{"id":"W4414183195","doi":"10.1038/s42256-025-01109-4","title":"Aligning generalization between humans and machines","year":2025,"lang":"en","type":"article","venue":"Nature Machine Intelligence","topic":"Reinforcement Learning in Robotics","field":"Computer Science","cited_by":11,"is_retracted":false,"has_abstract":false,"ca_institutions":"Vector Institute; University of British Columbia","funders":"Nederlandse Organisatie voor Wetenschappelijk Onderzoek","keywords":"Generalization; Abstraction; Cognition; Generative grammar; Key (lock); Human 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002625161,0.0004032358,0.0005383164,0.0004026197,0.0003371851,0.001084559,0.0008915106,0.001100018,0.003126814],"category_scores_gemma":[0.0142094,0.0003601982,0.0004241841,0.000309172,0.001462716,0.002997011,0.001921123,0.001933842,0.0006892026],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007939042,"about_ca_system_score_gemma":0.0007799389,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002424091,"about_ca_topic_score_gemma":0.002269272,"domain_scores_codex":[0.9988676,0.0004960618,0.00004538287,0.0003576782,0.0001457494,0.00008757869],"domain_scores_gemma":[0.9938263,0.003223249,0.000421174,0.001761162,0.0005643748,0.0002036093],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0004849,0.0003809365,0.01733518,0.0002913213,0.0003534855,0.0002724484,0.001296705,0.3726282,0.02322149,0.09569973,0.007492098,0.4805435],"study_design_scores_gemma":[0.00004209423,0.0002216321,0.006203203,0.00003431034,0.0000473039,0.0001216185,0.0002667333,0.7551463,0.005067538,0.2288994,0.003917234,0.00003254255],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3731948,0.0007812868,0.5964955,0.003031842,0.0002882336,0.0001332652,0.0001683598,0.002091637,0.02381493],"genre_scores_gemma":[0.9628608,0.0001228058,0.03437603,0.0002586508,0.00003514598,0.00003984274,0.0001018436,0.00009059846,0.00211424],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.003126814,"threshold_uncertainty_score":0.01388335,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.00935696220116273,"score_gpt":0.297988163620978,"score_spread":0.2886312014198152,"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."}}