{"id":"W4416014008","doi":"10.1609/aiide.v21i1.36855","title":"Game-Based Platforms for Studying Virtual Agent Believability","year":2025,"lang":"","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment","topic":"Social Robot Interaction and HRI","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Simon Fraser University","funders":"","keywords":"Virtual agent; Perception; Relevance (law); Presentation (obstetrics); Virtual machine; Reliability (semiconductor); Virtual reality; Adaptation (eye)","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":[0.002537546,0.0008730631,0.0003034122,0.001503689,0.0008466011,0.002371895,0.001134531,0.0007542752,0.005723421],"category_scores_gemma":[0.01628591,0.0005214976,0.0006448089,0.000718873,0.001122535,0.002264495,0.002262133,0.001400748,0.0006395059],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008087106,"about_ca_system_score_gemma":0.0005456559,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001178451,"about_ca_topic_score_gemma":0.002178472,"domain_scores_codex":[0.997552,0.001610643,0.0001599144,0.0001751442,0.000384303,0.0001179812],"domain_scores_gemma":[0.9886959,0.007928068,0.001067578,0.0009396841,0.0007231834,0.0006456301],"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":[0.003804541,0.009748764,0.1298356,0.004665565,0.0008319434,0.001372668,0.05157771,0.0396519,0.1251464,0.2872803,0.01477956,0.3313051],"study_design_scores_gemma":[0.00165141,0.0233042,0.2201779,0.001869932,0.000819694,0.00248018,0.03216199,0.2443563,0.06259441,0.2345547,0.1751291,0.0009002843],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6810744,0.0006989274,0.2717142,0.000813932,0.000227379,0.004876883,0.001389883,0.0006827126,0.03852163],"genre_scores_gemma":[0.8068536,0.0003755119,0.1829847,0.0002067037,0.00004095264,0.006019793,0.0004856834,0.0001053216,0.002927736],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.005723421,"threshold_uncertainty_score":0.01914674,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09585965332053883,"score_gpt":0.3712859447033267,"score_spread":0.2754262913827878,"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."}}