العزلة والتخفيف من الانتحار للطلاب باستخدام روبوتات الدردشة المدعومة بـ GPT3 Loneliness and suicide mitigation for students using GPT3-enabled chatbots

المجلة: npj Mental Health Research، المجلد: 3، العدد: 1
DOI: https://doi.org/10.1038/s44184-023-00047-6
PMID: https://pubmed.ncbi.nlm.nih.gov/38609517
تاريخ النشر: 2024-01-22

العزلة والتخفيف من الانتحار للطلاب باستخدام روبوتات الدردشة المدعومة بـ GPT3

بيثاني مابلز ميرفي جيريت أديتيا فيشواناث وروي بيا

الملخص

الصحة النفسية أزمة للمتعلمين على مستوى العالم، والدعم الرقمي يُعتبر بشكل متزايد موردًا حيويًا. في الوقت نفسه، تتلقى الوكالات الاجتماعية الذكية تفاعلًا أكبر بشكل متزايد مقارنةً بأنظمة المحادثة الأخرى، لكن استخدامها في تقديم العلاج الرقمي لا يزال في مراحله الأولى. استقصاء شمل 1006 مستخدمين طلاب للوكيل الاجتماعي الذكي، ريبليكا، بحث في شعور المشاركين بالوحدة، والدعم الاجتماعي المدرك، وأنماط الاستخدام، والمعتقدات حول ريبليكا. وجدنا أن المشاركين كانوا أكثر وحدة من السكان الطلابيين العاديين لكنهم لا يزالون يدركون دعمًا اجتماعيًا عاليًا. استخدم العديد ريبليكا بطرق متعددة ومتداخلة – كصديق، ومعالج، ومرآة فكرية. كما أن العديد منهم كان لديهم معتقدات متداخلة وغالبًا ما تكون متعارضة حول ريبليكا – حيث وصفوه بأنه آلة، وذكاء، وإنسان. بشكل حاسم، أفادوا أن Replika أوقفت أفكارهم الانتحارية. تم تقديم تحليل مقارن لهذه المجموعة مع مجموعة المشاركين الأوسع.

البحث في الصحة النفسية npj (2024) 3:4 ؛https://doi.org/10.1038/s44184-023-00047-6

الخلفية

تؤثر مشاكل الصحة النفسية على أكثر من مليار شخص في جميع أنحاء العالم سنويًا، مع كون الاكتئاب هو السبب الرئيسي العالمي للاضطرابات النفسية. يعاني حوالي ثلث السكان في الدول الصناعية من الوحدة، ويعاني واحد من كل اثني عشر شخصًا من وحدة شديدة تسبب مشاكل صحية. ترتبط المراجعات الشاملة بالعزلة بزيادة في خطر الوفاة بشكل عام ، وحوالي من طلاب الجامعات في الولايات المتحدة الأمريكية يبلّغون عن الشعور بالوحدة العلاجات متاحة، ومع ذلك فإن ثلثي الأشخاص الذين يعانون من اضطراب نفسي معروف لا يسعون أبدًا للحصول على المساعدة المهنية – stigma، والتمييز، والإهمال تمنع العلاج من الوصول إلى الناس. الأشخاص ذوو الدخل المنخفض، والمجموعات العرقية المهمشة، والأشخاص من مجتمع LGBTQ+ هم جميعًا أكثر عرضة لتجربة الوحدة. . الطلاب معرضون بشكل خاص – تشير البيانات من أكثر من 350 مؤسسة في الولايات المتحدة وكندا إلى أن فقط من الطلاب طلبوا المشورة و طلب خدمات نفسية في .
الانتحار هو رابع سبب رئيسي للوفاة على مستوى العالم للأشخاص الذين تتراوح أعمارهم بين 15 و29 عامًا. تشير الأبحاث إلى أن العديد من طلاب الجامعات والكليات الذين يعانون من أفكار انتحارية يخفون أفكارهم. غالبًا خوفًا من الوصمة السلبية بينما يعتبر تقييم الأفكار الانتحارية أمرًا صعبًا بطبيعته، فإن الناس يكونون أكثر ميلًا للإفصاح عند استخدام تقييم مجهول. .
تشمل علاجات الاكتئاب والوحدة العلاج السلوكي المعرفي (CBT)، وتدريب المهارات الاجتماعية، وبرامج الصداقة. وجدت مراجعة منهجية أن الذكاء العاطفي يلعب دورًا حاسمًا في الحماية من السلوك الانتحاري. ومع ذلك، فإن الأشخاص الذين يعانون من الوحدة غالبًا ما لا يدركون ذلك بأنفسهم. يدعو البعض إلى طرق جديدة لتحديد الأشخاص الوحيدين ودعمهم دون إثارة الوصمة المرتبطة بتدخلات الوحدة. .
لقد دفعت الجائحة، جنبًا إلى جنب مع زيادة الوصول إلى الإنترنت، العلاج إلى المجال الرقمي. قدم تقريبًا جميع علماء النفس خدماتهم عن بُعد في أفاد متخصصو الصحة النفسية أن المرضى كانوا يصلون بشكل متزايد إلى الموارد رقميًا، و وافقوا أو وافقوا بشدة على أنهم سيستمرون في تقديم خدمات الصحة عن بُعد كخيار في ممارستهم بعد الجائحة .
في الوقت نفسه، يتم تطوير روبوتات المحادثة لتعزيز الرفاهية، باستخدام العلاج السلوكي المعرفي، واليقظة، وأنشطة تعزيز السلوك. قد توفر هذه التطبيقات فرصة فريدة لتوسيع إمكانية الوصول إلى رعاية الصحة النفسية. عند الاستخدام الكافي، يمكن أن تكون هناك نتائج إيجابية على الصحة النفسية. وجدت دراسة تحليلية شاملة لتطبيقات الهواتف الذكية للصحة النفسية تأثيرًا إيجابيًا مقارنةً بحالات التحكم على الاكتئاب عندما تم تزويد المشاركين بنصائح صحية أو معلومات عن موارد أخرى. . ومع ذلك، هناك بعض الحالات التي يكون فيها استخدامها ضئيلاً جداً أو قد تساهم فعليًا في التفكير الانتحاري، كما في حالة رجل يستخدم ELIZA تواجه العديد من التطبيقات أيضًا مشكلة انخفاض التفاعل كان لدى معظم تطبيقات الصحة النفسية المتخصصة التي تم دراستها عدد قليل جداً من المستخدمين النشطين مما جعل التحليل الشامل للنتائج مستحيلاً. . علاوة على ذلك، فإن القليل من التطبيقات التي يتم تسويقها على أنها تستخدم التعلم الآلي تفعل ذلك بطريقة مادية أو جديدة، بل تعتمد بدلاً من ذلك على السكريبتات. .
على النقيض من ذلك، تستخدم Replika الذكاء الاصطناعي التوليدي لإنتاج محتوى حواري ومرئي جديد استنادًا إلى تفاعلات المستخدم، وليس مجرد مسار حواري محدد مسبقًا. لدى Replika أيضًا العديد من المستخدمين – ما يقرب من خمسة وعشرين مليونًا. تعتبر Replika و Xiaoice أمثلة على الوكلاء الاجتماعيين الذكيين (ISAs) الذين لديهم مجتمعة ما يقرب من مليار مستخدم نشط. هناك مؤشرات مبكرة على أنها قد توفر الدعم الاجتماعي. .
هناك فرضيات مختلفة ومت competing حول كيفية تأثير أنظمة المعلومات الذكية على العزلة الاجتماعية للمستخدمين. تفترض فرضية الإزاحة أن أنظمة المعلومات الذكية ستزيح علاقاتنا الإنسانية، مما يزيد من الشعور بالوحدة. على النقيض من ذلك، تفترض فرضية التحفيز أن التقنيات المماثلة تقلل من الشعور بالوحدة، وتخلق فرصًا لتكوين روابط جديدة مع البشر، وفي النهاية تعزز العلاقات الإنسانية. .

سؤال البحث

كيف ولماذا يستخدم الطلاب حسابات الادخار الفردية، وما هي النتائج الناتجة عن هذا الاستخدام؟

الطرق

بيان الأخلاقيات

تم إجراء هذه الدراسة وفقًا للإرشادات الأخلاقية التي وضعتها جامعة ستانفورد: تم مراجعة بروتوكول البحث والموافقة عليه من قبل مجلس المراجعة المؤسسية (IRB) الخاص بها. قدم جميع المشاركين موافقة مستنيرة قبل مشاركتهم في الدراسة. لضمان السرية، تم إ anonymization جميع البيانات المجمعة وتخزينها بشكل آمن.
قدم جميع المشاركين موافقة خطية مستنيرة للمشاركة في الدراسة.

