المواقف تجاه الذكاء الاصطناعي: القياس والارتباطات بالشخصية Attitudes towards AI: measurement and associations with personality

المجلة: Scientific Reports، المجلد: 14، العدد: 1
DOI: https://doi.org/10.1038/s41598-024-53335-2
PMID: https://pubmed.ncbi.nlm.nih.gov/38316898
تاريخ النشر: 2024-02-05

افتح

المواقف تجاه الذكاء الاصطناعي: القياس والارتباطات بالشخصية

يان-فيليب شتاين , تانجا ميسينغشلاغر , تيمو غنامبس , فابيان هوتماشر & ماركوس أبيل

لقد أصبح الذكاء الاصطناعي (AI) جزءًا لا يتجزأ من العديد من التقنيات المعاصرة، مثل منصات الوسائط الاجتماعية، والأجهزة الذكية، وأنظمة اللوجستيات العالمية. في الوقت نفسه، تُظهر الأبحاث حول قبول الجمهور للذكاء الاصطناعي أن العديد من الناس يشعرون بالقلق بشأن إمكانيات هذه التقنيات – وهي ملاحظة تم ربطها بكل من المتغيرات الديموغرافية والاجتماعية الثقافية للمستخدمين (مثل العمر، والتعرض السابق للوسائط). ومع ذلك، بسبب القياسات المتباينة وغالبًا ما تكون عشوائية للمواقف المتعلقة بالذكاء الاصطناعي، تظل الأدلة الحالية غير حاسمة. وبالمثل، لا يزال غير واضح ما إذا كانت المواقف تجاه الذكاء الاصطناعي تتأثر أيضًا بسمات شخصية المستخدمين. استجابةً لهذه الفجوات البحثية، نقدم مساهمة مزدوجة. أولاً، نقدم استبيانًا جديدًا مستندًا إلى علم النفس (ATTARI-12) يلتقط المواقف تجاه الذكاء الاصطناعي كهيكل واحد، مستقل عن السياقات أو التطبيقات المحددة. بعد أن لاحظنا موثوقية وصلاحية جيدة لمقياسنا الجديد عبر دراستين ( )، نفحص عدة سمات شخصية – الخمسة الكبار، والثالوث المظلم، وعقلية المؤامرة – كمتنبئين محتملين للمواقف المتعلقة بالذكاء الاصطناعي في دراسة ثالثة ( ). نجد أن الود والعمر الأصغر يتنبأان برؤية أكثر إيجابية تجاه التكنولوجيا الذكية الاصطناعية، بينما يرتبط القابلية للاعتقاد في المؤامرات بموقف أكثر سلبية. يتم مناقشة نتائجنا مع مراعاة القيود المحتملة والاتجاهات المستقبلية للبحث والممارسة.
يعد الذكاء الاصطناعي (AI) وعدًا لجعل حياة الإنسان أكثر راحة بكثير: قد يعزز التعاون بين الثقافات عبر أدوات الترجمة المتطورة، ويوجه العملاء عبر الإنترنت نحو المنتجات التي من المرجح أن يشتروا، أو ينفذ وظائف تبدو مملة للعمال البشر . في الوقت نفسه، يتم مواجهة انتشار تقنيات الذكاء الاصطناعي بمخاوف جدية. يُقال، على سبيل المثال، إن الذكاء الاصطناعي قد يؤدي إلى تقليص الوظائف البشرية ، وإنشاء أسلحة ذكية جديدة ، أو نقص متزايد في السيطرة على التقنيات الناشئة . وبالمثل، أثار العرض الأخير والمشهور للذكاء الاصطناعي الجديد الذي ينشئ نصوصًا وصورًا (مثل ChatGPT وMidjourney) مخاوف بشأن الترخيص الفني، والغش الأكاديمي، وقيمة الإبداع البشري .
من المهم أن الأفراد قد يختلفون في تقييمهم لهذه الفرص والمخاطر، وبالتالي، يحملون مواقف مختلفة تجاه الذكاء الاصطناعي. ومع ذلك، نلاحظ أن العلماء المهتمين بهذه الاختلافات غالبًا ما استخدموا نهجًا أساسيًا عشوائيًا لتقييم مواقف المشاركين تجاه الذكاء الاصطناعي ، أو ركزوا فقط على أنواع فرعية محددة للغاية من . في هذا السياق، تقدم الورقة الحالية مساهمة مزدوجة. أولاً، نقدم استبيانًا جديدًا حول المواقف المتعلقة بالذكاء الاصطناعي – ATTARI-12، الذي يدمج الثلاثية النفسية للإدراك، والعاطفة، والسلوك لتسهيل قياس شامل ولكنه اقتصادي – وندقق في صلاحيتها عبر دراستين. بعد ذلك، نربط مواقف المشاركين (كما تم قياسها بواسطة مقياسنا الجديد) بعدة سمات شخصية أساسية، وهي الخمسة الكبار، والثالوث المظلم، وعقلية المؤامرة.

هدف البحث 1: قياس المواقف تجاه الذكاء الاصطناعي

بشكل عام، يُستخدم مصطلح الذكاء الاصطناعي (AI) لوصف التقنيات التي يمكن أن تنفذ مهام قد يتوقع أن تتطلب ذكاءً بشريًا – أي، التقنيات التي تمتلك درجة معينة من الاستقلالية، والقدرة على التعلم والتكيف، والقدرة على التعامل مع كميات كبيرة من البيانات . مع تعمق الذكاء الاصطناعي في العديد من الأنظمة التقنية، وبالتالي، في حياة الناس اليومية، يظهر كواجب مهم للعديد من التخصصات العلمية (مثل علم النفس، وعلوم الاتصال، وعلوم الحاسوب، والفلسفة) لـ
لفهم استجابة المستخدمين – وقبولهم – للتكنولوجيا الذكية الاصطناعية بشكل أفضل. بالطبع، من أجل تحقيق هذا الهدف، من الضروري استخدام مقاييس نظرية ذات جودة نفسية عالية عند تقييم المواقف المتعلقة بالذكاء الاصطناعي. علاوة على ذلك، يحتاج الباحثون إلى اتخاذ قرار بشأن ما إذا كانوا يريدون دراسة نوع محدد فقط من التطبيقات (مثل السيارات ذاتية القيادة أو الروبوتات الذكية) أو التركيز على الذكاء الاصطناعي كمفهوم تكنولوجي مجرد يمكن تطبيقه في العديد من الإعدادات المختلفة. بينما تتمتع كلا الطريقتين بمزايا لا يمكن إنكارها، يمكن القول إن قابلية المقارنة بين جهود البحث المختلفة – خاصة تلك الموجودة عند تقاطع تخصصات مختلفة – تستفيد بوضوح من الأخيرة، أي، التركيز التجريبي على المواقف تجاه الذكاء الاصطناعي كمجموعة من القدرات التكنولوجية بدلاً من حالات الاستخدام المحددة.
تماشيًا مع هذا الحجة، أنتجت الجهود العلمية عدة مقاييس لتقييم المواقف تجاه الذكاء الاصطناعي كمفهوم أكثر شمولية. تشمل هذه مقياس المواقف العامة تجاه الذكاء الاصطناعي (GAAIS )، مقياس المواقف تجاه الذكاء الاصطناعي (ATAI )، مقياس قلق الذكاء الاصطناعي (AIAS )، مقياس تهديدات الذكاء الاصطناعي (TAI )، واستبيان المخاوف بشأن التكنولوجيا المستقلة . ومع ذلك، بعد الفحص الدقيق، نعتقد أن أيًا من هذه المقاييس لا يقدم خيارًا مرضيًا تمامًا لواحد على الأقل من خمسة أسباب. أولاً، العديد من المقاييس المذكورة أعلاه تستفسر فقط المشاركين عن الانطباعات السلبية والمخاوف بشأن الذكاء الاصطناعي، متجاهلة إمكانية وجود مواقف إيجابية بشكل واضح . ثانيًا، غالبًا ما يتم تقسيم المواقف المقاسة تجاه الذكاء الاصطناعي إلى عدة عوامل ، مما يعقد التعامل مع هذه المقاييس وتفسيرها من كل من منظور نظري وعملي. على سبيل المثال، يقدم مقياس GAAIS مقياسين فرعيين يسمى “القبول” و”الخوف”، على الرغم من أنه يمكن القول إن هذه تغطي ببساطة أقطاب مختلفة من نفس الطيف. ذات صلة، يقدم مقياس AIAS حلًا رباعي العوامل مع الأبعاد (المستحثة إحصائيًا) “التعلم”، “استبدال الوظائف”، “العمى الاجتماعي التقني”، و”تكوين الذكاء الاصطناعي” – مجموعات قد تكون فائدتها العملية مرتبطة بحالات استخدام أكثر تحديدًا. ثالثًا، بينما تحتوي بعض المقاييس على عدد منخفض نسبيًا من العناصر (مثل خمسة عناصر )، فإن مقاييس أخرى طويلة بالمقارنة (مثل 21 عنصرًا )، مما قد يعقد استخدامها في بيئات البحث التي تحتاج إلى تضمين المواقف تجاه الذكاء الاصطناعي كمفهوم واحد بين عدة مفاهيم أخرى. رابعًا، بالنسبة لأحد المقاييس التي تمت مراجعتها ، أدى العدد المنخفض من العناصر إلى نقص ملحوظ في الموثوقية. خامسًا، لا تعترف أي من المقاييس المتاحة بالجوانب المعرفية، والعاطفية، والسلوكية للمواقف في تصميماتها .
لذا، استنادًا إلى النقص المحدد في مقياس أحادي البعد، موثوق نفسيًا، ولكنه اقتصادي، يلتقط الجوانب الإيجابية والسلبية لموقف تجاه الذكاء الاصطناعي، كان هدفنا الأول هو تطوير مقياس يتجاوز هذه القيود. من أجل قابلية تطبيق واسعة بين التخصصات، ركزنا بشكل صريح على إنشاء مقياسنا على إدراك الذكاء الاصطناعي كمجموعة عامة من القدرات التكنولوجية، مستقلة عن تجسيدها المادي أو سياق استخدامها. لم يكن هذا المنظور مستندًا إلى الأبحاث الحديثة، التي أشارت إلى أن تقييم الناس للعقول الرقمية من المحتمل أن يتغير بمجرد دخول العوامل البصرية في اللعب – مما يبرز المزيد من قيمة معالجة الذكاء الاصطناعي كمفهوم غير مجسد.
حتى الآن، استكشفت معظم الدراسات حول قبول الجمهور للذكاء الاصطناعي دور المتغيرات الديموغرافية والاجتماعية الثقافية، كاشفة أن المخاوف بشأن الذكاء الاصطناعي تبدو أكثر انتشارًا بين النساء، وكبار السن، والأقليات العرقية، والأفراد ذوي المستويات التعليمية المنخفضة . علاوة على ذلك، تم إثبات أن القيم المتساوية وعدم الثقة في العلم ، والقلق من الارتباط ، والتعرض للخيال العلمي الكئيب تشكل مؤشرات ذات مغزى للمواقف السلبية تجاه الذكاء الاصطناعي. في الواقع، تتردد العديد من هذه النتائج في مجموعة كبيرة من الأدبيات المتعلقة بتفاعل الإنسان مع الروبوتات، التي تؤكد أيضًا على التأثير الحاسم للمتغيرات الديموغرافية، والمعايير الثقافية، واستخدام وسائل الإعلام على قبول الجمهور للآلات الروبوتية. . ومع ذلك، بالنظر إلى أن الدراسات من هذا التخصص نادراً ما توضح ما إذا كانت التأثيرات المبلغ عنها تنشأ من إدراك الأجسام الروبوتية أو العقول الروبوتية (أي، الذكاء الاصطناعي)، نقترح أنه يجب اعتبارها دليلاً ثانوياً فقط للموضوع الحالي.
بعيدًا عن الفهم الكبير للمحددات الاجتماعية والثقافية المتعلقة بإعجاب أو عدم إعجاب الأفراد بتكنولوجيا الذكاء الاصطناعي، فإن المعرفة حول تأثير عوامل شخصية المستخدمين على مواقفهم المتعلقة بالذكاء الاصطناعي أقل بكثير. حسب علمنا، فإن عددًا قليلاً فقط من الدراسات العلمية قد تناولت هذه القضية، وقد أسفرت معظمها عن نتائج محدودة جدًا – إما من خلال التركيز على أنواع محددة جدًا من أو من خلال قياس ردود الفعل تجاه الذكاء الاصطناعي المتمثل جسديًا (مثل الروبوتات)، مما يجعل من الصعب جدًا تحديد سبب التأثير. . بينما درست دراسة حديثة أخرى بالفعل المواقف تجاه الذكاء الاصطناعي كمفهوم تقني بغض النظر عن استخدامه أو تجسيده المحدد تركز هذا الجهد بشكل حصري على المواقف السلبية، مما أدى إلى إغفال جزء كبير (أي الجانب الإيجابي) من كيفية تفكير المستخدمين ومشاعرهم تجاه الأنظمة الذكية الاصطناعية.
لذلك، بناءً على العمل الذي تم مراجعته، كان هدفنا الثاني هو فحص عدة أبعاد شخصية مركزية كمتنبئات لمواقف الناس تجاه الذكاء الاصطناعي، بما في ذلك الخمسة الكبار المعروفة. وثلاثي الظلام في الشخصية بالإضافة إلى ذلك، ندرج البناء الجديد إلى حد ما لفكرة المؤامرة في عملنا. مع الأخذ في الاعتبار أهميته المعاصرة العالية في عالم يتزايد رقميته .

