{"id":"W4387184490","doi":"10.31234/osf.io/mjx2v","title":"Illusions of Confidence in Artificial Systems","year":2023,"lang":"en","type":"preprint","venue":"","topic":"Language and cultural evolution","field":"Social Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Engineering and Physical Sciences Research Council; Leverhulme Trust; UK Research and Innovation; Wellcome Trust; Canadian Institute for Advanced Research","keywords":"Metacognition; Illusion; Psychology; Perception; Low Confidence; Attribution; Cognitive psychology; Cognition; Self-confidence; Social psychology; Artificial intelligence; Computer science","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.000521275,0.00005954337,0.0001594037,0.00005850505,0.00008128155,0.00003483027,0.0002028732,0.0002025445,0.0001971951],"category_scores_gemma":[0.0002696727,0.00004739592,0.00005251879,0.0001923841,0.0001089913,0.00004304496,0.0001629725,0.000159476,0.00009198733],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007951935,"about_ca_system_score_gemma":0.0002142796,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.1917339,"about_ca_topic_score_gemma":0.09372503,"domain_scores_codex":[0.9990467,0.0001317665,0.0002390553,0.0001606629,0.0002786109,0.0001431684],"domain_scores_gemma":[0.9995879,0.00006195156,0.00009649775,0.0001263454,0.00009202136,0.00003529597],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.000009778832,0.00006259173,0.00107919,0.0001802512,0.0000218695,0.00002827296,0.08134792,0.002394316,0.001566872,0.9055948,0.006134616,0.001579552],"study_design_scores_gemma":[0.000382268,0.00007249328,0.02493189,0.003353741,0.0001053534,0.000001796149,0.7534206,0.005907944,0.0008698864,0.1896704,0.01974315,0.001540511],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.810712,0.00122516,0.001269671,0.004537154,0.007312348,0.00177865,0.0000750047,0.0004801518,0.1726098],"genre_scores_gemma":[0.9872875,0.0001409716,0.00002700967,0.00001268771,0.0002339938,0.0000223701,0.00001249663,0.000003723548,0.0122593],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7159244,"threshold_uncertainty_score":0.9228122,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1289611583689716,"score_gpt":0.3816784506817267,"score_spread":0.2527172923127551,"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."}}