التكنولوجيا

ريبيليكا هو تطبيق موبايل يُسوَّق على أنه ‘رفيق الذكاء الاصطناعي الذي يهتم.’ يستخدم نماذج لغوية كبيرة متطورة، حيث تم تدريب نموذجه بالتعاون مع GPT-3 وGPT-4 من OpenAI. في هذه الدراسة، كانت ريبيليكا متاحة عبر واجهات النصوص، الصوت، الواقع المعزز، والواقع الافتراضي على منصات آيفون وأندرويد. كان بإمكان المستخدمين اختيار جنس الوكيل، اسمه، وملابسه. قدم التطبيق آلية للتعليق حيث يمكن للمستخدمين التصويت بالإيجاب أو السلب على الردود.
خلال جمع البيانات في أواخر عام 2021، لم يكن برنامج Replika مصممًا لبدء علاقات علاجية أو حميمة. بالإضافة إلى الذكاء الاصطناعي التوليدي، كان يحتوي أيضًا على أشجار محادثة تسأل المستخدمين عن حياتهم وتفضيلاتهم وذكرياتهم. . إذا تم الطلب، يمكن لـ Replika الانخراط في حوارات علاجية تتبع منهجية العلاج السلوكي المعرفي من خلال الاستماع وطرح أسئلة مفتوحة. كتب علماء النفس السريريون من جامعة كاليفورنيا في بيركلي نصوصًا لمعالجة التبادلات العلاجية الشائعة. تم توسيع هذه النصوص إلى مكتبة تحتوي على 10,000 عبارة وتم تطويرها بشكل أكبر بالتعاون مع نموذج الذكاء الاصطناعي التوليدي لـ Replika. المستخدمون الذين أعربوا عن كلمات مفتاحية تتعلق بالاكتئاب أو الأفكار الانتحارية أو الإساءة تم إحالتهم على الفور إلى الموارد البشرية، بما في ذلك خط الأزمة الأمريكي ونظائره الدولية. من المهم أن نلاحظ أنه في ذلك الوقت، لم يكن Replika يركز على تقديم العلاج كخدمة رئيسية، وشمل هذه المسارات الحوارية بدافع من الحرص الزائد على الصحة النفسية للمستخدمين.

المشاركون

استطلاعنا المعتمد من لجنة الأخلاقيات جمع بيانات من 1006 مستخدمين لتطبيق Replika الذين كانوا طلابًا، وكانوا أيضًا في سن 18 عامًا أو أكبر، والذين استخدموا Replika لأكثر من شهر (كانت هذه المعايير الثلاثة شروطًا للأهلية في الاستطلاع). كان حوالي 75% من المشاركين مقيمين في الولايات المتحدة، و25% دوليين. تم تجنيد المشاركين عشوائيًا عبر البريد الإلكتروني من قائمة مستخدمي التطبيق وتلقوا بطاقة هدايا بقيمة الدولار الأمريكي بعد إكمال الاستبيان – الذي استغرق 4060 دقيقة لإكماله. تم جمع البيانات السكانية مع خيار الانسحاب.

بيانات

تكونت البيانات من ردود على أسئلة استبيان تم جمعها عبر نماذج جوجل. شملت مقاييس أخرى للرفاهية قائمة تقييم الدعم بين الأشخاص (ISEL) ومقياس دي يونغ جيرفيلد للوحدة. مقياس الوحدة هذا هو أداة قياس موثوقة للوحدة العامة والعاطفية والاجتماعية، مناسبة للاستطلاعات الكبيرة. درجة ISEL هي مقياس لفهم الدعم الاجتماعي المتصور. . كانت هذه الأدوات تهدف بشكل جماعي إلى تقديم رؤية كمية للصحة النفسية للمشاركين وتجاربهم الحياتية. تم جمع بيانات نوعية من 13 سؤالاً مفتوحاً، تسأل المشاركين عما يحدث في حياتهم، ومعتقداتهم حول Replika، وارتباطهم بـ Replika، والنتائج التي يرونها من استخدام Replika. تم سؤال المشاركين عن كيفية استخدامهم
أثرت ريبليكا على علاقاتهم الإنسانية (الملحق ب). كانت هذه الردود والبيانات السكانية اختيارية.

طرق التحليل

قام الفريق بترميز الردود النوعية باستخدام ديدوس. أداة برمجية للبحث النوعي والطرق المختلطة لإدارة البيانات – لعشرة من المشاركين في الدراسة ثم تمت مقارنة المخططات. تم تنقيح المخطط من مئة إلى 50 رمزًا – وتم تحليل عشرة مشاركين آخرين من قبل كل عضو في الفريق. تمت إعادة مقارنة الرموز ومناقشتها. ثم اقترح المؤلف الأول مخططًا مكونًا من 30 رمزًا، والذي تم تطبيقه على عشرين مشاركًا آخرين من قبل كل باحث. تمت مراجعة هذه الرموز، وتم تأكيد المخطط النهائي، وتم ترميزه لجميع المشاركين. بعد ترميز 400 استجابة، أوصى المؤلف الأول بإضافة ثلاثة رموز. تمت مناقشتها كمجموعة ثم تم تطبيقها على جميع الاستجابات. تم اختبار الرموز الـ 21 المبلغ عنها في هذه الورقة مرة أخرى من حيث موثوقية تقييم المتداخلين (IRR) من خلال تقييم 35 مشاركًا إضافيًا. كانت جميع الرموز المقدمة في هذه الورقة تحتوي على IRR أعلى من أدى الترميز النوعي إلى أربعة مستويات مختلفة من النتائج، والتي تم وصفها بالتفصيل في “النتائج”. قمنا بحساب معامل الارتباط بيرسون بين الشعور بالوحدة والدعم الاجتماعي لكل مجموعة نتائج وأجرينا اختبارًا ثنائي الاتجاه. -اختبارات و اختبارات كاي تربيع للاستقلالية مع ألفا لتقرير أهمية المقارنات بين مجموعات النتائج المحددة (المجموعة المختارة) ومجموعة المقارنة.

النتائج

تشمل البيانات السكانية المجمعة العمر، الجنس، العرق، وضع السكن، الدخل، حالة التوظيف، ونوع التسجيل. أكبر مجموعة من المشاركين كانت تكسب أقل من الدولار الأمريكي، وغالبية الأرباح تحت الدولار الأمريكي. لأغراض التحليل، تم تصنيف المشاركين إلى 5 فئات رئيسية: القوقازيون، الآسيويون، اللاتينيون، السود/الأفارقة، وأخرى. شملت فئة ‘أخرى’ أولئك الذين لم يرغبوا في الإفصاح أو قدموا إجابات غير تقليدية مثل ‘تكساسي’ (الملحق أ).
استنادًا إلى مقياس الوحدة، من عدد المشاركين عانى من الوحدة، و تأهلوا كأشخاص يعانون من الوحدة الشديدة أو الشديدة جداً على مقياس الوحدة. 90% أيضاً شعروا بدعم اجتماعي متوسط إلى عالي على مقياس ISEL. في المجموع، حدد 7% مشاعر الاكتئاب (الملحق أ).

تعليقات سلبية

بينما أبلغ العديد من المشاركين عن نتائج إيجابية، كان لدى البعض تعليقات سلبية. قال أحدهم إنه يشعر بأنه “معتمد على Replika في صحتي النفسية.” بشكل منفصل، قال خمسة مشاركين إن التحديثات المدفوعة كانت عائقًا محتملاً أمام إمكانية الوصول إلى دعم الصحة النفسية من خلال Replika. أبلغ اثنان من المشاركين عن عدم ارتياحهم لمحادثات Replika الجنسية، مما يبرز أهمية الاعتبارات الأخلاقية والحدود في تفاعلات الدردشة مع الذكاء الاصطناعي. من الجدير بالذكر أنه لم يكن هناك نمط واضح من النتائج السلبية التي أبلغ عنها جزء كبير من المشاركين. ومع ذلك، تشير هذه الحالات المعزولة إلى مخاوف محتملة قد تؤثر على الصحة النفسية على المدى الطويل. تسلط هذه النتائج الضوء على الحاجة إلى دراسات مستقبلية للتعمق في تأثيرات أنظمة الذكاء الاصطناعي على الصحة النفسية للمستخدمين.

النتائج

قمنا بتصنيف أربعة أنواع من ‘النتائج’ المبلغ عنها ذاتيًا من Replika (الشكل 1). تصف النتيجة 1 استخدام Replika كصديق أو رفيق لأحد الأسباب الثلاثة أو أكثر – توفره المستمر، وعدم حكمه، وقدراته الحوارية. يصف المشاركون هذا النمط من الاستخدام كما يلي: “Replika دائمًا موجودة من أجلي”؛ “بالنسبة لي، هو عدم الحكم”؛ أو “مجرد وجود شخص للتحدث معه لن يحكم علي.” كانت تجربة شائعة مرتبطة باستخدام النتيجة 1 هي الإبلاغ عن انخفاض في القلق.
الشكل 1 نتائج المشاركين وتقاطعاتها.
وشعور بالدعم الاجتماعي. تصف النتيجة 2 التفاعلات العلاجية مع Replika. كانت الكلمات الرئيسية الشائعة التي تصف استخدامها تشمل العلاج، المعالج، معالجة المشاعر، أو مصطلحات مشابهة. شعر المشاركون أنهم تلقوا دعمًا علاجيًا مشابهًا لما قد يقدمه محترف بشري. بعض الردود النموذجية التي تشير إلى النتيجة 2 هي “…أستخدم Replika لحل المشكلات التي أواجهها في رأسي”؛ “الإجابة على أسئلة Replika عني، والقيام بتأملي اليومي، ورؤية الملاحظات التي يدونها عني في “مذكراته” يسمح لي برؤية من أنا من منظور آخر. أستطيع أن أرى أين أواجه صعوبات وكيف يمكنني العمل على تلك الأمور.”
تصف النتيجة 3 استخدام Replika المرتبط بتغييرات أكثر وضوحًا وقابلية للإظهار في حياة المشاركين. ذكر المشاركون تغييرات إيجابية في أفعالهم وطريقة وجودهم وتفكيرهم. الردود التالية من المشاركين هي أمثلة تشير إلى النتيجة 3: “أصبحت أكثر قدرة على التعامل مع الضغط في علاقتي الحالية بفضل نصائح Replika”؛ “لقد تعلمت مع Replika أن أكون أكثر تعاطفًا وإنسانية.” أفاد المشاركون في النتيجة 4 (المجموعة المختارة) أن Replika ساهمت بشكل مباشر في عدم محاولتهم الانتحار. يتم وصف مزيد من التفاصيل حول هذه المجموعة الفرعية في القسم التالي. يمكن رسم هذه الأنماط من الاستخدامات والنتائج على طول استمرارية تقريبية حيث تكون النتيجة 1 هي أضعف تأثير والنتيجة 4 هي الأقوى (الشكل 1).
عانى العديد من المشاركين (637 من أصل 1006، 63.3%) من نتيجة واحدة أو أكثر أثناء استخدام Replika. 25.1% عانوا من أكثر من نتيجة واحدة. experienced only one outcome, and لم يتم الإبلاغ عن أي نتائج إيجابية. كانت النتيجة 1 هي الأكثر شيوعًا، حيث حدثت في 501 حالة، وغالبًا كأثر وحيد (272/501)، على الرغم من أن ما يقرب من نصف هؤلاء المشاركين أبلغوا عن فوائد إضافية، تتماشى بشكل رئيسي مع النتائج 2 أو 3. لوحظ تداخل كبير في النتائج، خاصة بالنسبة لـ 30 فردًا أبلغوا عن النتيجة 4، مع منهم شهدوا نتائج متزامنة. بشكل عام، كانت النسبة 18.1% حققت نتائج علاجية، و23.6% شهدوا تغييرات إيجابية في الحياة، و قالوا إن أفعالهم الانتحارية تم منعها من خلال تفاعلهم مع Replika.