نظرة عامة على الدراسات والتوقعات

في معالجة الاقتراحات البحثية الموضحة، نقدم ثلاث دراسات (انظر الجدول 1 للحصول على نظرة عامة موجزة عن الخصائص الرئيسية للدراسة). يمكن العثور على جميع المواد والبيانات التي تم الحصول عليها وأكواد التحليل لهذه الدراسات الثلاث في مستودع إطار العلوم المفتوحة لمشروعنا.https://osf.io/3j67a/تم جمع الموافقة المستنيرة من جميع المشاركين قبل أن يشاركوا في جهودنا البحثية.
الدراستان 1 و2 خدمتا بشكل رئيسي للتحقق من موثوقية وصلاحية مقياسنا الجديد للمواقف تجاه الذكاء الاصطناعي، ATTARI-12. وبالتالي، كانت هذه الدراسات موجهة بشكل رئيسي من خلال عدة خطوات تحليلية تفحص الخصائص الإحصائية لمقياسنا، بما في ذلك صلاحية التقارب، وموثوقية إعادة الاختبار، وقابلية التأثر بتحيز الرغبة الاجتماعية. في الدراسة 3، بدأنا بعد ذلك في ربط موقف المشاركين تجاه الذكاء الاصطناعي (كما تم قياسه من خلال الاستبيان المطور) بالسمات الشخصية الأساسية والعوامل الديموغرافية. في هذه الدراسة الثالثة والأخيرة، سنفصل جميع الاعتبارات النظرية والافتراضات في ما يلي.
الدراسة 1 الدراسة 2 الدراسة 3
نوع العينة MTurk (الولايات المتحدة) عينة الطالب MTurk (الولايات المتحدة)
تركيز الدراسة تطوير المقياس: الصلاحية العاملية، الموثوقية (الاتساق الداخلي)، صلاحية البناء تطوير المقياس: الموثوقية (موثوقية إعادة الاختبار)، الصلاحية البنائية مؤشرات الشخصية (الخمسة الكبار، مثلث الظلام، عقلية المؤامرة)
دراسة اللغة الإنجليزية ألماني الإنجليزية
حجم العينة النهائي ٤٩٠ 166 (في الوقت 1)، 163 في الوقت 2، 150 في كلا الوقتين 298
القياسات 1 2 (متابعة بعد أربعة إلى خمسة أسابيع) 1
الجدول 1. نظرة عامة على الدراسات التي تم إجراؤها.
في مواجهة نقص النتائج السابقة المتعلقة بمؤشرات الشخصية تجاه المواقف المتعلقة بالذكاء الاصطناعي، اعتبرنا أنه من المهم أن نؤسس تحقيقنا على رؤية أوسع لشخصية الإنسان. وقد قادنا هذا المنظور إلى تضمين نموذج الخمسة الكبار أولاً. التي ظلت الأكثر استخدامًا في تصنيف الشخصية البشرية لعدة عقود. كما يوحي الاسم، تتكون الخمسة الكبار من خمسة أبعاد أساسية للشخصية – الانفتاح على التجربة، والضمير، والانبساط، والود، والعصابية – والتي يُفترض أنها تغطي جزءًا كبيرًا من التباين الشخصي بين الأفراد. استنادًا إلى هذا المفهوم، قمنا بتطوير مجموعة أولى من الفرضيات حول المساهمة الفردية لكل سمة من سمات الخمسة الكبرى في مواقف الناس تجاه الذكاء الاصطناعي.
بالنسبة للانفتاح على التجربة – الذي يمكن تعريفه على أنه الميل إلى المغامرة، والفضول الفكري، والخيال – توقعنا وجود علاقة إيجابية واضحة مع المواقف المتعلقة بالذكاء الاصطناعي. نظرًا لأن تقنيات الذكاء الاصطناعي تعد بالعديد من الإمكانيات الجديدة للمجتمع البشري، يجب أن يشعر الأشخاص المنفتحون على التجربة بمزيد من الحماس والفضول تجاه الابتكارات المعنية (وهذا ما تعكسه النتائج الأخيرة حول السيارات ذاتية القيادة). نحن نفترض:
H1 كلما زادت انفتاح الشخص على التجارب، زادت المواقف الإيجابية التي يحملها تجاه الذكاء الاصطناعي.
يمكن وصف الضمير بأنه الميل إلى أن يكون الشخص مجتهدًا وفعالًا وحذرًا، وأن يتصرف بطريقة منضبطة أو حتى مثالية. نظرًا لأنه قد يكون من الصعب على البشر فهم الآليات الداخلية لنظام الذكاء الاصطناعي أو توقع سلوكه، يبدو من المحتمل أن الأشخاص ذوي الضمير سيعبرون عن آراء أكثر سلبية حول الذكاء الاصطناعي، وهو مفهوم تكنولوجي قد يجعل العالم أقل قابلية للفهم بالنسبة لهم. .
H2 كلما زادت درجة الوعي الذاتي لدى الشخص، زادت المواقف السلبية التي يحملها تجاه الذكاء الاصطناعي.
الصفة الثالثة من الصفات الخمس الكبرى، الانفتاح، تشمل الميل إلى الانخراط، والتحدث، والاجتماعية؛ ومن ثم، فإنها ترتبط سلبًا بالسلوك القلق والمقيد. استنادًا إلى هذا التعريف، ليس من المستغرب أن الأفراد المنفتحين أبلغوا عن مخاوف أقل بشأن التقنيات المستقلة في الدراسات السابقة. لذا، نفترض:
H3 كلما زادت انفتاحية الأشخاص، زادت المواقف الإيجابية التي يحملونها تجاه الذكاء الاصطناعي.
تظهر الوداعة العالية في الميل إلى إظهار سلوك دافئ وتعاوني وطيب القلب. فيما يتعلق بالموضوع المطروح، تشير الأبحاث السابقة إلى أن الوداعة قد تكون مرتبطة (بشكل معتدل) بوجهات نظر إيجابية حول الأتمتة أو أنواع معينة من وبناءً على ذلك، نفترض:
H4 كلما زادت درجة توافق الشخص، زادت المواقف الإيجابية التي يحملها تجاه الذكاء الاصطناعي.
السمة الأخيرة من سمات الخمسة الكبار، العصابية، مرتبطة بالسلوك الخجول والواعي للذات. الأشخاص الذين يحصلون على درجات عالية في هذه السمة يميلون إلى أن يكونوا أكثر عرضة للضغوط الخارجية ويواجهون صعوبة في التحكم في الدوافع العفوية. ماك دورمان وإنتيزاري اكتشف أن الأفراد العصبيين أبلغوا عن شعور أقوى بالغرابة بعد مواجهتهم لروبوت مستقل مقارنةً بأولئك الذين لديهم درجات أقل في هذه السمة. ظهرت نتائج مماثلة فيما يتعلق بآراء الناس حول السيارات ذاتية القيادة. وتطبيقات الذكاء الاصطناعي الضيق استنادًا إلى هذه الأدلة بالإضافة إلى الطبيعة القلقة عمومًا للأفراد العصبيين، فإن فرضيتنا هي كما يلي:
H5 كلما زادت درجة العصابية لدى الشخص، زادت المواقف السلبية التي يحملها تجاه الذكاء الاصطناعي.
على الرغم من أن الخمسة الكبار يُعتبرون مجموعة شاملة نسبيًا من سمات الشخصية البشرية، إلا أن الأبحاث قد أسفرت عن عدة مفاهيم أخرى تساعد في فهم الفروق بين الأفراد (وكيف ترتبط هذه الفروق بالمواقف والسلوكيات). ومن المهم، التطرق إلى الجوانب الأكثر سلبية في الطبيعة البشرية، حيث قام بولهاس وويليامز قدمت ثلاثية الظلام للشخصية، وهي تصنيف يجمع بين ثلاث سمات شخصية خبيثة: الميكافيلية، السيكوباتية، والنرجسية. نظرًا لخصائصها المعادية للمجتمع، تم ربط ثلاثية الظلام بشكل متكرر بالسلوكيات المنحرفة، والمشاكل بين الأفراد، وزيادة الصعوبات في مكان العمل. علاوة على ذلك، أصبحت جزءًا مهمًا من استكشاف الفروق بين الأفراد في تشكيل المواقف، بما في ذلك الآراء حول التكنولوجيا الحديثة. نتيجة لذلك، تدور مجموعة فرضياتنا الثانية حول التأثير المحتمل لهذه الصفات الشخصية الانتقامية إلى حد ما.
الميكافيلية هي بُعد شخصي يتضمن صفات متلاعبة، قاسية، وغير أخلاقية. بالنسبة لموضوع بحثنا، تأملنا في فكرتين متعارضتين حول كيفية ارتباط هذه السمة بمواقف الذكاء الاصطناعي. من ناحية، تشير الأدبيات الحديثة إلى أن الأفراد الميكافليين قد يشعرون بقلق أقل بشأن الاستخدامات غير الأخلاقية للذكاء الاصطناعي وبدلاً من ذلك يركزون على قيمته النفعية، مما قد ينعكس في مواقف أكثر إيجابية. من ناحية أخرى، تظل آفاق المراقبة المدفوعة بالذكاء الاصطناعي موضوعًا بارزًا في الخطاب العام. -ورأينا أنه من المحتمل أن يكون الأشخاص الذين يتمتعون بمستوى عالٍ من الميكافيلية حذرين من ذلك، ويرغبون في أن تظل أفعالهم الأكثر انحرافًا غير مكتشفة. بعد weighing كلا الحجتين، استقر رأينا في النهاية على الخيار الثاني وفرضنا:
H6 كلما زادت ميول الشخص الميكافيلية، زادت المواقف السلبية التي يحملها تجاه الذكاء الاصطناعي.
