],"category_scores_gemma":[0.08164881,0.0003922115,0.0002382369,0.0002654225,0.0003462749,0.001115208,0.0009616178,0.0007622453,0.0002152357],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000302481,"about_ca_system_score_gemma":0.000241894,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007102246,"about_ca_topic_score_gemma":0.0004190541,"domain_scores_codex":[0.9966127,0.001191114,0.0003336752,0.0006231784,0.001055012,0.000184199],"domain_scores_gemma":[0.9471826,0.03803673,0.006296588,0.003922124,0.003424541,0.001137474],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01387688,0.002222,0.4406779,0.0007685789,0.0005360142,0.0004848596,0.01096795,0.01861501,0.3447361,0.001651909,0.001022762,0.1644401],"study_design_scores_gemma":[0.0003437171,0.006556659,0.8863888,0.00008429191,0.0001874517,0.000701924,0.001298188,0.05461691,0.04626497,0.001942338,0.001385716,0.0002290215],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9915919,0.00007060541,0.007106403,0.00002861258,0.00001082952,0.00005347787,0.00004933355,0.00006608431,0.001022561],"genre_scores_gemma":[0.9980114,0.0000186071,0.001662853,0.00002200188,0.00000602101,0.00004299012,0.00009032089,0.00001122937,0.0001345789],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.002889308,"threshold_uncertainty_score":0.01528031,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02761251143653697,"score_gpt":0.3044137519517995,"score_spread":0.2768012405152625,"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."}}