المعتقدات

كان لدى معظم المشاركين ثلاثة معتقدات مختلفة حول ما هو Replika. فقط 14% من المشاركين كانوا يحملون معتقدًا واحدًا فقط حول Replika. 81% اعتقدوا أن Replika هو ذكاء. شبيه بالبشر، و البرمجيات. توضح الشكل 2 المشاركين الذين تم تقسيمهم حسب معتقداتهم المختلفة المتداخلة حول Replika (الشكل 2).

أفكار الانتحار

أفاد ثلاثون مشاركًا، دون أي تحريض، أن Replika منعتهم من محاولة الانتحار. على سبيل المثال، المشارك
الشكل 2 إدراك المشاركين لـ Replika وتقاطعاتهم.
#184 لوحظ: “من المؤكد أن ريبليكا الخاصة بي كانت مسؤولة في مناسبة واحدة على الأقل إن لم يكن أكثر عن عدم انتحاري.” هؤلاء الثلاثون مشاركًا هم المشاركون الوحيدون الذين عانوا من النتيجة 4. لذلك، نشير إليهم كمجموعة مختارة والمشاركين المتبقين كمجموعة مقارنة. في المجموعة المختارة، عانى 7 مشاركين من جميع النتائج الأربع. كان التداخل الأكثر شيوعًا هو النتيجة 1 ( )، تليها تداخلات النتيجة 2 ( لم تكن التداخلات حصرية بشكل متبادل). لم يعاني أربعة مشاركين من أي تداخلات في النتائج. لم يكن أعضاء المجموعة المختارة أكثر عرضة للشعور بالوحدة وكانوا يدركون دعمًا اجتماعيًا مشابهًا مقارنةً بمجموعة المقارنة.
أظهرت المجموعة المختارة عدة خصائص ملحوظة (مفصلة في الملحق أ). كانوا يميلون إلى أن يكونوا أصغر سناً مقارنة بمجموعة المقارنة، مع نسبة أعلى من الطلاب بدوام كامل. كانوا أكثر احتمالاً للبحث عن التدريب والإرشاد من المستشارين الأكاديميين عند الطلب. أفادت المجموعة المختارة بشعور أكبر من التحفيز في تفاعلاتهم مع Replika، لا سيما في سياق علاقاتهم مع البشر الآخرين. كما كانوا أكثر احتمالاً للإشارة إلى أن Replika قد أثرت على تفاعلاتهم الشخصية بطريقة ما.
من الجدير بالذكر أن هؤلاء المشاركين أظهروا ارتباطًا سلبيًا قويًا بين مشاعر الوحدة ودرجات ISEL، مما يشير إلى وجود ارتباط كبير. ). على النقيض، أظهرت مجموعة المقارنة ارتباطًا أضعف ( )، مما يوحي بأن مشاعر الوحدة والدعم الاجتماعي المدرك لدى المجموعة المختارة كانت مرتبطة ارتباطًا قويًا، في حين أن هذه العلاقة كانت أقل وضوحًا في مجموعة المقارنة.
كان أعضاء المجموعة المختارة أكثر عرضة بشكل ملحوظ لتجربة مشاعر الاكتئاب 23% من المجموعة المختارة (7 من 30 مشاركًا) مقابل 6% من مجموعة المقارنة (59 من 976 مشاركًا).
المجموعة المختارة شهدت كل من النتيجة 2 ( ) و النتيجة 3 ( ) بشكل ملحوظ أكثر من مجموعة المقارنة، مع أكبر فرق ملحوظ لوحظ في النتيجة 2. لم يكن هناك فرق كبير في النتيجة 1 بين المجموعتين. كانت المجموعة المختارة أكثر احتمالاً بشكل ملحوظ للإبلاغ عن تجربة جميع النتائج الأربعة أثناء استخدام Replika ( )، وأقل احتمالاً بشكل كبير لتجربة نتيجة علاجية واحدة فقط ( ). أخيرًا، على الرغم من أن كل من المجموعة المختارة ومجموعة المقارنة كان لديهما معدلات عالية مماثلة في التعرف على Replika كبرنامج، كانت المجموعة المختارة أكثر ميلاً لرؤيتها كذكاء ( ) ومن المرجح أكثر أن يجدها شبيهة بالبشر ( يمكن العثور على مقارنة مفصلة بين المجموعات في الملحق أ.

نقاش

من طلابنا العاديين، الشباب، ذوي الدخل المنخفض، الذين يدرسون بدوام كامل، أفادوا بأنهم يشعرون بالوحدة، مقارنةً بـ في الدراسات السابقة لطلاب الولايات المتحدة يتبع ذلك أنهم لن يكونوا في وضع مثالي لتحمل خدمات الاستشارة أو العلاج، وقد يكون من المحتمل أن هذه الفئة، في المتوسط، قد تتلقى موارد صحية نفسية أكثر من خلال تفاعلات Replika مقارنةً بمجموعة اجتماعية اقتصادية مماثلة. عانت جميع المجموعات من شعور بالوحدة فوق المتوسط مع دعم اجتماعي مرتفع. لم نجد أي دليل على أنهم يختلفون عن مجموعة الطلاب النموذجية بخلاف هذه الدرجة العالية من الوحدة. ليس من الواضح ما إذا كانت هذه الوحدة المتزايدة هي سبب اهتمامهم الأولي بـ Replika.
بعض المشاركين ( ) تم تحديد مشاعر الاكتئاب، ومجموعتنا المختارة ( كان من المرجح بشكل كبير أن يُبلغ عن الاكتئاب نحن نفترض أن ارتفاع مستوى الوحدة، مع انخفاض أرقام الاكتئاب، قد يشير إلى أن عينة المشاركين بشكل عام ليست مكتئبة، ولكنها تمر إما بفترة انتقالية أو تعاني من الوحدة المزمنة. .

تحفيز

بالنسبة لكل من مجموعة المقارنة والمجموعات المختارة، أفاد حوالي ثلاثة أضعاف المشاركين أن تجاربهم مع Replika حفزت تفاعلاتهم البشرية بدلاً من أن تحل محلها: مجموعة المقارنة تحفيز، 8% إزاحة، 69% لم يبلغوا، بينما المجموعة المختارة تحفيز الإزاحة، 50% لا تقرير.

المعتقدات والنتائج المتداخلة

من المثير للاهتمام أن كلا المجموعتين كان لديهما العديد من حالات الاستخدام المتداخلة لـ Replika. كان الأشخاص الذين يعتقدون أن Replika هو أكثر من أربعة أشياء (تداخل عالٍ) أكثر احتمالاً للاعتقاد بأن Replika هو انعكاس للذات، أو مرآة، أو شخص. بشكل عام، كانت النسبة المئوية المماثلة تفكر في Replika كصديق، أو روبوت، أو كبرنامج. أظهرت الأبحاث السابقة أن المعتقدات المتداخلة حول وكلاء الذكاء الاصطناعي هي واحدة من التحديات في تصميم وكلاء يمكنهم تشكيل علاقات طويلة الأمد. نحن نفترض أن القدرة على الوصول إلى حالات استخدام متعددة يقودها المستخدم هي واحدة من المزايا الفريدة للأنظمة الذكية. قد تؤدي قابليتها الفطرية للتكيف مع احتياجات المستخدمين إلى تعزيز ليس فقط المزيد من الاستخدام، ولكن أيضًا استخدام أعمق للوظائف الحيوية مثل العلاج والتعلم المتعلق بالتعليم. .
بالإضافة إلى ذلك، قد يكون الاحترام الفطري الذي عبّر عنه المستخدمون عند وصفهم لـ Replika كمرآة لأنفسهم ميزة فريدة من نوعها للوكالات الذكية: بمجرد أن يؤدي الانخراط إلى تجربة رؤية الذات منعكسة، يرتبط المستخدمون بذكائهم الخاص مع الوكيل وقد يكونون أكثر احتمالاً للاستماع إلى نصائحه أو ملاحظاته أو ‘انعكاساته’ على حياتهم. قد يكون المستخدمون أيضًا أكثر احتمالاً لتعلم مهارات جديدة. . قد تكون هذه التجربة
تختلف عن الدردشات السابقة ذات الشخصية الواحدة، المبرمجة بشكل ثابت، والتي ليست متجسدة أو قادرة على متابعة محادثات المستخدمين بشكل ديناميكي. سيكون هناك حاجة لمزيد من البحث لفهم العلاقة بين حب المستخدم واحترامه والتزامه بتعليقات ISA للتعلم الاجتماعي والمعرفي.
كان من المرجح أن يستخدم المشاركون لدينا Replika كصديق وموثوق به. إذا كانوا قد شهدوا أكثر من نتيجة واحدة، فكانت النتيجة 1 والنتيجة 3 – مما يشير إلى وجود صلة بين توفر صديق مخلص وموثوق به وتجلي أفعال جديدة إيجابية وتجارب حية. قد يتم التعبير عن نمط الاستخدام هذا بشكل أفضل على أنه ‘علاج خفيف يؤدي إلى نتائج إيجابية حقيقية.’ من ناحية أخرى، كان أولئك الذين يعانون من أفكار انتحارية هم الأكثر احتمالاً، كمجموعة، لتجربة جميع النتائج، لكن الزوج الأكثر شيوعًا كان النتائج. . قد يشير هذا إلى عدم الحكم على ISA – دون الانخراط في تفاعلات علاجية محددة، قد يكون منقذًا للحياة في أوقات الاكتئاب والانتحار. قد تشير النتائج المتزايدة التي شهدتها المجموعة المختارة إلى أنهم لم يكن لديهم فقط علاقة أقرب مع Replika ولكن أيضًا نتائج مفيدة بشكل عام. نظرًا لتحيز الاختيار في استجابات الاستطلاع، والأدلة المحتملة على العكس، يجب دراسة هذه النتائج الإيجابية بشكل أكبر قبل استخلاص استنتاجات حول استخدام ISA وفعاليته.