الشخص الذي يحصل على درجات عالية في السيكوباتية من المحتمل أن يكون ميالًا لسلوكيات البحث عن الإثارة وقد يشعر فقط بقليل من القلق. على عكس التلاعب المتعمد الذي يتميز به الأشخاص الميكافيلليين، فإن سمة الشخصية السيكوباتية تتضمن ميولًا معادية اجتماعيًا أكثر اندفاعًا. بناءً على ذلك، افترضنا وجود علاقة إيجابية بين السيكوباتية والمواقف تجاه الذكاء الاصطناعي – لا سيما بالنظر إلى أن متغيرنا التابع سيتضمن أيضًا ردود فعل الناس العاطفية تجاه الذكاء الاصطناعي، والتي ينبغي أن تكون أقل خوفًا بين أولئك الذين لديهم درجات عالية في السيكوباتية. افترضنا:
H7 كلما زادت درجة نفسية الشخص، زادت المواقف الإيجابية التي يحملها تجاه الذكاء الاصطناعي.
بشكل عام، يميل النرجسيون إلى إظهار سلوكيات أنانية وفخورة وغير متعاطفة، وغالبًا ما تكون مصحوبة بمشاعر العظمة والاستحقاق. بينما تشير بعض الأبحاث النوعية إلى أن الحواسيب الذكية قد توجه ضربة شديدة لـ”نرجسيتنا”. ، ص. 145)، بعض الإمكانيات التي تقدمها الذكاء الاصطناعي قد تبدو جذابة للغاية للأشخاص الذين يحصلون على درجات عالية في هذه السمة – مثل الفكرة المستقبلية لإدخال جوانب من الذات في خوارزمية حاسوبية، لتُحفظ للأبد. ومن المدهش أن استفسار المشاركين حول هذه الفكرة لم يكشف عن علاقة كبيرة بالنرجسية في دراسة حديثة. ثم مرة أخرى، يجب أن نلاحظ أن الأفراد الذين يعانون من نرجسية عالية يميلون أيضًا إلى إظهار سلوك مفتوح ومبادر. مما قد يشير إلى علاقة أكثر إيجابية مع المواقف المتعلقة بالذكاء الاصطناعي اعتمادًا على فرضياتنا الأخرى. باختصار، استقرينا على الافتراض التالي، الذي يحمل طابعًا إيجابيًا بحذر:
كلما زاد نرجسية الشخص، زادت المواقف الإيجابية التي يحملها تجاه الذكاء الاصطناعي.
أخيرًا، انتقلنا إلى استكشاف متغير سلوكي على نطاق أصغر بكثير من التجريد، والذي بدا واعدًا لنا من منظور معاصر: عقلية المؤامرة لدى الناس. ترتبط هذه السمة الشخصية ارتباطًا وثيقًا بعدم الثقة العامة في الناس والمؤسسات والأنظمة السياسية بأكملها، وتركز على القابلية لنظريات المؤامرة، أي التفسيرات حول أحداث مشهورة تتحدى المنطق السليم أو الحقائق المقدمة علنًا. أظهرت الأبحاث النفسية أن الأشخاص الذين يؤمنون بنظرية مؤامرة واحدة يكونون عادةً أكثر عرضة للإيمان بنظرية أخرى، مما يشير إلى ميل ثابت يشبه السمة للاشتراك في طريقة تفكير مفرطة الشك وعدم الثقة. علاوة على ذلك، أكدت الأبحاث الحديثة أن الميل لقبول المعلومات المشكوك فيها معرفيًا قد يشكل بُعدًا واحدًا يرتبط بمزيد من الصعوبات في التفكير التحليلي، وتقدير مبالغ فيه للمعرفة الذاتية، بالإضافة إلى الانفتاح على ‘هراء زائف عميق’. .
على الرغم من أنهم قد لا يكونون معروفين تمامًا مثل نظريات المؤامرة البارزة الأخرى (مثل، المتعلقة بـ الهجمات)، هناك في الواقع عدة معتقدات مؤامرة حول استخدام التكنولوجيا المتطورة والذكاء الاصطناعي. تتراوح النظريات المعنية من فكرة أن الذين في السلطة يريدون استبدال بعض الأفراد بالآلات الذكية. إلى الاعتقاد بأن لقاحات COVID-19 هي في الواقع حقن لرقائق نانوية متطورة لإنشاء نظام مراقبة عالمي بينما قد تبدو هذه المفاهيم غامضة بعض الشيء للكثيرين، فإن انتصار وسائل التواصل الاجتماعي كمصدر للأخبار قد منح مؤيدي نظريات المؤامرة منصة قوية لنشر أفكارهم. بالإضافة إلى ذلك، فإن الاستخدامات (غير) الموثقة جيدًا للبيانات الضخمة – على سبيل المثال، في فضيحة كامبريدج أناليتيكا – قد دفعت العديد من الناس على الأقل للتفكير في كيفية استخدام خوارزميات الكمبيوتر لأغراض شريرة. مع الأخذ في الاعتبار جميع هذه الملاحظات، توقعنا أن تتعلق عقلية المؤامرة بمزيد من المواقف السلبية تجاه الذكاء الاصطناعي:
H9 كلما زادت درجة عقلية المؤامرة لدى الشخص، زادت المواقف السلبية التي يحملها تجاه الذكاء الاصطناعي.
ختامًا للتحضير النظري لدراستنا، وضعنا فرضيات حول متغيرين ديموغرافيين أساسيين تم ربطهما سابقًا بالمواقف المتعلقة بالذكاء الاصطناعي في المنشورات العلمية. على وجه التحديد، وُجد أن النساء عمومًا يحملن مواقف أكثر سلبية تجاه الذكاء الاصطناعي مقارنة بالرجال، وهو ما يمكن تفسيره، من بين أسباب أخرى، بالحواجز الاجتماعية التي تحد من وصول النساء إلى (واهتمامهن بـ) المجالات التقنية. . وبالمثل، أشارت الدراسات إلى وجود قلق أكبر تجاه التكنولوجيا المدعومة بالذكاء الاصطناعي بين كبار السن، ربما بسبب نقص الفهم وسهولة الوصول. في التوقع لتكرار هذه النتائج السابقة، اقترحنا:
تمتلك النساء في H10 مواقف أكثر سلبية تجاه الذكاء الاصطناعي مقارنة بالرجال.
H11 الأفراد الأكبر سناً يحملون مواقف أكثر سلبية تجاه الذكاء الاصطناعي مقارنة بالأفراد الأصغر سناً.

الدراسة 1

الهدف الأساسي من الدراسة 1 كان تطوير مقياس نفسي موثوق يكون (أ) أحادي البعد، (ب) يتضمن عناصر تعكس الأسس الثلاثة أو الجوانب للمواقف (المعرفية، العاطفية، السلوكية) التي وجهت أبحاث المواقف في علم النفس على مدى العقود الماضية. و (ج) تتكون من عناصر مكتوبة بعبارات إيجابية وسلبية لضمان قياس المواقف على طيف كامل بين النفور والحماس (ولم تؤثر انحياز الاتفاق على النتائج بشكل منهجي). استنادًا إلى النظرية ومقاييس المواقف الموجودة في المجالات التطبيقية تم إنشاء عناصر أصلية. تم توجيه إنشاء العناصر من خلال الأهداف المتمثلة في تطوير عدد متساوٍ من العناصر ذات الصياغة السلبية والإيجابية وعدد متساوٍ من العناصر التي تمثل الجوانب الثلاثة للموقف (المعرفية، العاطفية، السلوكية). الالتزام بنظرية الذكاء الاصطناعي السابقة. وطرق القياس الحالية ، تم تقديم العناصر بعد تعليمات قدمت ظاهرة الذكاء الاصطناعي للمشاركين، من أجل تقليل الغموض الدلالي حول هدف الموقف. هذه التعليمات هي جزء لا يتجزأ من القياس.
في الخطوة الثانية، ناقش المؤلفون العناصر، واستبعدوا العناصر التي كانت غامضة في المحتوى، أو محددة جدًا من حيث هدف الموقف، أو تضمنت تداخلًا لغويًا كبيرًا مع عناصر أخرى. كانت المقياس المتبقي يتكون من اثني عشر عنصرًا، حيث تم تمثيل كل من الجوانب النفسية الثلاثة لمواقف البشر – المعرفية، والعاطفية، والسلوكية – بواسطة عنصرين إيجابيين وعنصرين سلبيين. ومع ذلك، كان من المتوقع أن تمثل جميع العناصر الاثني عشر عاملًا عامًا واحدًا: موقف الناس تجاه الذكاء الاصطناعي. يمكن العثور على المقياس الكامل (ATTARI-12) في الملحق.
للحصول على رؤى حول صلاحية مقياسنا الجديد، شملت الدراسة 1 أيضًا عناصر تتعلق بتطبيقات الذكاء الاصطناعي الأكثر تحديدًا. كنا نتوقع أن تكون المواقف العامة تجاه الذكاء الاصطناعي كما تم قياسها بواسطة ATTARI-12 مرتبطة بشكل إيجابي بالمواقف المحددة تجاه المساعدين الشخصيين الإلكترونيين (مثل أليكسا) والمواقف تجاه الروبوتات. علاوة على ذلك، قمنا بتقييم ميل المشاركين لتقديم إجابات مرغوبة اجتماعيًا لتحديد ما إذا كان أداتنا الجديدة ستتأثر بتحيز الرغبة الاجتماعية. استنادًا إلى صياغتنا الدقيقة أثناء بناء المقياس، توقعنا أن تكون مواقف الناس تجاه الذكاء الاصطناعي (وفقًا لـ ATTARI-12) غير مرتبطة بشكل كبير بالرغبة الاجتماعية. تم تسجيل الفرضيات وخطوات تحليل البيانات المخطط لها للدراسة 1 مسبقًا على https://aspredicted.org/8B5_GHZ (انظر أيضًا الملحق S1).

