{"id":"W6968549152","doi":"10.5281/zenodo.15285505","title":"Interpretable Concept Bottlenecks to Align Reinforcement Learning Agents","year":2024,"lang":"en","type":"article","venue":"Zenodo (CERN European Organization for Nuclear Research)","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"","keywords":"Reinforcement learning; Control (management); Key (lock); Process (computing); Action (physics); Matching (statistics)","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":["sts","scholarly_communication","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.0007806115,0.0001562461,0.0001314919,0.0003160319,0.001396238,0.002828607,0.00192994,0.00004951485,0.00770073],"category_scores_gemma":[0.0005042306,0.0001656117,0.00005998759,0.001085321,0.00007112409,0.0008293196,0.002328253,0.000294217,0.02451608],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002516987,"about_ca_system_score_gemma":0.000008449595,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003623423,"about_ca_topic_score_gemma":2.490944e-7,"domain_scores_codex":[0.9978858,0.0002257325,0.0003033374,0.0006039344,0.0004623263,0.0005188878],"domain_scores_gemma":[0.9987614,0.00004074691,0.00004797995,0.0005566066,0.0003388953,0.0002544208],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002447373,0.00006760986,0.000001215208,0.00006849305,0.00007334715,0.0001136643,0.01739933,0.02822369,0.00867821,0.1393614,0.4085723,0.3974163],"study_design_scores_gemma":[0.00005619757,0.0003562448,0.000007606729,0.00008269808,0.000004634226,0.00003911227,0.000389118,0.09480134,0.0076023,0.0004703845,0.8960051,0.0001852373],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002015012,0.0001185422,0.8521657,0.001676512,0.0004058971,0.0004387796,0.000009239499,0.002090024,0.1410803],"genre_scores_gemma":[0.9893811,0.00003794123,0.001485435,0.0004972625,0.0001207511,1.472013e-7,0.0001235608,0.0005543896,0.007799441],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9873661,"threshold_uncertainty_score":0.9999038,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04140303531784949,"score_gpt":0.2817263463880848,"score_spread":0.2403233110702353,"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."}}