{"id":"W4391998433","doi":"10.53841/bpsadm.2024.16.1.48","title":"Large-scale testing in the face of AI","year":2024,"lang":"en","type":"article","venue":"Assessment and Development Matters","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"First Nations University of Canada; Brock University","funders":"","keywords":"Expansive; Test (biology); Computer science; Construct (python library); Scale (ratio); Face validity; Set (abstract data type); Test design; Face (sociological concept); Construct validity; Data science; Psychology; Political science; Sociology; Psychometrics; Social science; Law","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":[0.0004140309,0.00005389057,0.00008231076,0.00007515649,0.00004325564,0.00002666369,0.0000325016,0.00002365226,0.0000331185],"category_scores_gemma":[0.000009603837,0.00003631919,0.00000997612,0.0001734213,0.0000184597,0.00005680417,0.00001375988,0.000122265,0.00001380724],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004584501,"about_ca_system_score_gemma":0.0002794418,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005617137,"about_ca_topic_score_gemma":0.00003288381,"domain_scores_codex":[0.9993963,0.00001928128,0.0002181623,0.0001100254,0.0001333437,0.000122853],"domain_scores_gemma":[0.9997171,0.000148626,0.000018231,0.00006454111,0.00002600976,0.00002550235],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001100465,0.0001385216,0.834257,0.0006353581,0.00002387709,0.0000189578,0.04548273,0.000007015284,0.0008505703,0.0004779882,0.008504148,0.1095928],"study_design_scores_gemma":[0.0001076876,0.0001486307,0.8774754,0.001060189,0.00003197925,0.0000435041,0.06707484,0.003356904,0.004934025,0.0009077934,0.04465831,0.0002007858],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.969564,0.0001738737,0.003541244,0.02530897,0.0001878885,0.00022126,6.407356e-7,0.00001993191,0.0009821871],"genre_scores_gemma":[0.9921955,0.00001522496,0.003830402,0.003790101,0.00003023629,0.00002407989,0.00001010165,0.000004804544,0.00009950352],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.109392,"threshold_uncertainty_score":0.1481054,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1316125422972995,"score_gpt":0.4430557148226341,"score_spread":0.3114431725253346,"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."}}