{"id":"W3204942516","doi":"10.1609/aiide.v17i1.18894","title":"Explaining Deep Reinforcement Learning Agents in the Atari Domain through a Surrogate Model","year":2021,"lang":"en","type":"article","venue":"Proceedings of the AAAI Conference on Artificial Intelligence and Interactive Digital Entertainment","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":9,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"Natural Sciences and Engineering Research Council of Canada; Alberta Machine Intelligence Institute","keywords":"Reinforcement learning; Artificial intelligence; Computer science; Suite; Domain (mathematical analysis); Transformation (genetics); Replicate; Surrogate model; Representation (politics); Deep learning; Machine learning; Mathematics; Law","routes":{"ca_aff":true,"ca_fund":true,"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":["metaepi_narrow","scholarly_communication"],"consensus_categories":[],"category_scores_codex":[0.0005851371,0.0003705223,0.0003566617,0.0001275329,0.0003331081,0.00113913,0.001560516,0.00008275819,0.00004145728],"category_scores_gemma":[0.0005554006,0.0002620255,0.0001806508,0.0006378042,0.0002487722,0.002340922,0.000913917,0.0006049118,0.00004433403],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001893018,"about_ca_system_score_gemma":0.0001012007,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005677059,"about_ca_topic_score_gemma":0.00003870727,"domain_scores_codex":[0.9969544,0.00006594676,0.000863683,0.0007557895,0.0007687798,0.0005913739],"domain_scores_gemma":[0.9983215,0.0002806459,0.0004390439,0.0003825388,0.0004886513,0.00008759725],"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":[0.0001899083,0.0005333663,0.0006179477,0.00004679353,0.00006278507,0.00002054999,0.05610255,0.01348206,0.004780481,0.893064,0.00005964245,0.03103993],"study_design_scores_gemma":[0.00009506739,0.0004374222,0.00007677315,0.0005289183,0.00001281013,0.00002761981,0.05547073,0.5702553,0.2000228,0.1723101,0.0003677362,0.0003946144],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.6087757,0.0000705208,0.3185605,0.007415887,0.0004466574,0.001106647,0.000005778237,0.00008283298,0.06353548],"genre_scores_gemma":[0.997676,0.0001073174,0.0008163754,0.0009229748,0.00003142947,0.00009175948,0.000004418082,0.00001617169,0.0003335874],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7207538,"threshold_uncertainty_score":0.9999832,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.07958754341074467,"score_gpt":0.3097414533509233,"score_spread":0.2301539099401787,"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."}}