{"id":{"repo_id":"mit","oai_identifier":"oai:dspace.mit.edu:1721.1/162985"},"canonical_url":"https://search.dev.ndltd.org/etd/mit/oai:dspace.mit.edu:1721.1/162985","repository":{"repo_id":"mit","name":"MIT","base_url":"https://dspace.mit.edu/oai/request"},"display":{"title":"All Therapies Are Equal - Unless You’re a Bot: Evaluating the Effectiveness of Four Therapy Schools for AI Chatbot Therapists","abstract":"This thesis tests two design questions for Large Language Model (LLM) Chatbot Therapists: Which therapeutic school suits an LLM best, and does an explicit Theory-of-Mind (ToM) reflection improve outcomes? We prompted GPT-4.1-mini to act as eight therapists — CBT, Narrative, Psychodynamic, and SFBT, each with and without a ToM step — and held 240 simulated sessions with scripted AI patients. SFBT achieved the greatest projected PHQ-9 improvement (around 4 points), significantly higher than CBT, Narrative, or Psychodynamic approaches. Immediate distress (SUDS) fell modestly and uniformly across schools. ToM reasoning did not alter either measure. The findings show that extra “thinking time” might not automatically translate into therapeutic gain, but also highlight a current strength of LLMs: executing brief, rule-based therapies.","abstract_html":"This thesis tests two design questions for Large Language Model (LLM) Chatbot Therapists: Which therapeutic school suits an LLM best, and does an explicit Theory-of-Mind (ToM) reflection improve outcomes? We prompted GPT-4.1-mini to act as eight therapists — CBT, Narrative, Psychodynamic, and SFBT, each with and without a ToM step — and held 240 simulated sessions with scripted AI patients. SFBT achieved the greatest projected PHQ-9 improvement (around 4 points), significantly higher than CBT, Narrative, or Psychodynamic approaches. Immediate distress (SUDS) fell modestly and uniformly across schools. ToM reasoning did not alter either measure. The findings show that extra “thinking time” might not automatically translate into therapeutic gain, but also highlight a current strength of LLMs: executing brief, rule-based therapies.","abstract_has_math":false,"creators":["Liu, Andi"],"institution":"Massachusetts Institute of Technology","degree_name":"Master","degree_level":null,"degree_discipline":null,"degree_department":"Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science","school":null,"contributors":[],"advisors":["Gabrieli, John"],"committee_chairs":[],"committee_members":[],"year":2025,"date_issued":"2025-05","date_published":"2025-05","updated_at":"2026-07-22T22:21:29Z","subjects":[],"languages":[],"rights":["In Copyright - Educational Use Permitted","Copyright retained by author(s)"],"rights_urls":["https://rightsstatements.org/page/InC-EDU/1.0/"],"identifier_entries":[]},"links":{"outbound_url":"https://hdl.handle.net/1721.1/162985","outbound_label":"Handle","outbound_source":"dc:identifier.uri"},"metadata_groups":[{"id":"people","label":"People","entries":[{"key":"dc:contributor.advisor","label":"Advisor","values":["Gabrieli, John"]},{"key":"dc:contributor.department","label":"Department","values":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science"]},{"key":"dc:creator","label":"Author","values":["Liu, Andi"]}]},{"id":"academic_context","label":"Academic Context","entries":[{"key":"dc:date.accessioned","label":"Dc Date Accessioned","values":["2025-10-06T17:38:12Z"]},{"key":"dc:date.available","label":"Dc Date Available","values":["2025-10-06T17:38:12Z"]},{"key":"dc:date.issued","label":"Date","values":["2025-05"]},{"key":"dc:publisher","label":"Institution","values":["Massachusetts Institute of Technology"]},{"key":"dc:type","label":"Dc Type","values":["Thesis"]},{"key":"thesis:degree_name","label":"Degree Name","values":["Master","Master of Engineering in Electrical Engineering and Computer Science"]}]},{"id":"language_rights","label":"Language and Rights","entries":[{"key":"dc:rights","label":"Dc Rights","values":["In Copyright - Educational Use Permitted","Copyright retained by author(s)"]},{"key":"dc:rights.uri","label":"Rights URI","values":["https://rightsstatements.org/page/InC-EDU/1.0/"]}]},{"id":"identifiers","label":"Identifiers","entries":[{"key":"dc:identifier.uri","label":"Identifier URI","values":["https://hdl.handle.net/1721.1/162985"]}]},{"id":"additional","label":"Additional Metadata","entries":[{"key":"dc:description.abstract","label":"Abstract","values":["This thesis tests two design questions for Large Language Model (LLM) Chatbot Therapists: Which therapeutic school suits an LLM best, and does an explicit Theory-of-Mind (ToM) reflection improve outcomes? We prompted GPT-4.1-mini to act as eight therapists — CBT, Narrative, Psychodynamic, and SFBT, each with and without a ToM step — and held 240 simulated sessions with scripted AI patients. SFBT achieved the greatest projected PHQ-9 improvement (around 4 points), significantly higher than CBT, Narrative, or Psychodynamic approaches. Immediate distress (SUDS) fell modestly and uniformly across schools. ToM reasoning did not alter either measure. The findings show that extra “thinking time” might not automatically translate into therapeutic gain, but also highlight a current strength of LLMs: executing brief, rule-based therapies."]},{"key":"dc:description.degree","label":"Dc Description Degree","values":["M.Eng."]},{"key":"dc:title","label":"Title","values":["All Therapies Are Equal - Unless You’re a Bot: Evaluating the Effectiveness of Four Therapy Schools for AI Chatbot Therapists"]}]}],"canonical_facts":{"dc:contributor.advisor":["Gabrieli, John"],"dc:contributor.department":["Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science"],"dc:creator":["Liu, Andi"],"dc:date.accessioned":["2025-10-06T17:38:12Z"],"dc:date.available":["2025-10-06T17:38:12Z"],"dc:date.issued":["2025-05"],"dc:description.abstract":["This thesis tests two design questions for Large Language Model (LLM) Chatbot Therapists: Which therapeutic school suits an LLM best, and does an explicit Theory-of-Mind (ToM) reflection improve outcomes? We prompted GPT-4.1-mini to act as eight therapists — CBT, Narrative, Psychodynamic, and SFBT, each with and without a ToM step — and held 240 simulated sessions with scripted AI patients. SFBT achieved the greatest projected PHQ-9 improvement (around 4 points), significantly higher than CBT, Narrative, or Psychodynamic approaches. Immediate distress (SUDS) fell modestly and uniformly across schools. ToM reasoning did not alter either measure. The findings show that extra “thinking time” might not automatically translate into therapeutic gain, but also highlight a current strength of LLMs: executing brief, rule-based therapies."],"dc:description.degree":["M.Eng."],"dc:identifier.uri":["https://hdl.handle.net/1721.1/162985"],"dc:publisher":["Massachusetts Institute of Technology"],"dc:rights":["In Copyright - Educational Use Permitted","Copyright retained by author(s)"],"dc:rights.uri":["https://rightsstatements.org/page/InC-EDU/1.0/"],"dc:title":["All Therapies Are Equal - Unless You’re a Bot: Evaluating the Effectiveness of Four Therapy Schools for AI Chatbot Therapists"],"dc:type":["Thesis"],"thesis:degree_name":["Master","Master of Engineering in Electrical Engineering and Computer Science"]},"updated_at":"2026-07-22T22:21:29Z"}