طريقة

بيان الأخلاقيات

في ألمانيا، لا يتطلب البحث النفسي موافقة أخلاقية مؤسسية طالما أنه لا يتضمن قضايا منظمة بموجب القانون. تم إجراء جميع الدراسات المبلغ عنها (الدراسة 1، الدراسة 2، والدراسة 3) وفقًا كاملًا لإعلان هلسنكي، بالإضافة إلى الإرشادات الأخلاقية المقدمة من الجمعية النفسية الألمانية (DGPs). بالطبع، كان ذلك يتضمن أيضًا الحصول على موافقة مستنيرة من جميع المشاركين قبل أن يتمكنوا من المشاركة في هذه الدراسة.

المشاركون والإجراء

تحليلات القوة باستخدام semPower (الإصدار 2.0.1) ) اقترح حجم عينة ضروري لا يقل عن 500 مستجيب للدراسة المخطط لها. لذلك، تم السعي لجذب ما لا يقل عن 600 مشارك من أجل الحصول على قاعدة بيانات مناسبة لتحليلاتنا – مع الاحتفاظ بهامش لاستبعاد المحتمل. وبالتالي، تم تجنيد ما مجموعه 601 مشارك من مجموعة المشاركين الأمريكية في MTurk. لضمان جودة بيانات مرضية، قمنا بتحديد طلبنا في MTurk ليكون على الأقل 500 مهمة مكتملة مسبقًا (المعروفة أيضًا باسم HITs)، بالإضافة إلى معدل موافقة HIT يزيد عن 98%. تم مكافأة كل مشارك لمشاركتهم، التي استمرت حوالي ثلاث إلى خمس دقائق.
وفقًا لمعايير الاستبعاد المسجلة مسبقًا (أي، وقت الإكمال، الفشل في اجتياز واحد على الأقل من اختبارين للانتباه)، تم استبعاد 111 مشاركًا من تحليلاتنا. بشكل أكثر تحديدًا، أكمل 31 مشاركًا الاستبيان في أقل من لم يصف المشاركون الدراسة كما هو مطلوب، وأبلغ أحد المشاركين عن سنة ميلاد تختلف عن العمر المبلغ عنه بأكثر من ثلاث سنوات. وبالتالي، كانت العينة النهائية تتكون من 490 مشاركًا (212 أنثى، 273 ذكر، 5 آخرون أو لم يتم الرد). كان عمر المشاركين بين 19 و72 عامًا. سنوات، للحصول على معلومات ديموغرافية إضافية، يرجى الرجوع إلى الملحق S2. بعد أن قدم المشاركون موافقتهم المستنيرة، أكملوا أول اختبار انتباه قبل الانتقال إلى استبيان ATTARI-12. بعد ذلك، قمنا بتقييم مواقفهم تجاه المساعدين الشخصيين الإلكترونيين والروبوتات، قبل تقديم مقياس للإجابة الاجتماعية المرغوبة. أخيرًا، كان على المشاركين تلخيص موضوع الدراسة كاختبار انتباه إضافي والرد على عدة أسئلة سوسيو ديموغرافية.

تدابير

المواقف تجاه الذكاء الاصطناعي. قمنا بتطبيق مقياس ATTARI-12 الذي تم إنشاؤه حديثًا، باستخدام تنسيق إجابة من خمسة نقاط لالتقاط ردود المشاركين. أعارض بشدة موافق بشدة). يتم الإبلاغ عن المقاييس الوصفية وموثوقية المقياس في قسم النتائج.
المواقف تجاه المساعدين الصوتيين الشخصيين (PVA). أشار المشاركون إلى مواقفهم تجاه المساعدين الصوتيين الشخصيين من خلال ثلاثة عناصر تمييز دلالي (مع خمس نقاط تدرج) تم إنشاؤها لغرض هذه الدراسة (على سبيل المثال، “أكرهه – أحبه”؛ كرونباخ’s ; انظر الملحق S3).
المواقف تجاه الروبوتات. تم قياس التقييم العام للناس تجاه الروبوتات باستخدام ثلاثة عناصر تم استخدامها سابقًا في أبحاث قبول الروبوتات. . تم الإجابة على العناصر بمقياس من أربع نقاط (كرونباخ ، على سبيل المثال، “الروبوتات شيء جيد للمجتمعات لأنها تساعد الناس”، مع أختلف تمامًا مع أوافق تمامًا، انظر الملحق S3).
الرغبة الاجتماعية. تم قياس ميل المشاركين لتقديم إجابات مرغوبة اجتماعيًا باستخدام مقياس الرغبة الاجتماعية. . في هذا المقياس، يتعين على المشاركين الإجابة عما إذا كانت 16 من الأفعال الاجتماعية المرغوبة أو غير المرغوبة تتطابق مع سلوكهم الخاص على مقياس ثنائي (“صحيح” أو “خطأ”). يتم جمع عدد الاستجابات الاجتماعية المرغوبة لكل مشارك، مما ينتج عنه نطاق يتراوح بين 0 و 16 (كرونباخ ، ).

