{"id":"W4399364360","doi":"10.1145/3630106.3658948","title":"Visibility into AI Agents","year":2024,"lang":"en","type":"article","venue":"","topic":"Blockchain Technology Applications and Security","field":"Computer Science","cited_by":52,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; Mila - Quebec Artificial Intelligence Institute","funders":"","keywords":"Computer science; Visibility; Accountability; Software deployment; Computer security; Risk analysis (engineering); Key (lock); Identifier; Corporate governance; Intrusiveness; Delegation; Process management; Implementation; Supply chain; Knowledge management; Business; Software engineering","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":[0.01239408,0.0005400263,0.000567966,0.002056444,0.001931915,0.006807251,0.001237155,0.001764546,0.007195239],"category_scores_gemma":[0.08181301,0.0006420286,0.0004362924,0.00174125,0.00388776,0.01207174,0.006012283,0.002589767,0.001206086],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001595218,"about_ca_system_score_gemma":0.002478643,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00150506,"about_ca_topic_score_gemma":0.001239625,"domain_scores_codex":[0.9836591,0.008689507,0.001228555,0.001727274,0.003867482,0.0008280298],"domain_scores_gemma":[0.8654122,0.06839023,0.02055819,0.03021719,0.01274589,0.002676408],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0006421584,0.0002641949,0.05628987,0.0007992804,0.0001443852,0.0004669922,0.01294835,0.01816837,0.01526158,0.5462362,0.006182601,0.342596],"study_design_scores_gemma":[0.0001652924,0.001055305,0.03469511,0.001308436,0.0002911969,0.00159263,0.007069316,0.08345673,0.03313555,0.5767641,0.2602246,0.0002416938],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.282194,0.002014803,0.5395922,0.007580565,0.0003384081,0.0006408558,0.0005479094,0.00267508,0.1644162],"genre_scores_gemma":[0.9475823,0.000473876,0.04461204,0.0004416677,0.0001198187,0.0002116325,0.0002206942,0.000163059,0.006174865],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.01239408,"threshold_uncertainty_score":0.06554687,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01082722238158726,"score_gpt":0.2877113813373082,"score_spread":0.276884158955721,"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."}}