{"id":"W4390604658","doi":"10.1073/pnas.2304406120","title":"Impossibility theorems for feature attribution","year":2024,"lang":"en","type":"article","venue":"Proceedings of the National Academy of Sciences","topic":"Explainable Artificial Intelligence (XAI)","field":"Computer Science","cited_by":88,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Interpretability; Spurious relationship; Computer science; Feature (linguistics); Attribution; Impossibility; Task (project management); Simple (philosophy); Field (mathematics); Artificial intelligence; Machine learning; Theoretical computer science; Mathematics; Psychology","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":[0.02224123,0.001793482,0.001757084,0.002965678,0.002929997,0.004811173,0.003952444,0.005428411,0.0100464],"category_scores_gemma":[0.1131402,0.001252176,0.004433489,0.001463609,0.01488499,0.01409505,0.008905073,0.01069216,0.001244219],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002017912,"about_ca_system_score_gemma":0.001442023,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008017879,"about_ca_topic_score_gemma":0.0007142805,"domain_scores_codex":[0.9805372,0.009320838,0.001272793,0.004655444,0.003234,0.0009797051],"domain_scores_gemma":[0.8643887,0.1103137,0.004011571,0.01682046,0.003239285,0.001226348],"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":[0.00004009061,0.00003165983,0.0009626675,0.0001582895,0.00006754632,0.000161725,0.0003804411,0.006921298,0.0002121406,0.97823,0.002549926,0.01028428],"study_design_scores_gemma":[0.0000148369,0.00001023687,0.000113264,0.0000365362,0.00001151816,0.00007521448,0.00002969443,0.01796254,0.0001957315,0.9805405,0.0009975805,0.00001247526],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.02284993,0.0006022546,0.9453617,0.007035542,0.0002102609,0.0001667858,0.0004799427,0.0008183876,0.0224752],"genre_scores_gemma":[0.6706937,0.0007347434,0.3172348,0.002654453,0.0009271415,0.001074593,0.0009868692,0.0005300259,0.00516365],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.02224123,"threshold_uncertainty_score":0.1176243,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06421974591361419,"score_gpt":0.3520020040083907,"score_spread":0.2877822580947765,"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."}}