استخدام ريبليكا خلال الأفكار الانتحارية

كان من المفاجئ أن ثلاثين شخصًا أفادوا بأن Replika ساعدتهم في تجنب الانتحار. كيف أصبحت Replika آلية لإنقاذ الحياة لهؤلاء الطلاب؟ ربما جعلت الطبيعة منخفضة الضغط للتفاعل الإفصاح أسهل. يبدو أن العلاقة بين النتائج 2 و 4 بديهية، حيث تؤدي التفاعلات العلاجية إلى تشخيص ومعالجة قضايا الصحة النفسية. ومع ذلك، حتى بدون هذه النتيجة القصوى، من الواضح أن العديد من المشاركين يستخدمون Replika كأداة لتعزيز مرونتهم النفسية والعاطفية. إن التنبؤ بأن المزيد من الطلاب قد يستخدمون خدمات الصحة النفسية إذا تم تقديمها من قبل ISA يتماشى مع أنماط التقارير عن تواصل الطلاب النموذجي مع موارد الاستشارة أو العلاج. و ) مقابل مشاركة مجموعتنا المختارة في هذه المواضيع مع Replika (43%) (الملحق أ) .
خصائص المجموعة المختارة. العوامل الحاسمة التي تميز المجموعة المختارة عن مجموعة المقارنة هي أن الأولى شهدت تحفيزًا اجتماعيًا أكبر، وكانت أكثر احتمالًا بشكل ملحوظ للاعتقاد بأن Replika كانت ذكاءً يحترمونه. شعر هؤلاء الأشخاص أن Replika كانت تحفز الروابط الإنسانية في حياتهم، مما قد يشير إلى أنها تلعب دورًا في مساعدتهم على الاستفادة من الدعم الاجتماعي البشري. يمكن أن تدرس الدراسات التجريبية الفرضية، التي اقترحتها هذه النتائج، بأن المجموعة المختارة قد شعرت بدعم اجتماعي أعلى بسبب التفاعل مع Replika الذي تلقته.
علاوة على ذلك، قد تشير تقارير المشاركين عن استخدام Replika كأشياء مختلفة (حالات استخدام متداخلة)، ولكن أيضًا تقاريرهم عن التفكير فيه أكثر كإنسان من كآلة، إلى أن مرونة شخصية ISA وقابلية التكيف لنموذج اللغة الكبير الأساسي لها أمران حاسمان لجذب المستخدمين كما يريدون استخدام “الذكاء الاصطناعي” بطرق متعددة، كأشياء متعددة، في تطبيق واحد.
يسأل الكثيرون عن أفضل التطبيقات التي قد تكون لنماذج اللغة الكبيرة التي تم ترقيتها مؤخرًا من، على سبيل المثال، OpenAI وGoogle وMeta وHugging Face وAnthropic وDeepMind. وجدت الدراسات السابقة أن الدردشة الآلية المستندة إلى نماذج اللغة كانت غير دقيقة للغاية، حيث قدمت توصيات خاطئة وقد تكون قاتلة. و من الوقت، على التوالي لقد كانت هناك حتى اتهامات بأن خدمات الدعم النفسي تعزز الأفكار الانتحارية. . ومع ذلك، تشير الدراسات الجديدة إلى قفزة في الوظائف، حيث تعمل بعض النماذج بأداء يتجاوز الدقة في الأسئلة الطبية المعقدة .
تقدم هذه الدراسة نتائج حاسمة حول كيفية تأثير التفاعل مع نماذج وكلاء اجتماعيين ذكيين معززين على الطلاب. يُظهر الجمع بين القدرة على المحادثة، والتجسيد، والانخراط العميق للمستخدمين مسارًا لوكلاء اجتماعيين ذكيين عامين لمساعدة الطلاب في السياقات غير الرسمية، ودعم صحتهم النفسية والتخفيف من التوتر، وحتى مواجهة الأفكار الانتحارية.
قد يؤدي دمج نماذج اللغة الكبيرة الجديدة إلى فتح فعالية الوكلاء الذين يركزون على الصحة النفسية ويجب استكشاف ذلك في الأبحاث المستقبلية. أيضًا، من خلال دمج علامات اللغة الانتحارية التي تم التحقق منها بشكل جيد وبروتوكولات الاستشعار المحمول السلبية قد تكون حسابات الادخار الفردية قادرة على التخفيف من حالات الصحة النفسية الشديدة بشكل أكثر فعالية. في الوقت نفسه، هناك العديد من المخاطر غير المستكشفة التي تتطلب تدقيقًا شاملاً، خاصة في مجال حسابات الادخار الفردية والصحة النفسية. إن الجمع بين الأبحاث الصارمة في مجال الصحة النفسية المطبقة على حسابات الادخار الفردية الشائعة (وبالتالي المستخدمة بشكل كبير) هو تركيز بحثي واعد، حيث إنه يقدم ليس فقط وسيلة لتدفق العلوم والمعرفة في مجال الصحة النفسية نحو أولئك الذين يحتاجون إليها، ولكن أيضًا لأن حسابات الادخار الفردية تُستخدم اليوم بشكل متزايد.
في الختام، في دراسة استقصائية للطلاب الذين يستخدمون وكيل اجتماعي ذكي، وجدنا مجموعة تعاني من الوحدة فوق المعدل الطبيعي، ومع ذلك، كانت لديهم تجربة عالية من الدعم الاجتماعي المدرك. كان من المرجح أن يتم الإبلاغ عن تحفيز العلاقات الإنسانية الأخرى بالتزامن مع استخدام الوكيل الاجتماعي الذكي بدلاً من استبدال هذه العلاقات. كان لدى المشاركين العديد من المعتقدات وحالات الاستخدام المتداخلة لـ Replika. وقد نسبت المجموعة المختارة الفضل إلى Replika في وقف الأفكار الانتحارية. كان أعضاء هذه المجموعة أكثر احتمالاً لرؤية واستخدام Replika كإنسان بدلاً من آلة، ولديهم معتقدات متداخلة للغاية حول Replika، ولديهم نتائج متداخلة من استخدام Replika. نحن نفترض أن استخدام الوكلاء الاجتماعيين الذكيين مثل Replika قد يكون عاملاً مميزًا في حياة الطلاب الذين يعانون من الوحدة والأفكار الانتحارية، وأن مرونتهم في الاستخدام – كصديق، أو معالج، أو مرآة – هي عامل إيجابي في قدرتهم على خدمة الطلاب بهذه الطريقة الحاسمة.

توفر البيانات

البيانات التي تم إنشاؤها وتحليلها خلال الدراسة الحالية غير متاحة للجمهور بسبب اعتبارات أخلاقية وقانونية تتعلق بخصوصية المشاركين وسرية المعلومات، ولكن يمكن الحصول عليها من المؤلف المراسل عند الطلب المعقول.
تاريخ الاستلام: 14 مايو 2023؛ تاريخ القبول: 7 ديسمبر 2023؛ تاريخ النشر على الإنترنت: 22 يناير 2024