النتائج

تم توجيه تحليلنا لصلاحية العوامل من خلال الافتراض بأن العناصر تمثل بناءً واحدًا: موقف الناس تجاه الذكاء الاصطناعي. ولهذا الغرض، قمنا بمقارنة نماذج مختلفة، تتسم بتقييد أقل بشكل متزايد، في تحليل العوامل التأكيدي: النموذج الأول (أ) حدد عاملاً واحدًا، وبالتالي افترض أن الفروق الفردية في استجابات العناصر يمكن تفسيرها من خلال بناء موقف عام واحد. ومع ذلك، فإن هذا الافتراض غالبًا ما يكون قويًا جدًا في الممارسة العملية لأن جوانب المحتوى المحددة أو صياغة العناصر قد تؤدي إلى تعددية أبعاد طفيفة. لذلك، قام النموذج (ب) بتقدير هيكل ثنائي العوامل S-1 الذي، بالإضافة إلى العامل العام، شمل عاملين محددين متعامدين للعناصر المعرفية أو العاطفية. هذه العوامل المحددة التقطت التباين الفريد الناتج عن مجالي المحتوى الاثنين اللذين لم يتم حسابهما بواسطة عامل الموقف العام. وفقًا لما ذكره عيد وزملاؤه. لم يتم تحديد عامل محدد للعناصر السلوكية التي عملت، بالتالي، كمرجع. نظرًا لأن ATTARI-12 شمل عناصر مكتوبة بصياغة إيجابية وسلبية، استكشفت النموذج (ج) التأثيرات المحتملة للصياغة من خلال تقييم نموذج S-1 ثنائي العامل الذي شمل عاملًا محددًا متعامدًا للعناصر المكتوبة بصياغة سلبية. أخيرًا، قمنا بدمج النموذجين (ب) و(ج) لدراسة التأثيرات المشتركة لمحتوى وطرق القياس.
تم تلخيص جودة ملاءمة هذه النماذج في الجدول 2. باستخدام التوصيات المعتمدة لتفسير هذه المؤشرات. أظهر نموذج العامل الواحد ملاءمة أقل للبيانات. كما أن الاعتراف بجوانب المحتوى المختلفة (المعرفية، العاطفية، السلوكية) لم يحسن من ملاءمة النموذج. ومع ذلك، أظهر نموذج ATTARI-12 تأثيرات ملحوظة في الصياغة كما يتضح من النموذج (ج). أظهر هذا النموذج ملاءمة جيدة للبيانات مع مؤشر ملاءمة مقارن (CFI) قدره 0.98، وخطأ الجذر التربيعي المتوسط (RMSEA) قدره 0.03، وجذر متوسط البقايا المعيارية (SRMR) قدره 0.03. على الرغم من تعددية الأبعاد الملحوظة، أظهرت جميع العناصر ارتباطات جيدة إلى ممتازة مع العامل العام كما يتضح من تحميلات العوامل العالية التي تتراوح بين 0.48 و 0.88 (انظر الجدول 3). لفحص عواقب هيكل القياس متعدد الأبعاد بمزيد من التفصيل، قمنا بفحص مقدار التباين المنهجي في ATTARI-12 الذي تم شرحه بواسطة العامل العام.
CFI RMSEA SRMR إيه آي سي بيك تعويض
(أ) نموذج العامل الواحد 434.77 ٥٤ 0.85 0.15 0.08 ١٤,٥٦٤ ١٤٬٦٦٤
نماذج Bifactor S-1 مع عامل عالمي واحد وعوامل محددة متعامدة لـ …
(ب) جوانب المحتوى 327.39 ٤٦ 0.89 0.14 0.07 ١٤,٤١٢ 1456 (أ) ١٠٢.٩٥ ٨ <0.001
(ج) نص العنصر ١٠٩.٦٤ ٤٨ 0.98 0.06 0.03 ١٤٠٨٦ ١٤٢١٢ (أ) ٢٦٨.٣٥ ٦ <0.001
(د) جوانب المحتوى وصياغة العناصر 93.91 40 0.98 0.06 0.03 ١٤٠٧٦ ١٤٢٣٦ (ب) 191.45 ٦ <0.001
(ج) ١٦.٠٧ ٨ 0.041
الجدول 2. جودة الملاءمة لنماذج العوامل التأكيدية المتنافسة لـ ATTARI-12 (الدراسة 1؛ المشاركون في لوحة MTurk الأمريكية). إحصائية اختبار قوية ، درجات الحرية، مؤشر الملاءمة المقارن (CFI)، جذر متوسط مربع خطأ التقريب (RMSEA)، الجذر القياسي لمتوسط البقايا (SRMR)، معيار معلومات أكايك (AIC)، معيار معلومات بايزي (BIC)، نموذج المقارنة (Comp.) إحصائية اختبار فرق مربع كاي المقاسة .
عنصر تحميل على العامل العام تحميل على عامل محدد التباين المتبقي
1 0.79 (0.04)/0.84 0.26
2 0.66 (0.05) /0.59 0.61 (0.06) /0.54 0.46
٣ 0.81 (0.04) /0.77 0.45
٤ 0.66 (0.05) /0.57 0.54 (0.06) /0.48 0.58
٥ 0.91 (0.04) /0.87 0.27
٦ 0.64 (0.05) /0.65 0.55
٧ 0.77 (0.05) /0.64 0.56 (0.06) /0.46 0.56
٨ 0.55 (0.05) /0.48 0.52 (0.06) /0.45 0.75
9 0.71 (0.05) / 0.65 0.69
10 0.67 (0.05) / 0.59 0.70 (0.06) / 0.61 0.37
11 0.95 (0.04) / 0.88 0.27
12 0.84 (0.05) / 0.69 0.59 (0.06) / 0.48 0.43
الجدول 3. نمط تحميل العوامل لـ ATTARI-12 (الدراسة 1؛ المشاركون في لوحة MTurk الأمريكية). تم تقديم تحميلات العوامل غير المعيارية (مع الأخطاء المعيارية بين قوسين) / تحميلات العوامل المعيارية والتباينات المتبقية. تم تحديد العوامل الكامنة عن طريق تقييد تباينات العوامل الكامنة إلى 1.
عامل أو عامل الصياغة المحددة (انظر ). بينما أوضح عامل الصياغة عن من التباين، معظم التباين المشترك ( ) تم عكسه بواسطة العامل العام. ونتيجة لذلك، كانت موثوقية أوميغا الهرمية للعامل العام لكن فقط للعامل المحدد. تؤكد هذه النتائج أن جميع العناصر الاثني عشر تشكل مقياسًا أحادي البعد بشكل أساسي يقوم بتشغيل عامل عام قوي يعكس الموقف الشامل تجاه الذكاء الاصطناعي. استنادًا إلى التحليلات، لم يكن هناك حاجة لاستبعاد أي عناصر من المقياس بسبب خصائص نفسية غير مرضية.
تظهر النتائج الوصفية والارتباطية على المقياس المركب لـ ATTARI-12 في الجدول 4. كانت موثوقية المقياس من حيث الاتساق الداخلي ممتازة، حيث كانت قيمة كرونباخ “. كانت التوزيعة تقارب التوزيع الطبيعي، الانحراف ( ) و kurtosis ( ) كانت ضمن الحدود المتوقعة للمقاييس التي تم اختبارها نفسيًا (انظر أيضًا الشكل 1). ومع ذلك، نعترف بأن التوزيع كان مائلًا قليلاً إلى اليسار. كما هو متوقع، كانت هناك علاقة إيجابية بين المقياس والمواقف تجاه المساعدين الشخصيين الإلكترونيين ( ) ومع المواقف تجاه الروبوتات ( لم تكن درجات ATTARI-12 مرتبطة بشكل كبير بتحيز الرغبة الاجتماعية. ).
باختصار، قاس مقياس ATTARI-12 أحادية البعد المواقف تجاه الذكاء الاصطناعي بطريقة موثوقة، وتؤكد مؤشرات الصلاحية البنائية المتقاربة (المواقف تجاه تطبيقات الذكاء الاصطناعي المحددة) والمتباينة (الرغبة الاجتماعية) صلاحية المقياس.
1 ٢ ٣ ٤ ٥
كرونباخ M انحراف كورت
1 أتاري-12 0.93 3.66 0.83 -0.63 0.22
٢ الرغبة الاجتماعية 0.85 8.74 ٤.٢٤ -0.25 -0.87 0.037
٣ الموقف تجاه PVAs 0.94 3.72 1.09 -0.81 0.02 0.599*** 0.125**
٤ الموقف تجاه الروبوتات 0.77 3.24 0.57 -0.75 0.81 0.676*** -0.021 0.435***
٥ عمر ٣٩.٧٨ 11.06 0.81 -0.08 – 0.039 0.090* -0.023 -0.036
٦ جنس -0.087 0.042 0.099* -0.058
الجدول 4. الإحصائيات الوصفية والارتباطات من الدرجة صفر (الدراسة 1؛ المشاركون في لوحة MTurk الأمريكية). لتحليلات تتعلق بالجنس، لجميع التحليلات الأخرى. تم ترميز الجنس بشكل وهمي ذكر؛ الإناث). PVAs = المساعدات الصوتية الشخصية.
الشكل 1. توزيعات نتائج ATTARI-12 في الدراسة 1 والدراسة 2 (في الوقت 1 والوقت 2).

الدراسة 2

كان هدف الدراسة الثانية هو تطوير نسخة باللغة الألمانية من المقياس والحصول على مزيد من الفهم حول موثوقية المقياس وصحته. على وجه الخصوص، قمنا بفحص موثوقية إعادة الاختبار للمقياس، من خلال تقديم العناصر في نقطتين زمنيتين. كما قمنا بتقييم مدى رغبة المشاركين (طلاب الجامعات) في العمل مع (أو بدون) الذكاء الاصطناعي في مسيراتهم المهنية المستقبلية. كنا نتوقع وجود ارتباط إيجابي بين المتغير الأخير والمواقف تجاه الذكاء الاصطناعي.

طريقة

تكونت هذه الدراسة من جزئين، تم إجراؤهما عبر الإنترنت بمتوسط تأخير قدره 31.40 يومًا. ; المدى 26-36 يومًا). شارك طلاب تخصص العلوم الاجتماعية في جامعة وورزبورغ، ألمانيا، للحصول على رصيد دراسي. وفقًا للحسابات مع برنامج الطاقة يعتمد على ارتباط متوقع صغير إلى معتدل القوة، و كان مطلوبًا حد أدنى من حجم العينة يبلغ 123 مشاركًا. ومع ذلك، نظرًا لتوقعنا انسحاب بعض المشاركين من الوقت 1 إلى الوقت 2، تم استهداف عينة ابتدائية من 180 مشاركًا. في النهاية، أكمل 166 مشاركًا الاستبيان في الوقت 1، وقدم إجمالي 150 طالبًا بيانات في كلا النقطتين الزمنيتين. كانت هذه العينة النهائية المكونة من 150 مشاركًا مع بيانات كاملة بمتوسط عمر يبلغ 21.21 عامًا. ; النطاق: من 18 إلى 41 عامًا) وتكونت بشكل أساسي من النساء (113 أنثى، 36 ذكر، 1 آخر). كان تقريبًا جميع المشاركين يتحدثون الألمانية كلغة أم (انظر الملحق S4 للحصول على بيانات وصفية مفصلة والملحق S8 للإصدار باللغة الألمانية من المقياس). من المهم أنه نظرًا لأن المشاركين في الدراسة 2 تم تجنيدهم من بين الطلاب الألمان (على عكس استخدام أعضاء لوحة الاستطلاع الناطقة باللغة الإنجليزية في الدراسة 1)، فنحن واثقون من أن دراستينا للتحقق من الصحة كانت مبنية على عينات مختلفة تمامًا.
في الوقت 1، أجاب المشاركون على اختبار ATTARI-12، بالإضافة إلى أربعة عناصر قيست اهتمامهم بمهنة تتعلق بتكنولوجيا الذكاء الاصطناعي (كان هناك عنصران مشفران عكسيًا، مثل: “أفضل وظيفة لا يلعب فيها الذكاء الاصطناعي أي دور”). كانت العناصر تتضمن مقياسًا من 5 نقاط. أعارض بشدة أوافق بشدة)، كرونباخ (انظر الملحق S5). في الفهرس الناتج، كانت الدرجات الأعلى تشير إلى طموحات أعلى في الوظائف المتعلقة بالذكاء الاصطناعي. في وقت القياس الثاني للدراسة، كان يجب الإجابة على ATTARI-12 مرة أخرى. علاوة على ذلك، تم تصميم الدراسة في البداية لتشمل أيضًا مقياسًا للرغبة الاجتماعية في هذه المرحلة ( ، في ترجمته إلى الألمانية)؛ ومع ذلك، لم تحقق التقييمات المعنية موثوقية مرضية (كرونباخ لذلك، تم استبعاده من تحليلنا.

النتائج

تظهر النتائج الرئيسية في الجدول 5. كانت موثوقية النسخة الألمانية المستخدمة في هذه الدراسة من حيث الاتساق الداخلي جيدة جداً، معامل كرونباخ T1 ; معامل كرونباخ T2 . كانت التوزيعة تقارب التوزيع الطبيعي، الانحراف (الانحراف انحراف ) و kurtosis ( kurtosis التفرطح كانت ضمن الحدود المتوقعة للمقاييس النفسية السليمة (الشكل 1). من المهم أن كانت العلاقة بين الاختبار وإعادة الاختبار كبيرة، ، مما يدعم موثوقية المقياس. كانت الارتباطات مع مقياس الذكاء الاصطناعي-المهنة ذات دلالة وبلغت حوالي باختصار، أظهر مقياس ATTARI-12 أحادي البعد (الإصدار الألماني) خصائص نفسية جيدة.

الدراسة 3

بعد أن وجدنا دعمًا تجريبيًا لصحة مقياسنا الجديد، تابعنا هدفنا البحثي الرئيسي الثاني: استكشاف العلاقات المحتملة بين سمات الشخصية والمواقف تجاه الذكاء الاصطناعي. لضمان الشفافية، قررنا تسجيل الدراسة 3 مسبقًا قبل جمع البيانات، بما في ذلك جميع الفرضيات والتحليلات المخطط لها.https://aspredicted.org/VRU_BJI).

