{"id":"W4405290985","doi":"10.1098/rsos.241875","title":"When children can explain why they believe a claim, they suggest a better empirical test for that claim","year":2024,"lang":"en","type":"article","venue":"Royal Society Open Science","topic":"Child and Animal Learning Development","field":"Psychology","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"Universitetet i Oslo","keywords":"Test (biology); Empirical research; Psychology; Object (grammar); Developmental psychology; Social psychology; Computer science; Mathematics; Artificial intelligence; Statistics","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":[0.004574151,0.0005501487,0.0004017092,0.0007763744,0.0002473543,0.001701123,0.0005887132,0.001076707,0.005276759],"category_scores_gemma":[0.02571538,0.0005308474,0.0005282903,0.0002807137,0.001352842,0.001799815,0.001234442,0.0009766156,0.0006874707],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004480633,"about_ca_system_score_gemma":0.0006598227,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008647945,"about_ca_topic_score_gemma":0.001955373,"domain_scores_codex":[0.9983268,0.0003705588,0.0001606901,0.0003841996,0.0005404562,0.0002172846],"domain_scores_gemma":[0.9769419,0.01141665,0.00767324,0.002120159,0.001118491,0.0007296412],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001388509,0.001617491,0.6552665,0.001155497,0.0002299231,0.004022585,0.03532614,0.001966986,0.1400191,0.01435001,0.002089183,0.1425681],"study_design_scores_gemma":[0.0001343773,0.00265286,0.8954286,0.0004368448,0.0004233536,0.003641837,0.01155086,0.003598101,0.04668467,0.01646129,0.01881518,0.0001721607],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.994459,0.0002183404,0.00222298,0.0001416581,0.000008817994,0.0000268389,0.00008325195,0.00005584615,0.002783243],"genre_scores_gemma":[0.9892267,0.0003249983,0.009176095,0.00007984526,0.00001148339,0.00006203896,0.0002120463,0.00001384606,0.000892945],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005276759,"threshold_uncertainty_score":0.02419072,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.03606854729262957,"score_gpt":0.3387746782709916,"score_spread":0.302706130978362,"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."}}