{"id":"W3214288572","doi":"10.1080/13546783.2021.1999327","title":"Investigating lay evaluations of models","year":2021,"lang":"en","type":"article","venue":"Thinking & Reasoning","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University","funders":"","keywords":"Overfitting; Distrust; Computer science; Cognitive psychology; Mean squared prediction error; Mean squared error; Machine learning; Psychology; Function (biology); Econometrics; Artificial intelligence; Statistics; Mathematics; Artificial neural network","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":[0.02638787,0.0008062787,0.0004066475,0.001821069,0.001154869,0.005990946,0.000926666,0.001991208,0.005032013],"category_scores_gemma":[0.2210287,0.0005046646,0.0009014738,0.0005391347,0.003906877,0.006565655,0.002994403,0.002474402,0.000558911],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001373812,"about_ca_system_score_gemma":0.0006138688,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002422007,"about_ca_topic_score_gemma":0.001384764,"domain_scores_codex":[0.9808359,0.01370668,0.0006000296,0.001150984,0.00327991,0.0004265486],"domain_scores_gemma":[0.8116134,0.1511018,0.01377825,0.0114555,0.01082206,0.001228919],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.002448055,0.0004164785,0.1924009,0.001846562,0.0006281912,0.001517767,0.5118659,0.01525751,0.02182634,0.07509713,0.006789158,0.1699059],"study_design_scores_gemma":[0.0003284988,0.002749663,0.180677,0.003319792,0.0007506663,0.002615324,0.4060939,0.1009803,0.01816596,0.2184583,0.06508172,0.0007789839],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9196752,0.0009383404,0.04220827,0.002479979,0.0001197772,0.0001059054,0.0001323665,0.000117427,0.03422265],"genre_scores_gemma":[0.9946077,0.0001621974,0.004399324,0.000165407,0.00002297142,0.00003300945,0.00005942098,0.00002990891,0.0005200877],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02638787,"threshold_uncertainty_score":0.1395541,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05356244137974382,"score_gpt":0.3362210250929328,"score_spread":0.282658583713189,"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."}}