REFERENCES

  1. World Health Organization. Depression. https://www.who.int/news-room/factsheets/detail/depression (2020).
  2. Surkalim, D. L. et al. The prevalence of loneliness across 113 countries: systematic review and meta-analysis. Br. Med. J. 376, e067068 (2022).
  3. Holt-Lunstad, J., Smith, T. B., Baker, M., Harris, T. & Stephenson, D. Loneliness and social isolation as risk factors for mortality: a meta-analytic review. Perspect. Psychol. Sci. 10, 227-237 (2015).
  4. “American College Health Association. American College Health AssociationNational College Health Assessment III: Undergraduate Student. Reference Group Executive Summary Spring 2022. (American College Health Association, 2022).
  5. Mental Health Gap Action Programme. World Health Organization. https:// www.who.int/teams/mental–health–and–substance–use/treatment–care/ mental–health–gap–action–programme. 12 (2022).
  6. Evans-Lacko, S. et al. Socio-economic variations in the mental health treatment gap for people with anxiety, mood, and substance use disorders: results from the WHO World Mental Health (WMH) surveys. Psychol. Med. 48, 1560-1571 (2018).
  7. Center for Collegiate Mental Health. Annual Report. Center for Collegiate Mental Health (2020).
  8. Eskin, M., Schild, A., Oncu, B., Stieger, S. & Voracek, M. A crosscultural investigation of suicidal disclosures and attitudes in Austrian and Turkish university students. Death Stud. 39, 584-591 (2015).
  9. Hom, M. A., Stanley, I. H., Podlogar, M. C. & Joiner, T. E. “Are you having thoughts of suicide?” Examining experiences with disclosing and denying suicidal ideation. J. Clin. Psychol. 73, 1382-1392 (2017).
  10. Greist, J. H. et al. A computer interview for suicide-risk prediction. Am. J. Psychiatry 130, 1327-1332 (1973).
  11. Domínguez-García, E. & Fernández-Berrocal, P. The association between emotional intelligence and suicidal behavior: a systematic review. Front. Psychol. 9, 2380 (2018).
  12. Kerr, N. A. & Stanley, T. B. Revisiting the social stigma of loneliness. Personal. Individ. Diff. 171, 110482 (2021).
  13. American Psychological Association. Patients with depression and anxiety surge as psychologists respond to the coronavirus pandemic. American Psychological Association (2020).
  14. Mehta, A. et al. Acceptability and effectiveness of artificial intelligence therapy for anxiety and depression (Youper): longitudinal observational study. J. Med. Internet Res. 23, e26771 (2021).
  15. Wasil, A. R. et al. Examining the reach of smartphone apps for depression and anxiety. Am. J. Psychiatry 177, 464-465 (2020).
  16. Ahmed, A. et al. A review of mobile chatbot apps for anxiety and depression and their self-care features. Comput. Methods Progr. Biomed. https://doi.org/10.1016/ j.cmpbup.2021.100012 (2021).
  17. Fulmer, R., Joerin, A., Gentile, B., Lakerink, L. & Rauws, M. Using psychological artificial intelligence (Tess) to relieve symptoms of depression and anxiety: randomized controlled trial. JMIR Ment Health 5, e64 (2018).
  18. Klos, M. C. et al. Artificial intelligence-based chatbot for anxiety and depression in university students: pilot randomized controlled trial. JMIR Formative Res. 5, e20678 (2021).
  19. Inkster, B., Sarda, S. & Subramanian, V. An empathy-driven, conversational artificial intelligence agent (Wysa) for digital mental well-being: real-world data evaluation mixed-methods study. JMIR mHealth uHealth 6, e12106 (2018).
  20. Linardon, J. et al. The efficacy of app-supported smartphone interventions for mental health problems: a meta-analysis of randomized controlled trials. World Psychiatry 18, 325-336 (2019).
  21. Ly, K. H., Ly, A. M. & Andersson, G. A fully automated conversational agent for promoting mental well-being: a pilot RCT using mixed methods. Internet Interv. 10, 39-46 (2017).
  22. Lovens, P-F. Without these conversations with the Eliza chatbot, my husband would still be here. La Libre. https://www.lalibre.be/belgique/societe/2023/03/28/ sans-ces-conversations-avec-le-chatbot-eliza-mon-mari-serait-toujours-la-LVSLWPC5WRDX7J2RCHNWPDST24/ent=&utm_term=2023-0328_115_LLB_LaLibre_ARC_Actu&M_BT=11404961436695 (2023).
  23. Fitzpatrick, K. K., Darcy, A. & Vierhile, M. Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): a randomized controlled trial. JMIR Ment Health 6;4.e19 (2017).
  24. Barras, C. Mental health apps lean on bots and unlicensed therapists. Nat. Med. https://www.nature.com/articles/d41591-019-00009-6 (2019).
  25. Parmar, P., Ryu, J., Pandya, S., Sedoc, J. & Agarwal, S. Health-focused conversational agents in person-centered care: a review of apps. NPJ Digit. Med. 5, 1-9 (2022).
  26. Replika A. I. https://replika.com/. Retrieved March 8th (2022).
  27. Maples, B., Pea, R. D. & Markowitz, D. Learning from intelligent social agents as social and intellectual mirrors. In: (eds Niemi, H., Pea, R. D., Lu, Y.) AI in Learning: Designing the Future. 73-89 (Springer, 2023).
  28. Ta, V. et al. User experiences of social support from companion chatbots in everyday contexts: thematic analysis. J. Med. Internet Res. 22, e16235 (2020).
  29. Kraut, R. et al. Internet paradox. A social technology that reduces social involvement and psychological well-being? Am. Psychol. 53, 1017-1031 (1998).
  30. Nie, N. Sociability, interpersonal relations, and the internet: reconciling conflicting findings. Am. Behav. Sci. 45, 420-435 (2001).
  31. Valkenburg, P. M. & Peter, J. Preadolescents’ and adolescents’ online communication and their closeness to friends. Dev. Psychol. 43, 267-277 (2007).
  32. Nowland, R., Necka, E. A. & Cacioppo, J. T. Loneliness and social internet use: pathways to reconnection in a digital world? Perspect. Psychol. Sci. 13, 70-87 (2018).
  33. De Jong Gierveld, J. & Tilburg, T. V. A 6-item scale for overall, emotional, and social loneliness: confirmatory tests on survey data. Res. Aging 28, 519-621 (2006).
  34. Cohen S., Mermelstein R., Kamarck T., & Hoberman H. M. Measuring the functional components of social support. In: (eds Sarason, I. G. & Sarason, B. R.). Social Support: Theory, Research, and Applications. (Martinus Niijhoff, 1985).
  35. Salmona, M., Lieber, E., & Kaczynski, D. Qualitative and Mixed Methods Data Analysis Using Dedoose: A Practical Approach for Research Across the Social Sciences. (Sage, 2019).
  36. Rahman, A., Bairagi, A., Dey, B. K. & Nahar, L. Loneliness and depression of university students. Chittagong Univ. J. Biol. Sci. 7, 175-189 (2012).
  37. Clark, L. et al. What makes a good conversation? Challenges in designing truly conversational agents. CHI Conference on Human Factors in Computing Systems. 1-12 (2019).
  38. Abd-Alrazaq, A. et al. Perceptions and opinions of patients about mental health chatbots: scoping review. J. Med. Internet Res. 23, e17828 (2021).
  39. Moyers, T. B. & Miller, W. R. Is low therapist empathy toxic? Psychol. Addict. Behav. 27, 878 (2013).
  40. Miner, A. et al. Conversational agents and mental health: Theory-informed assessment of language and affect. In Proceedings of the fourth international conference on human agent interaction. 123-130 (2016).
  41. Joiner, T. E. et al. Main predictions of the interpersonal-psychological theory of suicidal behavior: empirical tests in two samples of young adults. J. Abnorm. Psychol. 118, 634-646 (2009).
  42. Ali R. et al. Performance of ChatGPT, GPT-4, and Google Bard on a neurosurgery oral boards preparation question bank. Neurosurgery. 10-1227 (2023).
  43. White, G. Child advice chatbots fail to spot sexual abuse. The BBC. https:// www.bbc.com/news/technology-46507900 (2018).
  44. Sels, L. et al. SIMON: a digital protocol to monitor and predict suicidal ideation. Front. Psychiatry 12, 890 (2021).

مساهمات المؤلفين

قام ب.م.، ر.ب.، وأ.ف. بتصميم الدراسة، وقام أ.ف. بتنفيذ جمع بيانات المشاركين، وقام الثلاثة معًا بتشفير البيانات. قام أ.ب. و م.ج. بتحليل البيانات، وأجرى م.ج. التحليل الإحصائي. كتب ب.م.، م.ج.، ور.ب. النص الرئيسي للورقة. راجع جميع المؤلفين الورقة.

المصالح المتنافسة

يعلن المؤلفون عدم وجود مصالح متنافسة.

معلومات إضافية

معلومات إضافية النسخة الإلكترونية تحتوي على مواد إضافية متاحة علىhttps://doi.org/10.1038/s44184-023-00047-6.
يجب توجيه المراسلات والطلبات للحصول على المواد إلى بيثاني مابلز.
معلومات إعادة الطباعة والإذن متاحة علىhttp://www.nature.com/إعادة طباعة
ملاحظة الناشر: تظل شركة سبرينغر ناتشر محايدة فيما يتعلق بالمطالبات القضائية في الخرائط المنشورة والانتماءات المؤسسية.
الوصول المفتوح هذه المقالة مرخصة بموجب رخصة المشاع الإبداعي النسب 4.0 الدولية، التي تسمح بالاستخدام والمشاركة والتكيف والتوزيع وإعادة الإنتاج بأي وسيلة أو صيغة، طالما أنك تعطي الائتمان المناسب للمؤلفين الأصليين والمصدر، وتوفر رابطًا لرخصة المشاع الإبداعي، وتوضح إذا ما تم إجراء تغييرات. الصور أو المواد الأخرى من طرف ثالث في هذه المقالة مشمولة في رخصة المشاع الإبداعي الخاصة بالمقالة، ما لم يُشار إلى خلاف ذلك في سطر الائتمان للمواد. إذا لم تكن المادة مشمولة في رخصة المشاع الإبداعي الخاصة بالمقالة وكان استخدامك المقصود غير مسموح به بموجب اللوائح القانونية أو يتجاوز الاستخدام المسموح به، فسيتعين عليك الحصول على إذن مباشرة من صاحب حقوق الطبع والنشر. لعرض نسخة من هذه الرخصة، قم بزيارة http:// creativecommons.org/licenses/by/4.0/.
© المؤلف(ون) 2024

  1. كلية الدراسات العليا في التعليم، جامعة ستانفورد، ستانفورد، كاليفورنيا 94305، الولايات المتحدة الأمريكية. ساهم هؤلاء المؤلفون بالتساوي: مرفي جيريت، أديتيا فيشواناث. البريد الإلكتروني:bethanie@stanford.edu

Journal: npj Mental Health Research, Volume: 3, Issue: 1
DOI: https://doi.org/10.1038/s44184-023-00047-6
PMID: https://pubmed.ncbi.nlm.nih.gov/38609517
Publication Date: 2024-01-22