طريقة

المشاركون

حساب مسبق لحجم العينة الأدنى بواسطة برنامج الطاقة (بافتراض تأثير صغير إلى معتدل من قوة ، ونتيجةً لذلك، كان هناك حد أدنى قدره 221 مشاركًا في تحليل الانحدار الخطي الهرمي مع أحد عشر متنبئًا. نظرًا لأننا كنا نعتزم فحص عيّنتنا لعدة مؤشرات جودة البيانات، وبالتالي، أردنا بعض المرونة لاستبعاد محتمل، قمنا بتجنيد 353 مشاركًا أمريكيًا عبر مجموعة المشاركين في أمازون MTurk (العمر: سنوات، ; 112 أنثى، 239 ذكر، 2 آخر؛ الدفع ). على وجه التحديد، استخدمنا المعايير التالية من MTurk لضمان جودة البيانات العالية : (أ) على الأقل 100 HITs معتمدة؛ (ب) معدل الموافقة على HIT . علاوة على ذلك، فإن خدمة الويب “آي ف hub.info” تم استخدامه للتحقق مما إذا كانت عناوين IP الفردية للمستخدمين تشير بشكل صحيح إلى الولايات المتحدة كموقع حالي.
أتاري-12 T1 أتاري-12 T2
كرونباخ M SD انحراف كورت
أتاري-12 تي 1 0.90 3.51 0.64 – 0.58 0.07
أATTARI-12 T2 0.89 3.49 0.63 -0.48 -0.24 0.804
المهنة-الذكاء الاصطناعي T1 0.85 3.11 0.92 -0.08 -0.78 0.635 0.563
الجدول 5. الوصفيات والارتباطات من الدرجة صفر (الدراسة 2؛ طلاب الجامعات الألمان). لكلا المتغيرين المقيمين في الوقت 1 (T1). لـ ATTARI-12 تم تقييمه في الوقت 2 (T2). للعلاقات بين متغيرات T1 و T2. جميع العلاقات ذات دلالة إحصائية عند (ذو طرفين). القيمة المطبوعة بخط عريض تمثل موثوقية إعادة الاختبار المحورية.
عند فحص البيانات المجمعة، تم اتخاذ عدة تدابير لاستبعاد عمال MTurk الذين تشير إجاباتهم إلى الاستجابة غير المبالية وغير المنتبهة. أولاً، طلبنا من جميع المشاركين تحديد أعمارهم وسنة ميلادهم في صفحتين منفصلتين من الاستبيان؛ إذا انحرفت هذه الإجابات بأكثر من عامين عن بعضها البعض، تم استبعاد المشارك. ثانيًا، كتحقق محدد ضد العمال الآليين، طُلب من المشاركين تسمية نوع من الخضار (“الباذنجان”) الذي تم تصويره في صورة كبيرة الحجم. باستثناء الأخطاء المطبعية الطفيفة، أدت جميع الإجابات التي لم تشبه الإجابة الصحيحة باللغة الإنجليزية إلى حذف البيانات. . ثالثًا، طُلب من المشاركين اختيار الموضوع الدراسي الصحيح (“الذكاء الاصطناعي”) في سؤال اختيار من متعدد، وهو مهمة لم يتم تنفيذها بشكل صحيح من قبل آخرين. المشاركين.
بالإضافة إلى ذلك، اعتبرنا في البداية استبعاد جميع المشاركين الذين ملأوا الاستبيان في أقل من أربع دقائق، وهو الحد الأدنى الذي قمنا بقياسه كوقت أدنى للاستجابة بانتباه. عند النظر إلى البيانات التي حصلنا عليها، كان من الممكن أن يؤدي ذلك إلى استبعاد 81 مشاركًا إضافيًا. ومع ذلك، حيث لاحظنا أن العديد من عمال MTurk قد أنهوا الاستبيان في أقل من أربع دقائق بقليل، قررنا تخفيف معيار الاستبعاد الأولي لدينا إلى مدة أدنى قدرها ثلاث دقائق، مما أدى إلى استبعاد المشاركين. للتأكد من أن هذا الانحراف عن خطة التحليل المسجلة مسبقًا لن يغير نتائجنا بشكل كبير، قمنا بإعادة جميع التحليلات المخطط لها باستخدام معيار الأربع دقائق الأولي. من خلال القيام بذلك، لم نجد أي اختلافات إحصائية ملحوظة (نتائج هذا التحليل الإضافي مقدمة في الملحق S6).
لذا، في الملخص، كانت عيّنتنا النهائية تتكون من 298 مشاركًا بمتوسط عمر 39.29 سنة. ; تتراوح أعمارهم بين 22 و 73 عامًا). كان التوازن بين الجنسين مائلًا قليلاً نحو الرجال (102 أنثى، 195 ذكر، 1 آخر). فيما يتعلق بالعرق، حدد معظم المشاركين أنفسهم كأشخاص بيض ( )، تليها الآسيوية ( )، أسود ( ) ، ولاتيني ( ). كان مستوى التعليم مرتفعًا نسبيًا، حيث حصل معظم المشاركين على درجة البكالوريوس ( ماجستير ) أو دكتوراه ( وأخيرًا، نلاحظ أن عيّنتنا كانت متوازنة إلى حد كبير من حيث التوجه السياسي. للحصول على نظرة شاملة على البيانات الوصفية التي تم الحصول عليها، يرجى الرجوع إلى الملحق S7.

إجراءات

تم تقديم جميع عناصر التدابير التالية على مقياس من 5 نقاط ( أعارض بشدة؛ أوافق بشدة).
المواقف تجاه الذكاء الاصطناعي. لقياس المتغير الناتج الرئيسي، قمنا بتطبيق مقياس ATTARI-12 بنسخته الإنجليزية. أظهرت تحليل بياناتنا أن المقياس مرة أخرى حقق اتساقًا داخليًا ممتازًا في هذه المرة (ألفا كرونباخ ).
الخمس الكبار. تم تقييم الخمس الكبار باستخدام استبيان الخمس الكبار. ، والتي تتكون من 44 عنصرًا (الانفتاح: عشرة عناصر، مثل “أنا شخص فضولي بشأن العديد من الأشياء المختلفة”; الضمير الحي: تسعة عناصر، مثل “… يقوم بعمل شامل”; الانبساط: 8 عناصر، مثل “… يتحدث كثيرًا”; التوافق: تسعة عناصر، مثل “… يعتبر لطيفًا ومهتمًا تقريبًا بالجميع”; العصابية: 8 عناصر، مثل “… يمكن أن يكون متوترًا”). لقد لاحظنا اتساقًا داخليًا جيدًا جدًا إلى ممتاز لجميع الأبعاد الخمسة المقاسة (الانفتاح: ألفا كرونباخ ; الضمير: كرونباخ ; الانفتاح: ألفا كرونباخ ; القابلية للتوافق: ألفا كرونباخ ; العصابية: ألفا كرونباخ ).

الثالوث المظلم

قمنا بتقييم سمات الشخصية الثلاثية المظلمة لدى المشاركين باستخدام مقياس الثلاثية المظلمة القصير (SD3). ). تتضمن هذه الأداة تسعة عناصر عن الميكافيلية (مثل: “ليس من الحكمة أن تخبر أسرارك”)، وستة عناصر عن السيكوباتية (مثل: “غالبًا ما يقول الناس إنني خارج عن السيطرة.”)، وتسعة عناصر عن النرجسية (مثل: “تميل العديد من الأنشطة الجماعية إلى أن تكون مملة من دوني.”). أشارت تحليلات الموثوقية إلى وجود اتساق داخلي جيد إلى جيد جدًا لجميع المقاييس الثلاثة (الميكافيلية: ألفا كرونباخ ; السيكوباتية: كرونباخ ; النرجسية: كرونباخ ).

عقلية المؤامرة

استبيان عقلية المؤامرة ) يقدم مقياسًا لمدى تعرض الناس العام لنظريات المؤامرة والمعتقدات. يتكون من خمسة عناصر تلخص شكوكًا متأصلة حول كيفية عمل المجتمع والحكومات والمنظمات السرية (على سبيل المثال، “أعتقد أن العديد من الأمور المهمة تحدث في العالم، والتي لا يتم إبلاغ الجمهور عنها أبدًا.”؛ “أعتقد أن الأحداث التي تبدو ظاهريًا غير مرتبطة غالبًا ما تكون نتيجة أنشطة سرية.”). مع بياناتنا، لاحظنا قيمة جيدة جدًا لمقياس كرونباخ. 0.86 لـ CMQ.

