{"id":"W4403540278","doi":"10.2196/62963","title":"Describing the Framework for AI Tool Assessment in Mental Health and Applying It to a Generative AI Obsessive-Compulsive Disorder Platform: Tutorial","year":2024,"lang":"en","type":"review","venue":"JMIR Formative Research","topic":"Digital Mental Health Interventions","field":"Psychology","cited_by":18,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Mental health; Psychological intervention; Transparency (behavior); Credibility; Agency (philosophy); Scrutiny; Accountability; Psychology; Computer science; Psychiatry; Computer security; Sociology; Political science","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01272693,0.002369263,0.00103886,0.007361727,0.002050529,0.007472216,0.003163577,0.004173354,0.01091619],"category_scores_gemma":[0.01486245,0.0012559,0.002730737,0.004367807,0.008434322,0.008505072,0.005479745,0.005591345,0.005047066],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005800981,"about_ca_system_score_gemma":0.006247671,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01139436,"about_ca_topic_score_gemma":0.01127606,"domain_scores_codex":[0.9929193,0.003828849,0.001077476,0.0005540575,0.001195377,0.0004249642],"domain_scores_gemma":[0.9927643,0.004707661,0.0006268364,0.0004369783,0.001199813,0.0002644381],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003048574,0.0001459866,0.002199414,0.001736789,0.00003934773,0.0007085527,0.005874554,0.002586825,0.002237197,0.7183282,0.02662176,0.2394909],"study_design_scores_gemma":[0.00002061654,0.0002075011,0.003643336,0.003771068,0.0000461122,0.002179363,0.00350852,0.01703464,0.002001309,0.3062067,0.6612063,0.0001744939],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"review","genre_scores_codex":[0.00200907,0.009059539,0.94295,0.01167321,0.0008292155,0.002016312,0.0005155507,0.001465897,0.02948133],"genre_scores_gemma":[0.02140013,0.010083,0.9497119,0.002408823,0.0005405118,0.003938595,0.000646434,0.0003507242,0.01091992],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.01272693,"threshold_uncertainty_score":0.06730723,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3058027182502824,"score_gpt":0.590659697631615,"score_spread":0.2848569793813325,"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."}}