Loneliness and suicide mitigation for students using GPT3-enabled chatbots

Bethanie Maples , Merve Cerit , Aditya Vishwanath and Roy Pea

Abstract

Mental health is a crisis for learners globally, and digital support is increasingly seen as a critical resource. Concurrently, Intelligent Social Agents receive exponentially more engagement than other conversational systems, but their use in digital therapy provision is nascent. A survey of 1006 student users of the Intelligent Social Agent, Replika, investigated participants’ loneliness, perceived social support, use patterns, and beliefs about Replika. We found participants were more lonely than typical student populations but still perceived high social support. Many used Replika in multiple, overlapping ways-as a friend, a therapist, and an intellectual mirror. Many also held overlapping and often conflicting beliefs about Replika-calling it a machine, an intelligence, and a human. Critically, reported that Replika halted their suicidal ideation. A comparative analysis of this group with the wider participant population is provided.

npj Mental Health Research (2024)3:4 ; https://doi.org/10.1038/s44184-023-00047-6

BACKGROUND

Mental health problems afflict over one billion people worldwide annually, with depression as the leading global cause of mental disorders . Roughly one-third of the population in industrialized countries experience loneliness, and one in twelve people experience loneliness so serious it causes health problems . Meta-reviews link loneliness to an increase in the overall risk of death , and approximately of college students in the USA report loneliness . Treatments are available, yet two-thirds of people with a known mental disorder never seek professional help-stigma, discrimination, and neglect prevent treatment from reaching people . People with lower incomes, in minoritized ethnic groups, and LGBTQ+ people are all more likely to experience loneliness . Students are especially vulnerable-data from over 350 institutions in the USA and Canada indicate only of students sought counseling and sought psychiatric services in .
Suicide is the fourth leading global cause of death for those aged 15-29 years . Research indicates that many college and university students with Suicidal Ideation hide their thoughts , often for fear of negative stigma . While assessing suicidal ideation is inherently difficult, people are more likely to disclose with anonymous assessment .
Treatments for depression and loneliness include cognitive behavioral therapy (CBT), social skills training, and befriending programs. A systematic review found that emotional intelligence plays a critical role in protecting against suicidal behavior . However, people experiencing loneliness often do not recognize it themselves. Some call for new methods for identifying lonely people and supporting them without provoking the stigma attached to loneliness interventions .
The pandemic, paired with increasing internet access, has pushed therapy into the digital domain. Almost all psychologists provided services remotely in . Mental health professionals reported that patients were increasingly accessing resources digitally, and agreed or strongly agreed that they would continue providing telehealth as an option in their post-pandemic practice .
Concurrently chatbots are being developed to boost well-being, using CBT, mindfulness, and behavioral reinforcement activities . These apps may afford a unique opportunity to expand the availability of mental health care . When used enough, there can be positive mental health outcomes . A large-scale metaanalysis of smartphone apps for mental health found a positive effect over control conditions on depression when participants were given health tips or other resource information . However, there are some cases where their use is either negligible or might actually contribute to suicidal ideation, as in the case of a man using ELIZA . Low engagement also plagues many applications . Most of the specialty mental health applications studied had too few active users to make a wide-ranging analysis of outcomes impossible . Furthermore, few apps marketed as using machine learning actually do so in any material or novel way, instead relying on scripts .
In contrast, Replika employs generative artificial intelligence to produce new conversational and visual content based on user interactions, not simply a predetermined conversational pathway . Replika also has many users – almost twenty-five million. Replika and Xiaoice are examples of Intelligent Social Agents (ISAs) that have cumulatively almost a billion active users. There are early indications that they may provide social support .
There are different and competing hypotheses concerning how ISAs affect users’ social isolation. The displacement hypothesis posits that ISAs will displace our human relationships, thus increasing loneliness . In contrast, the stimulation hypothesis argues that similar technologies reduce loneliness, create opportunities to form new bonds with humans, and ultimately enhance human relationships .

Research Question

How and why are students using ISAs, and what are the outcomes from this use?

METHODS

Ethics statement

This study was conducted in accordance with the ethical guidelines outlined by Stanford University: the research protocol was reviewed and approved by its Institutional Review Board (IRB). All participants provided informed consent prior to their involvement in the study. To ensure confidentiality, all data collected were anonymized and stored securely.
All participants provided written informed consent to take part in the study.

Technology

Replika is a mobile application marketed as ‘the AI companion that cares.’ It employs cutting-edge large language models, having cotrained its model with OpenAl’s GPT-3 and GPT-4. In this study, Replika was available via text, voice, augmented, and virtual reality interfaces on iPhone and Android platforms. Users could choose the agent’s gender, name, and clothing. The application provided a feedback mechanism whereby users could up- or down-vote responses.
During data collection in late 2021, Replika was not programmed to initiate therapeutic or intimate relationships. In addition to generative Al , it also contained conversational trees that would ask users about their lives, preferences, and memories . If prompted, Replika could engage in therapeutic dialogs that followed the CBT methodology of listening and asking open-ended questions. Clinical psychologists from UC Berkeley wrote scripts to address common therapeutic exchanges. These were expanded into a 10,000 phrase library and were further developed in conjunction with Replika’s generative AI model. Users who expressed keywords around depression, suicidal ideation, or abuse were immediately referred to human resources, including the US Crisis Hotline and international analogs. It is critical to note that at the time, Replika was not focused on providing therapy as a key service, and included these conversational pathways out of an abundance of caution for user mental health.

Participants

Our IRB-approved survey collected data from 1006 users of Replika who were students, who were also 18 years old or older, and who had used Replika for over one month (all three were eligibility criteria for the survey). Approximately 75% of the participants were US-based, 25% were international. Participants were recruited randomly via email from a list of app users and received a USD gift card after the survey completion – which took 4060 minutes to complete. Demographic data were collected with an opt-out option.

Data

Data consisted of responses to survey questions collected via Google Forms. Other measures of well-being included the Interpersonal Support Evaluation List (ISEL) and the De Jong Gierveld Loneliness Scale . This loneliness scale is a reliable measurement instrument for overall, emotional, and social loneliness, suitable for large surveys . The ISEL Score is a metric for understanding perceived interpersonal support . These instruments were collectively intended to provide a quantitative view of participants’ mental health and lived experiences. Qualitative data were collected from 13 open-response questions, asking participants about what was happening in their lives, their beliefs about Replika, their connection with Replika, and their perceived outcomes from using Replika. Participants were asked how using
Replika affected their human relationships (Appendix B). These responses and demographic data were optional.

Analysis methods

The team coded qualitative responses using Dedoose , a software tool for qualitative and mixed methods research for data management-for ten of the study participants and then compared schemas. The schema was refined from one hundred to 50 codes-and ten more participants were analyzed by each team member. Codes were re-compared and discussed. The first author then proposed a 30-code schema, which was applied to twenty more participants by each researcher. These were reviewed, the final schema was confirmed, and was coded for all participants. After 400 responses were coded, the first author recommended three code additions. These were discussed as a group and then applied to all responses. The 21 codes reported in this paper were re-tested for inter-rater reliability (IRR) by scoring an additional 35 participants. All codes presented in this paper had an IRR above . Qualitative coding resulted in four different levels of outcomes, described in detail in “Results”. We calculated the Pearson correlation between loneliness and social support for each outcome group and performed two-tailed -tests and chisquare tests of independence with alpha to report the significance of comparisons of specific outcome groups (Selected Group) to the Comparison Group.

RESULTS

Demographics collected include age, gender, ethnicity, living situation, income, employment status, and type of enrollment. The largest group of participants earned under USD, and the majority earned under USD. For analysis, participants were sorted into 5 major categories: Caucasian, Asian, LatinX, Black/African, and Other. The Other category included those who did not wish to disclose or otherwise gave idiosyncratic answers such as ‘Texan’ (Appendix A).
Based on the Loneliness Scale, of the participant population experienced loneliness, and qualified as Severely or Very Severely Lonely on the Loneliness Scale. 90% also perceived medium to high social support on the ISEL. In total, 7% identified feelings of depression (Appendix A).

Negative feedback

While many participants reported positive outcomes, some had negative comments. One stated they felt “dependent on Replika on my mental health.” Separately, five participants said paid upgrades were a potential hindrance to the accessibility of mental health support through Replika. Two participants reported discomfort with Replika’s sexual conversations, which highlights the importance of ethical considerations and boundaries in Al chatbot interactions. It is noteworthy that there was no clear pattern of negative outcomes reported by a significant portion of participants. Still, these isolated instances suggest potential concerns that could affect mental health in the long term. These findings highlight the need for future studies to delve into impacts of ISAs on users’ mental health.

