{"id":"W3172982719","doi":"10.1109/lra.2021.3083465","title":"Expectations Vs. Reality: Unreliability and Transparency in a Treasure Hunt Game With Icub","year":2021,"lang":"en","type":"article","venue":"IEEE Robotics and Automation Letters","topic":"Artificial Intelligence in Games","field":"Computer Science","cited_by":16,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"European Commission","keywords":"iCub; Transparency (behavior); Affect (linguistics); Robot; Treasure; Humanoid robot; Computer science; Human–computer interaction; Psychology; Quality (philosophy); Simulation; Artificial intelligence; Computer security; Communication; Geography","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":[0.003188743,0.0005917001,0.00054335,0.0002806853,0.0008023339,0.00236187,0.0006218316,0.001222915,0.002817938],"category_scores_gemma":[0.03145745,0.0003493789,0.0002789299,0.0001185888,0.002030199,0.001945454,0.001971216,0.001827122,0.0002461167],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009036856,"about_ca_system_score_gemma":0.0003864976,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002502996,"about_ca_topic_score_gemma":0.002043212,"domain_scores_codex":[0.9964861,0.002286008,0.0001475096,0.0002915603,0.0005316601,0.0002571146],"domain_scores_gemma":[0.9765822,0.01804689,0.002365444,0.000996431,0.0004950198,0.001513971],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.02759987,0.01661603,0.3127157,0.002290634,0.000966783,0.0072493,0.268811,0.04711853,0.1602905,0.04025527,0.006416341,0.1096699],"study_design_scores_gemma":[0.002540269,0.02748245,0.4206751,0.0004995018,0.0006282819,0.002561839,0.1105368,0.335508,0.02546371,0.05645049,0.01644599,0.001207613],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9961075,0.00001723937,0.001938516,0.0001162965,0.000009222327,0.00003476857,0.00001458573,0.00002073121,0.001740994],"genre_scores_gemma":[0.9986572,0.000007586048,0.0009548488,0.00004584463,0.000002183676,0.0000357795,0.00001401651,0.000006416013,0.0002762489],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.003188743,"threshold_uncertainty_score":0.01686388,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02469985435777006,"score_gpt":0.261043406057689,"score_spread":0.2363435516999189,"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."}}