{"id":"W4390722476","doi":"10.48550/arxiv.2401.03529","title":"Quantifying stability of non-power-seeking in artificial agents","year":2024,"lang":"en","type":"preprint","venue":"arXiv (Cornell University)","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Shutdown; Computer science; Metric (unit); Stability (learning theory); Core (optical fiber); Shut down; Power (physics); Mathematical optimization; Economics; Engineering; Mathematics; Operations management; Machine learning","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":[0.008807587,0.001172897,0.001065958,0.0020178,0.0009476453,0.002441945,0.002095385,0.002034015,0.002586994],"category_scores_gemma":[0.07574522,0.0008533177,0.001306839,0.0007794489,0.005859689,0.005587686,0.003876803,0.002855456,0.0002933756],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002899197,"about_ca_system_score_gemma":0.001478042,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002079812,"about_ca_topic_score_gemma":0.0007734472,"domain_scores_codex":[0.994155,0.002258561,0.000434992,0.001258956,0.001456383,0.000436183],"domain_scores_gemma":[0.8979132,0.07705282,0.01167658,0.006213259,0.004483449,0.002660595],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0002038562,0.00008533595,0.00764832,0.0002218245,0.0002130641,0.0002480836,0.0008496326,0.7238564,0.006064658,0.2444734,0.0004065401,0.01572884],"study_design_scores_gemma":[0.00001799269,0.000163089,0.0006849958,0.0000346878,0.00003106994,0.0000890919,0.00008453034,0.8083546,0.002773282,0.1870662,0.0006674028,0.00003306709],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.255891,0.0004818882,0.7360859,0.001518786,0.00004092225,0.0001178874,0.0001577662,0.000482043,0.005223778],"genre_scores_gemma":[0.9577084,0.0001548864,0.0406335,0.0001227128,0.00002764667,0.0001124675,0.00009642389,0.00009669471,0.00104729],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.008807587,"threshold_uncertainty_score":0.04657954,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1637298112787587,"score_gpt":0.2461775063420691,"score_spread":0.08244769506331043,"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."}}