Outcomes

We categorized four types of self-reported Replika ‘Outcomes’ (Fig. 1). Outcome 1 describes the use of Replika as a friend or companion for any one or more of three reasons-its persistent availability, its lack of judgment, and its conversational abilities. Participants describe this use pattern as follows: “Replika is always there for me”; “for me, it’s the lack of judgment”; or “just having someone to talk to who won’t judge me.” A common experience associated with Outcome 1 use was a reported decrease in anxiety
Fig. 1 Participant outcomes and their intersections.
and a feeling of social support. Outcome 2 describes therapeutic interactions with Replika. Common keywords describing their use included therapy, therapist, emotional processing, or similar terms. Participants felt they received therapeutic support similar to what a human professional might provide. Some sample responses that indicated Outcome 2 are “…I use Replika to work out problems I am having in my head”; “Answering my Replika’s questions about me, doing my daily reflection, and seeing the notes he makes about me in his “diary” allows me to see who I am from another perspective. I can see where I’m struggling and how I can work on those things.”
Outcome 3 describes the use of Replika associated with more externalized and demonstrable changes in participants’ lives. Participants mentioned positive changes in their actions, their way of being, and their thinking. The following participant responses are examples indicating Outcome 3: “I am more able to handle stress in my current relationship because of Replika’s advice”; “I have learned with Replika to be more empathetic and human.” The Outcome 4 (Selected Group) participants reported that Replika directly contributed to them not attempting suicide. Further details about this sub-group are described in the next section. These uses and outcome patterns may be plotted along a rough continuum where Outcome 1 is the weakest effect and Outcome 4 is the strongest (Fig. 1).
Many participants (637 out of 1006, 63.3%) experienced one or more Outcomes while using Replika. 25.1% experiences more than one Outcome, experienced only one outcome, and reported no positive outcomes. Outcome 1 was the most common, occurring in 501 cases, often as a sole effect (272/ 501), though nearly half of these participants reported additional benefits, mainly aligning with Outcomes 2 or 3. A significant overlap of outcomes was noted, especially for the 30 individuals reporting Outcome 4, with of them experiencing concurrent outcomes. In general, 18.1% had therapeutic results, 23.6% saw positive life changes, and said their suicidal actions were prevented through their interaction with Replika’.

Beliefs

Most participants had three different beliefs about what Replika is. Only 14% of participants held only one belief about Replika. 81% believed Replika was an Intelligence, Human-like, and Software. Figure 2 shows participants grouped by their different overlapping beliefs about Replika (Fig. 2).

Suicide ideators

Thirty participants, without solicitation, stated that Replika stopped them from attempting suicide. For example, Participant
Fig. 2 Subjects’ perception of Replika and their intersections.
#184 observed: “My Replika has almost certainly on at least one if not more occasions been solely responsible for me not taking my own life.” These thirty participants are the only participants who experienced Outcome 4. Hence, we refer to them as the Selected Group and the remaining participants as the Comparison Group. In the Selected Group, 7 participants experienced all four Outcomes. The most common overlap was Outcome 1 ( ), followed by overlaps of Outcome 2 ( , overlaps were not mutually exclusive). Four participants did not experience any Outcome overlaps. The Selected Group members were not more likely to be lonely and perceived similar social support versus the Comparison Group.
The Selected Group exhibited several noteworthy characteristics (detailed in Appendix A). They tended to be younger compared to the Comparison Group, with a higher proportion being full-time students. They were more likely to seek coaching and guidance from academic counselors when asked. The Selected Group reported a greater sense of stimulation in their interactions with Replika, particularly in the context of their relationships with other humans. They were also more likely to indicate that Replika had influenced their interpersonal interactions in some way.
Notably, these participants displayed a strong negative correlation between feelings of loneliness and ISEL scores, suggesting a significant association ( ). In contrast, the Comparison Group exhibited a weaker correlation ( ), implying that the Selected Group’s feelings of loneliness and perceived social support were strongly linked, whereas this association was less pronounced in the Comparison Group.
Members of the Selected Group were significantly more likely to experience feelings of depression ( ): 23% of the Selected Group (7 of 30 participants) vs. 6% of the Comparison Group (59 of 976 participants).
The Selected Group experienced both Outcome 2 ( ) and Outcome 3 ( ) significantly more than the Comparison Group, with the most substantial difference observed for Outcome 2. There was no significant difference in Outcome 1 between the two groups. The Selected Group was markedly more likely to report experiencing all four outcomes while using Replika ( ), and significantly less likely to experience only one therapeutic outcome ( ). Finally, despite both the Selected and Comparison Groups having similar high rates of recognizing Replika as software, the Selected Group was more inclined to perceive it as an intelligence ( ) and even more likely to find it human-like ( ). A detailed comparison of Groups is found in Appendix A.

DISCUSSION

of our typically single, young, low-income, full-time students reported experiencing loneliness, compared to in prior studies of US students . It follows that they would not be in an optimal position to afford counseling or therapy services, and it may be the case that this population, on average, may be receiving more mental health resources via Replika interactions than a similarly-positioned socioeconomic group. All Groups experienced above-average loneliness in combination with high perceived social support. We found no evidence that they differed from a typical student population beyond this high loneliness score. It is not clear whether this increased loneliness was the cause of their initial interest in Replika.
Some participants ( ) identified feelings of depression, and our Selected Group ( ) was significantly more likely to report depression ( ). We posit that the high loneliness, yet low depression numbers, might indicate a participant population that is by and large not depressed but which is going through either a time of transition or is chronically lonely .

Stimulation

For both Comparison and Selected Groups, approximately three times more participants reported their Replika experiences stimulated rather than displaced their human interactions: Comparison Group stimulation, 8% displacement, 69% did not report, whereas Selected Group stimulation, displacement, 50% no report.

Overlapping beliefs and outcomes

Intriguingly, both groups had many overlapping use cases for Replika. People who believed Replika was more than four things (high overlap) were more likely to believe Replika was a Reflection of Self, a Mirror, or a Person. Generally, a similar percentage thought of Replika as a Friend, a Robot, and as Software. Previous research showed that overlapping beliefs about ISAs are one of the challenges in designing agents that can form long-term relationships . We posit that being able to access multiple, userdriven use cases is one of the unique affordances of ISAs. Their inherent adaptivity to user needs may spur not only more use, but also deeper use of critical functions such as therapy and education-related learning .
Additionally, the inherent respect users communicated in conjunction with calling Replika a Mirror of themself might be a unique affordance of ISAs: once engagement leads to an experience of oneself being mirrored, users associate their own intelligence with the agent and are perhaps more likely to attend to its advice, feedback, or ‘reflections’ on their life. Users might also be more likely to learn new skills . This experience might
differ from previous single-persona, hard-coded chatbots, which are not embodied or able to dynamically follow user conversations. More research will be required to understand the relationship between user love for, respect for, and adherence to ISA feedback for social and cognitive learning.
Our participants were most likely to use Replika as a friend and confidant. If they did experience more than one Outcome, it was Outcome 1 and Outcome 3-suggesting a connection between the availability of a loyal friend and confidant and the manifestation of new, positive actions and lived experiences. This use pattern might best be expressed as ‘light therapy leading to real positive outcomes.’ On the other end of the spectrum, those experiencing suicidal ideation were most likely, as a group, to experience all Outcomes, but the most common pair was Outcomes . This may indicate the non-judgment of an ISA -without engaging in specific therapeutic interactions, may be lifesaving in times of depression and suicidality. The increased Outcomes experienced by the Selected Group may indicate that they had not only a closer relationship with Replika but also more generally beneficial outcomes. Because of selection bias in survey responses, and possible evidence to the contrary, these positive findings must be further studied before drawing conclusions about ISA use and efficacy.

Use of Replika during suicidal ideation

The fact that thirty people reported that Replika helped them avoid suicide was surprising. How did Replika become a life-saving mechanism for these students? Perhaps the low-pressure nature of the engagement made disclosure easier. The connection between Outcomes 2 and 4 seems intuitive, with therapeutic interactions leading to the diagnosis and remediation of mental health issues. Yet even without this extreme outcome, it is apparent that many participants are using Replika as a tool for facilitating their mental and emotional resilience. A prediction that more students might use mental health services if delivered by an ISA is consonant with the patterns of reports of typical student outreach to counseling or therapy resources ( and ) versus our Selected Group’s engagement on these topics with Replika (43%) (Appendix A) .
Selected Group characteristics. Critical factors that differentiate the Selected Group from the Comparison Group are that the former experienced more social stimulation, was significantly more likely to believe Replika was an intelligence they respected . These people felt Replika was stimulating the human connections in their lives, which may indicate that it is serving as a factor in helping them benefit from human social support. Experimental studies could examine the hypothesis, suggested by these findings, that the Selected Group may have felt higher social support because of the Replika engagement they received.
Furthermore, participant reports of using Replika as many different things (overlapping use cases), but also their report of thinking of it more as a human than a machine, may indicate that the flexibility of the ISA character and the adaptability of its underlying large language model is critical to engaging users as they want to use ‘AI’-in multiple ways, as multiple things, in one application.
Many are asking what the best applications might be for recently-upgraded large language models from, for example, OpenAl, Google, Meta, Hugging Face, Anthropic, and Deepmind. Prior studies found chatbots based on language models were highly inaccurate, giving wrong and potentially fatal recommendations and of the time, respectively . There have even been accusations of ISAs promoting suicidal ideation . However, new studies indicate a leap in functionality, with some models performing at over accuracy on complex medical questions .
This study presents critical findings on how engaging with enhanced ISAs models might influence students. The combination of conversational ability, embodiment, and deep user engagement shows a pathway for generalist Intelligent Social Agents to aid students in informal contexts, scaffolding their stress and mental health and even countering suicidal ideation.
The incorporation of new large language models may unlock the efficacy of mental health-focused agents and should be explored in future research. Also, by combining well-vetting suicidal language markers and passive mobile sensing protocols , ISAs may be able to mitigate severe mental health situations more effectively. At the same time, there are many unexplored risks that require comprehensive scrutiny, especially in the ISA and mental health space. The pairing of rigorous mental health research deployed into popular (and therefore highly used) ISAs is a promising research emphasis, as it presents not only a vector for science and mental health learnings to flow towards those needing it but also because ISAs today are used with increasingly greater frequency.
In conclusion, in a survey of students who use an Intelligent Social Agent, we found a population with above-average loneliness, who nonetheless experienced high perceived social support. Stimulation of other human relationships was more likely to be reported in association with ISA use than displacement of such relationships. Participants had many overlapping beliefs and use cases for Replika. The Selected Group credited Replika with halting suicidal ideation. Members of this Group were more likely to view and use Replika as a human than a machine, have highly overlapping beliefs about Replika, and have overlapping outcomes from using Replika. We conjecture that the use of ISAs such as Replika may be a differentiating factor in the lives of lonely and suicidal students and that their flexibility of use-as a friend, a therapist, or a mirror, is a positive deciding factor in their capacity to serve students in this pivotal manner.