النتائج

الجدول 6 يلخص المعلومات الوصفية والارتباطات من الدرجة صفر بين جميع متغيرات الدراسة. في المتوسط، كانت مواقف المشاركين المتعلقة بالذكاء الاصطناعي إيجابية قليلاً، مع الأخذ في الاعتبار أن قيمة 3 تشير إلى النقطة المحايدة في مقياسنا المكون من 5 نقاط، والذي تم تجميعه من أعداد متساوية من العناصر السلبية (المعكوسة) والإيجابية.
مع تحديد تحليل البيانات الرئيسي لدينا ليشمل الانحدار المتعدد، تأكدنا أولاً من تلبية جميع الافتراضات اللازمة. كانت المتبقيات مستقلة وموزعة بشكل طبيعي، ولم يتم العثور على أي مشاكل تتعلق بالتعدد الخطي أو التباين غير المتجانس. علاوة على ذلك، أظهرت قيم مسافة كوك أن لا حالات مؤثرة كانت تؤثر على نموذجنا. وبالتالي، تابعنا مع الانحدار الخطي الهرمي كالإجراء الأساسي لتحليل بياناتنا. باستخدام درجة ATTARI للمشاركين كمعيار، قمنا أولاً بإدخال أعمارهم وجنسهم كمتنبئين (الخطوة 1)، قبل إضافة الخمسة الكبار (الخطوة 2)، والثالوث المظلم (الخطوة 3)، وعقلية المؤامرة (الخطوة 4) إلى نموذج انحدار موسع. تقدم الجدول 7 الحسابات الرئيسية والمعاملات لهذه التحليل الانحداري الهرمي. كما يتضح هنا، أسفرت الخطوة الأولى من الانحدار عن معادلة غير دالة. 294) ، مع 0.02. بالمقابل، الخطوة الثانية من الإجراء (بما في ذلك الـ Big
متغير M SD 1 ٢ ٣ ٤ ٥ ٦ ٧ ٨ 9 10 11 12
1 عمر ٣٩.٢٩ 11.08
2 جنس -0.23***
٣ أتاري-12 3.60 0.81 -0.12* 0.04
٤ الانفتاح على التجربة ٣.٥٦ 0.74 0.02 0.01 0.14*
٥ الضمير الحي 3.97 0.72 0.25*** – 0.12* 0.08
٦ الانفتاح 2.77 1.00 0.04 0.07 0.04 0.27*** 0.25***
٧ الود 3.77 0.73 -0.15** 0.22*** 0.16**
٨ العصابية ٢.٦٤ 1.01 -0.13* -0.10 -0.07 -0.14* -0.60*** -0.44 -0.53***
9 المكيافيلية 2.88 0.81 -0.19** 0.17** -0.11 -0.05 -0.25*** 0.07 -0.50*** 0.25***
10 علم النفس المرضي ٢.١٩ 0.77 -0.25*** -0.12* -0.02 -0.36*** -0.50*** 0.14*
11 النرجسية ٢.٥٠ 0.74 -0.15** -0.06 0.23** -0.03 -0.18** -0.16** 0.62***
12 عقلية المؤامرة 3.36 0.94 -0.08 -0.06 0.06 -0.07 0.07 0.15*
الجدول 6. الإحصائيات الوصفية والارتباطات لمتغيرات الدراسة (الدراسة 3؛ المشاركون في لوحة MTurk الأمريكية).
ملاحظة. مُرمز بالجنس بـ ” 0 “ أنثى، ” 1 “ ذكر.
الخطوة 1 الخطوة 2 الخطوة 3 الخطوة 4
عمر – 0.12 -1.96 – 0.15* -2.52 -0.16** – 2.68 – 0.17** -2.86
جنس 0.01 0.24 0.06 0.95 0.07 1.15 0.04 0.68
الانفتاح على التجربة 0.09 1.53 0.10 1.67 0.11 1.85
الضمير الحي – 0.01 0.09 -0.03 -0.35 -0.01 -0.15
الانفتاح 0.01 0.08 0.08 0.93 0.07 0.85
الود 3.95 0.25** 3.12 3.34
العصابية 0.08 1.01 0.07 0.81 0.10 1.24
المكيافيلية 0.04 0.48 0.09 1.08
علم النفس المرضي -0.06 -0.64 -0.05 -0.49
النرجسية -0.09 – 1.02 -0.05 -0.60
عقلية المؤامرة -0.22*** -3.68
0.02 (0.08***) (0.01) (0.04***)
الجدول 7. الانحدار الهرمي الذي يتنبأ بمواقف المشاركين تجاه الذكاء الاصطناعي (الدراسة 3). ، . مُرمز بالجنس مع ” 0 “ أنثى، ” 1 “ ذكر.
خمسة) أسفر عن نتيجة مهمة، عند المتابعة مع الخطوة 3، وجدنا أن إضافة سمات الثلاثي المظلم لم تؤدِ إلى زيادة ملحوظة في ومع ذلك، أدى إدخال عقلية المؤامرة لدى المشاركين كمتنبئ نهائي في الخطوة 4 إلى حل يتمتع بتباين مفسر أعلى بشكل ملحوظ، . في ما يلي، سنلقي نظرة فاحصة على القيمة التنبؤية لكل متغير تم إدخاله.

القيمة التنبؤية للخمس الكبار

استجابةً للفرضيات H1 إلى H5، وجهنا انتباهنا أولاً إلى أبعاد الشخصية الخمسة الكبرى. استنادًا إلى الخطوة الثانية من الانحدار، التي كانت مصممة لتضمين هذه المتغيرات، وُجد أن التوافق فقط هو الذي تنبأ بشكل كبير بمواقف المشاركين تجاه الذكاء الاصطناعي. )، مع معامل بيتا إيجابي قدره 0.29. وقد استمرت هذه العلاقة الإيجابية أيضًا بعد إدخال سمات شخصية أخرى في الخطوتين 3 و 4 من الانحدار. ومن ثم، نستنتج أن ارتفاع مستوى التوافق كان مرتبطًا بوجهات نظر أكثر إيجابية حول تكنولوجيا الذكاء الاصطناعي في عيّنتنا – مما يوفر دعمًا تجريبيًا للفرضية 4. من بين الأبعاد الأخرى في الخمسة الكبار، كانت الانفتاح على التجربة قد فاتتها قليلاً عتبة الدلالة الإحصائية في الخطوة النهائية من الانحدار، لذا قد يكون من الضروري إجراء مزيد من الاستكشاف للاختبار المعني قبل رفضه تمامًا. ومع ذلك، ظلت السمات الخمس الكبرى المتبقية، وهي الضمير الحي (H2) والانفتاح (H3) والعصابية (H5)، غير ذات دلالة واضحة كمتنبئات.

القيمة التنبؤية للثالوث المظلم

بالانتقال إلى السمات الشخصية المظلمة مثل الميكافيلية، والاعتلال النفسي، والنرجسية (كما تم إدخاله خلال الخطوة 3)، نلاحظ أن أيًا من المتنبئين الثلاثة لم يقترب من العتبة التقليدية للدلالة الإحصائية. وبالتالي، لا يمكننا قبول الفرضيات H6 إلى H8؛ وفقًا لبياناتنا، لم تكن السمات الشخصية الأكثر شرًا تنبؤية بمواقف الناس العامة تجاه الذكاء الاصطناعي.

القيمة التنبؤية لعقلية المؤامرة

في ختام تحقيقنا في التأثيرات الشخصية، ركزنا على عقلية المؤامرة لدى المشاركين كما أضيفت خلال الخطوة النهائية من الانحدار. وجدنا أن هذه المتغيرات ظهرت كمتنبئ ذو دلالة أخرى، . دعمًا للفرضية 9، نبلغ عن أن عقلية المؤامرة الأقوى كانت مرتبطة بوجهات نظر أكثر سلبية تجاه التكنولوجيا الذكية الاصطناعية.

القيمة التنبؤية للعمر والجنس

أخيرًا، تم استكشاف تأثير عمر المشاركين وجنسهم على مواقفهم تجاه الذكاء الاصطناعي. لهذا الغرض، ركزنا على نموذج الانحدار النهائي لدينا مع جميع المتنبئين المدخلين في المعادلة. بينما ظل تأثير الجنس غير ذي دلالة، تم فحص أن العمر الأعلى كان مرتبطًا بشكل كبير بتدني الدرجات في مقياس ATTARI-12، أي، بمزيد من الإدراكات، والمشاعر، والنوايا السلوكية السلبية تجاه الذكاء الاصطناعي ( ). في ضوء ذلك، نجحت بياناتنا في تكرار النتائج السابقة المتعلقة بهذا المتغير الاجتماعي الديموغرافي، مما أدى إلى إجابة إيجابية للفرضية 11. في الوقت نفسه، نرفض الفرضية 10 بشأن تأثير الجنس.

