{"id":"W4391759979","doi":"10.1016/j.ijer.2024.102323","title":"Running a double-blind true social experiment with a goal oriented adaptive AI-based conversational agent in educational research","year":2024,"lang":"en","type":"article","venue":"International Journal of Educational Research","topic":"Online Learning and Analytics","field":"Computer Science","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"Université de Montréal","funders":"","keywords":"Psychology; Mathematics education; Cognitive psychology; Computer science; Applied psychology","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":[0.003659386,0.0001806491,0.0002118282,0.002878318,0.0003040906,0.0007627708,0.001470767,0.0001011799,0.0008677811],"category_scores_gemma":[0.0006211478,0.00015724,0.0001413734,0.00187326,0.0003581402,0.0008472301,0.000218792,0.00175439,0.0001846886],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00177334,"about_ca_system_score_gemma":0.01151916,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000174999,"about_ca_topic_score_gemma":0.00004023195,"domain_scores_codex":[0.9933702,0.0004190545,0.0006742685,0.000492523,0.00452342,0.0005204882],"domain_scores_gemma":[0.9910788,0.003031879,0.0001635939,0.0001946413,0.005302552,0.0002285663],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003513986,0.002689356,0.007088132,0.00004715561,0.0005118583,0.0001709292,0.007765902,0.007761563,0.0006167809,0.9444823,0.02115672,0.00419536],"study_design_scores_gemma":[0.03734022,0.005540663,0.0597622,0.006617246,0.00009054715,0.001477223,0.01627732,0.5131014,0.004361829,0.1249895,0.2282121,0.002229786],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.3886841,0.001597345,0.01867935,0.5810568,0.003469978,0.0006432264,0.00003298618,0.00002312944,0.005813077],"genre_scores_gemma":[0.9844725,0.00002031669,0.01000377,0.0001758112,0.002316673,0.00007712143,0.00004742056,0.00002296047,0.002863408],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8194928,"threshold_uncertainty_score":0.9940846,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1791202511191977,"score_gpt":0.5047955688891547,"score_spread":0.325675317769957,"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."}}