{"id":"W4391334942","doi":"10.1145/3630106.3659037","title":"Black-Box Access is Insufficient for Rigorous AI Audits","year":2024,"lang":"en","type":"preprint","venue":"","topic":"Adversarial Robustness in Machine Learning","field":"Computer Science","cited_by":57,"is_retracted":false,"has_abstract":true,"ca_institutions":"Mila - Quebec Artificial Intelligence Institute; University of Toronto","funders":"","keywords":"Black box; Audit; Computer science; Business; Computer security; Accounting; Artificial intelligence","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":["metaepi_narrow","scholarly_communication","open_science"],"consensus_categories":[],"category_scores_codex":[0.0006691749,0.0005495542,0.000548732,0.0003928898,0.0001987238,0.001943404,0.005059938,0.0004614261,0.0002523299],"category_scores_gemma":[0.0002329383,0.0004901103,0.0003779608,0.0005072142,0.0001009193,0.0003417326,0.01813798,0.001763517,0.000388176],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002608626,"about_ca_system_score_gemma":0.0005857387,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001270121,"about_ca_topic_score_gemma":0.00002123935,"domain_scores_codex":[0.9962175,0.0000902928,0.0005712826,0.001811874,0.0006716991,0.000637278],"domain_scores_gemma":[0.9972416,0.0002648666,0.0002725735,0.001756849,0.0002964417,0.0001676857],"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":[0.00002392524,0.0002184538,0.0001109612,0.001513231,0.0002997715,0.00008961181,0.005134565,0.5188875,0.00001966294,0.2753588,0.1143609,0.08398262],"study_design_scores_gemma":[0.0002365816,0.00005432054,0.00006311611,0.0001883166,0.0000505778,0.000006912508,0.00001911034,0.8743169,0.0004382124,0.09719174,0.02681469,0.0006195099],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.002850659,0.0002231434,0.9616169,0.0126512,0.007190969,0.00101643,0.00003439653,0.001167345,0.0132489],"genre_scores_gemma":[0.7457602,0.00004761915,0.2270221,0.007596431,0.001897213,0.0003751934,0.00005246334,0.0001669433,0.0170819],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7429095,"threshold_uncertainty_score":0.9997551,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0312995521654295,"score_gpt":0.3489467783865863,"score_spread":0.3176472262211568,"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."}}