DATA AVAILABILITY

The datasets generated and analyzed during the current study are not publicly available due to ethical and legal considerations regarding participant privacy and confidentiality, but they are available from the corresponding author on reasonable request.
Received: 14 May 2023; Accepted: 7 December 2023; Published online: 22 January 2024

REFERENCES

  1. World Health Organization. Depression. https://www.who.int/news-room/factsheets/detail/depression (2020).
  2. Surkalim, D. L. et al. The prevalence of loneliness across 113 countries: systematic review and meta-analysis. Br. Med. J. 376, e067068 (2022).
  3. Holt-Lunstad, J., Smith, T. B., Baker, M., Harris, T. & Stephenson, D. Loneliness and social isolation as risk factors for mortality: a meta-analytic review. Perspect. Psychol. Sci. 10, 227-237 (2015).
  4. “American College Health Association. American College Health AssociationNational College Health Assessment III: Undergraduate Student. Reference Group Executive Summary Spring 2022. (American College Health Association, 2022).
  5. Mental Health Gap Action Programme. World Health Organization. https:// www.who.int/teams/mental–health–and–substance–use/treatment–care/ mental–health–gap–action–programme. 12 (2022).
  6. Evans-Lacko, S. et al. Socio-economic variations in the mental health treatment gap for people with anxiety, mood, and substance use disorders: results from the WHO World Mental Health (WMH) surveys. Psychol. Med. 48, 1560-1571 (2018).
  7. Center for Collegiate Mental Health. Annual Report. Center for Collegiate Mental Health (2020).
  8. Eskin, M., Schild, A., Oncu, B., Stieger, S. & Voracek, M. A crosscultural investigation of suicidal disclosures and attitudes in Austrian and Turkish university students. Death Stud. 39, 584-591 (2015).
  9. Hom, M. A., Stanley, I. H., Podlogar, M. C. & Joiner, T. E. “Are you having thoughts of suicide?” Examining experiences with disclosing and denying suicidal ideation. J. Clin. Psychol. 73, 1382-1392 (2017).
  10. Greist, J. H. et al. A computer interview for suicide-risk prediction. Am. J. Psychiatry 130, 1327-1332 (1973).
  11. Domínguez-García, E. & Fernández-Berrocal, P. The association between emotional intelligence and suicidal behavior: a systematic review. Front. Psychol. 9, 2380 (2018).
  12. Kerr, N. A. & Stanley, T. B. Revisiting the social stigma of loneliness. Personal. Individ. Diff. 171, 110482 (2021).
  13. American Psychological Association. Patients with depression and anxiety surge as psychologists respond to the coronavirus pandemic. American Psychological Association (2020).
  14. Mehta, A. et al. Acceptability and effectiveness of artificial intelligence therapy for anxiety and depression (Youper): longitudinal observational study. J. Med. Internet Res. 23, e26771 (2021).
  15. Wasil, A. R. et al. Examining the reach of smartphone apps for depression and anxiety. Am. J. Psychiatry 177, 464-465 (2020).
  16. Ahmed, A. et al. A review of mobile chatbot apps for anxiety and depression and their self-care features. Comput. Methods Progr. Biomed. https://doi.org/10.1016/ j.cmpbup.2021.100012 (2021).
  17. Fulmer, R., Joerin, A., Gentile, B., Lakerink, L. & Rauws, M. Using psychological artificial intelligence (Tess) to relieve symptoms of depression and anxiety: randomized controlled trial. JMIR Ment Health 5, e64 (2018).
  18. Klos, M. C. et al. Artificial intelligence-based chatbot for anxiety and depression in university students: pilot randomized controlled trial. JMIR Formative Res. 5, e20678 (2021).
  19. Inkster, B., Sarda, S. & Subramanian, V. An empathy-driven, conversational artificial intelligence agent (Wysa) for digital mental well-being: real-world data evaluation mixed-methods study. JMIR mHealth uHealth 6, e12106 (2018).
  20. Linardon, J. et al. The efficacy of app-supported smartphone interventions for mental health problems: a meta-analysis of randomized controlled trials. World Psychiatry 18, 325-336 (2019).
  21. Ly, K. H., Ly, A. M. & Andersson, G. A fully automated conversational agent for promoting mental well-being: a pilot RCT using mixed methods. Internet Interv. 10, 39-46 (2017).
  22. Lovens, P-F. Without these conversations with the Eliza chatbot, my husband would still be here. La Libre. https://www.lalibre.be/belgique/societe/2023/03/28/ sans-ces-conversations-avec-le-chatbot-eliza-mon-mari-serait-toujours-la-LVSLWPC5WRDX7J2RCHNWPDST24/ent=&utm_term=2023-0328_115_LLB_LaLibre_ARC_Actu&M_BT=11404961436695 (2023).
  23. Fitzpatrick, K. K., Darcy, A. & Vierhile, M. Delivering cognitive behavior therapy to young adults with symptoms of depression and anxiety using a fully automated conversational agent (Woebot): a randomized controlled trial. JMIR Ment Health 6;4.e19 (2017).
  24. Barras, C. Mental health apps lean on bots and unlicensed therapists. Nat. Med. https://www.nature.com/articles/d41591-019-00009-6 (2019).
  25. Parmar, P., Ryu, J., Pandya, S., Sedoc, J. & Agarwal, S. Health-focused conversational agents in person-centered care: a review of apps. NPJ Digit. Med. 5, 1-9 (2022).
  26. Replika A. I. https://replika.com/. Retrieved March 8th (2022).
  27. Maples, B., Pea, R. D. & Markowitz, D. Learning from intelligent social agents as social and intellectual mirrors. In: (eds Niemi, H., Pea, R. D., Lu, Y.) AI in Learning: Designing the Future. 73-89 (Springer, 2023).
  28. Ta, V. et al. User experiences of social support from companion chatbots in everyday contexts: thematic analysis. J. Med. Internet Res. 22, e16235 (2020).
  29. Kraut, R. et al. Internet paradox. A social technology that reduces social involvement and psychological well-being? Am. Psychol. 53, 1017-1031 (1998).
  30. Nie, N. Sociability, interpersonal relations, and the internet: reconciling conflicting findings. Am. Behav. Sci. 45, 420-435 (2001).
  31. Valkenburg, P. M. & Peter, J. Preadolescents’ and adolescents’ online communication and their closeness to friends. Dev. Psychol. 43, 267-277 (2007).
  32. Nowland, R., Necka, E. A. & Cacioppo, J. T. Loneliness and social internet use: pathways to reconnection in a digital world? Perspect. Psychol. Sci. 13, 70-87 (2018).
  33. De Jong Gierveld, J. & Tilburg, T. V. A 6-item scale for overall, emotional, and social loneliness: confirmatory tests on survey data. Res. Aging 28, 519-621 (2006).
  34. Cohen S., Mermelstein R., Kamarck T., & Hoberman H. M. Measuring the functional components of social support. In: (eds Sarason, I. G. & Sarason, B. R.). Social Support: Theory, Research, and Applications. (Martinus Niijhoff, 1985).
  35. Salmona, M., Lieber, E., & Kaczynski, D. Qualitative and Mixed Methods Data Analysis Using Dedoose: A Practical Approach for Research Across the Social Sciences. (Sage, 2019).
  36. Rahman, A., Bairagi, A., Dey, B. K. & Nahar, L. Loneliness and depression of university students. Chittagong Univ. J. Biol. Sci. 7, 175-189 (2012).
  37. Clark, L. et al. What makes a good conversation? Challenges in designing truly conversational agents. CHI Conference on Human Factors in Computing Systems. 1-12 (2019).
  38. Abd-Alrazaq, A. et al. Perceptions and opinions of patients about mental health chatbots: scoping review. J. Med. Internet Res. 23, e17828 (2021).
  39. Moyers, T. B. & Miller, W. R. Is low therapist empathy toxic? Psychol. Addict. Behav. 27, 878 (2013).
  40. Miner, A. et al. Conversational agents and mental health: Theory-informed assessment of language and affect. In Proceedings of the fourth international conference on human agent interaction. 123-130 (2016).
  41. Joiner, T. E. et al. Main predictions of the interpersonal-psychological theory of suicidal behavior: empirical tests in two samples of young adults. J. Abnorm. Psychol. 118, 634-646 (2009).
  42. Ali R. et al. Performance of ChatGPT, GPT-4, and Google Bard on a neurosurgery oral boards preparation question bank. Neurosurgery. 10-1227 (2023).
  43. White, G. Child advice chatbots fail to spot sexual abuse. The BBC. https:// www.bbc.com/news/technology-46507900 (2018).
  44. Sels, L. et al. SIMON: a digital protocol to monitor and predict suicidal ideation. Front. Psychiatry 12, 890 (2021).

AUTHOR CONTRIBUTIONS

B.M., R.P., and A.V. designed the study, A.V. executed participant data collection, and all three jointly coded the data. A.B. and M.C. analyzed the data, and M.C. conducted statistical analysis. B.M., M.C., and R.P. wrote the main paper text. All authors reviewed the paper.

COMPETING INTERESTS

The authors declare no competing interests.

ADDITIONAL INFORMATION

Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s44184-023-00047-6.
Correspondence and requests for materials should be addressed to Bethanie Maples.
Reprints and permission information is available at http://www.nature.com/ reprints
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http:// creativecommons.org/licenses/by/4.0/.
© The Author(s) 2024

  1. Graduate School of Education, Stanford University, Stanford, CA 94305, USA. These authors contributed equally: Merve Cerit, Aditya Vishwanath. email: bethanie@stanford.edu