{"id":"W1992252434","doi":"10.1371/journal.pone.0007263","title":"Performance of Humans vs. Exploration Algorithms on the Tower of London Test","year":2009,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Cognitive Abilities and Testing","field":"Psychology","cited_by":12,"is_retracted":false,"has_abstract":true,"ca_institutions":"Centre for Interdisciplinary Research in Rehabilitation; École de Technologie Supérieure; Institut Universitaire de Gériatrie de Montréal","funders":"","keywords":"Computer science; Test (biology); Task (project management); Algorithm; Focus (optics); Bounded function; Execution time; Machine learning; Artificial intelligence; Mathematics; Parallel computing; Biology","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":[],"consensus_categories":[],"category_scores_codex":[0.0001651316,0.00006366881,0.0001275722,0.00003392624,0.00004709428,0.000003947453,0.00008280187,0.00002960891,0.0006289885],"category_scores_gemma":[0.0002121101,0.00004737344,0.00002790064,0.00009725679,0.00005237383,0.0000574119,0.00001114608,0.00008943903,0.00003131146],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000008821283,"about_ca_system_score_gemma":0.000007639694,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002555972,"about_ca_topic_score_gemma":0.00000301316,"domain_scores_codex":[0.9994084,0.00002854631,0.0001821042,0.000112716,0.0001480538,0.0001201492],"domain_scores_gemma":[0.999155,0.0003930498,0.00009775978,0.0002082658,0.0001322826,0.00001368083],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001215136,0.03963204,0.34823,0.0004527101,0.0006893861,0.00001471536,0.04533606,0.00005420679,0.2616132,0.02242065,0.004043975,0.2762979],"study_design_scores_gemma":[0.001421915,0.009984491,0.8134134,0.0009244546,0.000170473,0.000001698757,0.002864107,0.002599127,0.1655296,0.002593627,0.0001571296,0.0003398882],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9623585,0.00003905359,0.00001834324,0.0007000819,0.00002649754,0.0001801785,0.000009856988,0.00001763563,0.0366499],"genre_scores_gemma":[0.9985486,0.00001570193,0.0001956459,0.0002123462,0.00007620899,0.00001646302,0.000005083018,0.000005936652,0.0009240538],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4651835,"threshold_uncertainty_score":0.6886983,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1000508283801219,"score_gpt":0.2827535394975817,"score_spread":0.1827027111174598,"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."}}