نقاش عام

في عالم تشكله التقنيات المستقلة التي من المفترض أن تجعل الحياة البشرية أكثر أمانًا وصحة وراحة، من المهم فهم كيف يقيم الناس مفهوم التكنولوجيا الذكية الاصطناعية – وتحديد العوامل التي تفسر التباين الملحوظ بين الأفراد في هذا الصدد. وبالتالي، بدأ المشروع الحالي لاستكشاف دور السمات الشخصية الأساسية كمتنبئين بمواقف الناس المتعلقة بالذكاء الاصطناعي. من أجل بناء بحثنا على قياس صالح وموثوق للنتيجة المعنية، قمنا أولاً بتطوير استبيان جديد – ATTARI-12 – وأكدنا جودته النفسية عبر دراستين. مميزين أداتنا عن البدائل الموجودة، نلاحظ أن مقياسنا أحادي البعد يقيم المواقف تجاه الذكاء الاصطناعي على طيف كامل بين النفور والحماس. علاوة على ذلك، تتضمن عناصر ATTARI-12 الثالوث الكلاسيكي للموقف البشري (الإدراك، العاطفة، السلوك)، وبالتالي تظهر كطريقة مفهومية سليمة لقياس تقييم الناس للذكاء الاصطناعي. أخيرًا، نظرًا لأن تعليمات ATTARI-12 لا تركز على حالات استخدام محددة بل على فهم أوسع للذكاء الاصطناعي كمجموعة من القدرات التكنولوجية، نعتقد أنه قد يكون مناسبًا للاستخدام عبر العديد من التخصصات والإعدادات البحثية المختلفة.
باستخدام قياسنا الجديد المطور، انتقلنا إلى هدف بحثنا الثاني: التحقيق في الروابط المحتملة بين مواقف الأفراد المتعلقة بالذكاء الاصطناعي وسماتهم الشخصية. بشكل محدد، ركزنا على تصنيفين مركزيين من مجال علم نفس الشخصية (الخمسة الكبار، الثلاثي المظلم)، بالإضافة إلى عقلية المؤامرة كصفة ذات صلة معاصرة عالية. من خلال هذه الوسائل، وجدنا تأثيرات كبيرة لاثنين من المتنبئين الشخصيين المستكشفين: كانت الصداقة (واحدة من سمات الخمسة الكبار) مرتبطة بشكل كبير بمواقف أكثر إيجابية تجاه الذكاء الاصطناعي، بينما تنبأت عقلية المؤامرة الأقوى بالعكس. نعتقد أن كلا من هذين الاكتشافين قد يقدمان دلالات مثيرة للباحثين والمطورين ومستخدمي التكنولوجيا المستقلة.
بعد شخصية عادة ما تترافق مع نظرة متفائلة للحياة ، تم العثور على أن الصداقة تتنبأ بقبول الناس للابتكارات في العديد من المجالات المختلفة . من الناحية النفسية، هذا منطقي تمامًا: وفقًا لطبيعتهم الخيرية، يميل الأفراد الودودون إلى رؤية الآخرين بشكل أكثر إيجابية أيضًا – وقد يفكرون بالتالي أكثر في الفرص بدلاً من المخاطر لاختراع جديد عند تشكيل رأيهم. نظرًا لأن قياسنا المحدد لمواقف المشاركين طلب منهم النظر في الجوانب الإيجابية والسلبية للذكاء الاصطناعي، قد تكون هذه النزعة للتركيز أكثر على الجانب الإيجابي قد لعبت دورًا مهمًا في التأثير الملحوظ. علاوة على ذلك، نفترض أن الطبيعة الموثوقة التي غالبًا ما تميز الأفراد الودودين تقدم تفسيرًا معقولًا آخر لنتيجتنا. عندما يواجهون مفهومًا معقدًا مثل الذكاء الاصطناعي، الذي قد يبدو غامضًا أو غير مفهوم تمامًا للمستخدم العادي، يصبح من المهم أكثر للناس أن يفهموا نوايا صانعي القرار والمساهمين وراءه. بينما قد لا تلهم الشركات التكنولوجية التي تطور أنظمة الذكاء الاصطناعي دائمًا الثقة اللازمة من خلال أفعالها (مثل، جمع البيانات بشكل مفرط، سياسات الشركات المعقدة)، فإن وجود ميل أقوى للثقة بالآخرين من المحتمل أن يعوض عن ذلك – مما يقدم سببًا آخر لظهور الصداقة كمتنبئ مهم في الدراسة الحالية.
يمكن تقديم حجة مشابهة بشكل مدهش عند تناول العامل الشخصي الثاني الذي حقق دلالة ملحوظة في تحليلاتنا، أي، عقلية المؤامرة لدى الناس. حسب التعريف، هذه الصفة الشخصية مرتبطة أيضًا بإدراكات الثقة (أو بالأحرى، نقصها)، وهو مفهوم يتردد صداه في نتائجنا: كلما عبر المشاركون عن شكوكهم بشأن الحكومات، والمنظمات، والتغطية الإخبارية ذات الصلة (كما تشير إليه عناصر مثل “أعتقد أن العديد من الأمور المهمة جدًا تحدث في العالم، والتي لا يتم إبلاغ الجمهور عنها أبدًا”)، كلما كانت مواقفهم تجاه الذكاء الاصطناعي أكثر سلبية. في قراءتنا، يوضح هذا كيف تؤثر الشفافية والموثوقية المدركة بشكل حاسم على قبول الجمهور للذكاء الاصطناعي؛ تمامًا كما أن وجود شخصية ودودة وسهلة التصديق تنبأت بمواقف أكثر إيجابية، فإن الميل للاشتباه في قوى شريرة على الساحة العالمية مرتبط بوجهات نظر أكثر سلبية حول التكنولوجيا المستقلة.
على المستوى المجتمعي، نعتقد أن تحديدنا لمعتقدات المؤامرة كعقبة أمام مواقف الذكاء الاصطناعي هو نتيجة في الوقت المناسب بشكل خاص، حيث يرتبط بالنقاش المستمر حول الأخبار الزائفة في المجتمع الحديث. في السنوات القليلة الماضية، بدأ العديد من الناس في التوجه إلى وسائل التواصل الاجتماعي كمصدر للأخبار والتعليم، مما مهد الطريق لانتشار غير عادي للمعلومات المضللة . أصبحت المساحات الافتراضية، التي يؤكد فيها الأفراد ذوو التفكير المماثل مواقفهم وريبة حول قضايا مختلفة (ما يسمى بغرف الصدى؛ مثل، المنتديات النقاشية أو مجموعات الرسائل النصية) شائعة جدًا، ومن خلال هذه الوسائل، تحولت إلى مصدر للقلق بين علماء الإعلام. في الواقع، تشير الأبحاث إلى أن استخدام منصات مثل فيسبوك، تويتر، أو يوتيوب لا يوفر فقط وسيلة فعالة للغاية لنشر المعتقدات ما بعد الحقائق، بل قد يزيد حتى من قابلية الناس للإصابة بنظريات المؤامرة في المقام الأول . مع الأخذ في الاعتبار أن الموضوعات المعقدة والغامضة هي التي يسعى الناس للحصول على معلومات ما بعد الحقائق بشأنها ، قد تتزايد الآراء السلبية حول التكنولوجيا الذكية
قد تتزايد بسهولة في المجال الرقمي. من المؤكد أن الادعاء الملحوظ في وسائل التواصل الاجتماعي بأن لقاحات COVID-19 حقن سرًا تكنولوجيا النانو الذكية الاصطناعية في المرضى غير الراغبين يوضح هذا بشكل مثالي.
استنادًا إلى الحجج المقدمة، من المنطقي أن يعتمد النجاح في اعتماد التقنيات الذكية على نطاق واسع أيضًا على ما إذا كان يمكن دحض النظريات ما بعد الحقيقة حول الذكاء الاصطناعي علنًا. لتحقيق ذلك، قد يكون من الضروري زيادة ثقة الجمهور، ليس فقط في مفهوم الذكاء الاصطناعي نفسه، ولكن أيضًا في المنظمات والأنظمة التي تستخدمه. على سبيل المثال، قد يسعى المطورون إلى تعليم المستخدمين حول قدرات وحدود التقنيات المستقلة بطريقة مفهومة، بينما قد تسعى الشركات إلى وضع سياسات وإفصاحات أكثر سهولة. في الوقت نفسه، نلاحظ أن التغلب على المخاوف والشكوك المتأصلة حول الذكاء الاصطناعي سيظل على الأرجح تحديًا كبيرًا في المستقبل؛ ونتوقع أن تستمر التعقيدات والتطورات الكبيرة في الحواسيب الذكية في توفير أرض خصبة لنظريات المؤامرة. كما لا يمكن استبعاد أن المحاولات لجعل الذكاء الاصطناعي أكثر شفافية في المستقبل قد تؤدي أيضًا إلى نتائج عكسية، حيث قد يفسر منظرو المؤامرة هذه المحاولات كدليل فعلي على وجود مؤامرة في المقام الأول. علاوة على ذلك، فإن حقيقة أن الأفلام والبرامج التلفزيونية الشعبية تقدم غالبًا التكنولوجيا المعتمدة على الذكاء الاصطناعي بطريقة خطيرة أو مخيفة (على سبيل المثال، في وسائل الإعلام الخيالية الديستوبية قد يجعل ذلك من الصعب أكثر إقامة مستويات كافية من الثقة – خاصة بين أولئك الذين يميلون إلى أن يكون لديهم عقلية أكثر تشككًا في البداية.
بعيدًا عن الدلالات المثيرة التي تقدمها مؤشرات الشخصية الهامة، أظهرت عدة سمات أخرى تم فحصها عدم وجود ارتباط كبير بمواقف المشاركين تجاه الذكاء الاصطناعي. على وجه التحديد، وجدنا أن ثلاثة من الخمسة الكبار (الضمير الحي، الانفتاح، والعصابية) وجميع متغيرات الثالوث المظلم لم تكن كافية لشرح التباين الملحوظ في درجات ATTARI-12 التي تم الحصول عليها. مع الأخذ في الاعتبار النتائج المذكورة أعلاه، يشير هذا إلى أن شخصية المستخدم قد تكون لها إمكانيات قيمة ولكن محدودة لشرح قبول الجمهور لتكنولوجيا الذكاء الاصطناعي. بطبيعتها كسمات شاملة على مستوى ميتا، قد تكون بعض أبعاد الخمسة الكبار مجردة جدًا للتعامل مع الفروق الدقيقة التي تميز تجارب الناس مع – ومواقفهم تجاه – تكنولوجيا الذكاء الاصطناعي. على سبيل المثال، قد ينطوي كون الشخص أكثر انفتاحًا على جوانب تزيد من قبول الذكاء الاصطناعي (مثل كونه أقل قلقًا وتقييدًا بشكل عام) وتقلل منه (مثل تقدير الاتصال الاجتماعي الحقيقي مع البشر الآخرين، والذي قد يتعرض للتحدي من قبل تكنولوجيا الذكاء الاصطناعي الجديدة). وبالمثل، قد تكون السمات الشخصية الأكثر انحرافًا المدرجة في الثالوث المظلم مرتبطة برؤى إيجابية وسلبية تجاه الذكاء الاصطناعي، مما يمنع التنبؤ الكبير في اتجاه معين. تمامًا كما يمكن رؤية العديد من الإمكانيات التي تقدمها التكنولوجيا الذكية الجديدة كأدوات مفيدة للتلاعب أو تعزيز الذات، مما يشير إلى ارتباط إيجابي بالمكيافيلية والنرجسية – قد تثير أيضًا مخاوف بين الأشخاص الذين يحصلون على درجات عالية في هذه السمات، حيث قد تجعل الذكاء الاصطناعي من الصعب التصرف بدوافع خبيثة بطريقة غير مكتشفة أو مقبولة جيدًا.

القيود والاتجاهات المستقبلية

نود أن نشير إلى القيود المهمة التي يجب أخذها بعين الاعتبار عند تفسير النتائج المقدمة لمشروعنا، وخاصة الدراسة الثالثة والأخيرة التي كانت تهدف إلى معالجة فرضياتنا الرئيسية. على الرغم من أننا قمنا بتجنيد عينة متنوعة نسبيًا من حيث العمر والجنس والخلفية الاجتماعية والاقتصادية – وطبقنا عدة تدابير لضمان جودة البيانات العالية – فإن بياناتنا حول العلاقة بين مواقف الذكاء الاصطناعي وشخصية المستخدم تستند إلى مجموعة واحدة فقط من المشاركين. علاوة على ذلك، نظرًا لأن عينة دراستنا الثالثة تم تجنيدها بنفس الطريقة المستخدمة في الدراسة الأولى (MTurk)، لا يمكننا استبعاد أن بعض المشاركين قد شاركوا في كلا الجهدين البحثيين (على الرغم من أنه، استنادًا إلى مجموعة المشاركين الكبيرة جدًا في اللوحة المختارة، نعتبر أنه من غير المحتمل للغاية حدوث استجابات مكررة). وبالتالي، يُشجع على جهود التكرار مع عينات مختلفة من أجل تعزيز الأدلة المستخلصة. ولا سيما، يتعلق هذا بتجنيد عينات ذات توزيع جنسي أكثر توازنًا، حيث أسفرت عملية تجنيدنا عن أغلبية ملحوظة من المشاركين الذكور. أيضًا، بالنظر إلى أن العوامل الاجتماعية والاقتصادية والتعليمية تؤثر بشكل ملحوظ على آراء الناس تجاه التكنولوجيا (لا سيما فيما يتعلق بالذكاء الاصطناعي؛ انظر إن تحويل التركيز إلى خلفيات مختلفة يجب أن يكون الأكثر إضاءة فيما يتعلق بإمكانية تعميم عملنا. نحن نعتقد أن هذا قد يكون مهمًا بشكل خاص عند التركيز أكثر على بُعد عقلية المؤامرة – مع الأخذ في الاعتبار أن نقص التعليم، ومهارات الإعلام، والقدرة على التفكير التحليلي قد أظهرت جميعها أنها تتنبأ بالعرضة للمعلومات ما بعد الحقائق. .
على هذا النحو، نقترح إجراء دراسات مع عينات من ثقافات مختلفة من أجل تحديد ما إذا كانت نتائجنا متسقة عبر الحدود الوطنية. مؤخرًا، أبلغت دراسة تركزت على عينة من المشاركين الكوريين الجنوبيين أيضًا عن ارتباطات إيجابية بين المواقف تجاه الذكاء الاصطناعي وأبعاد الخمسة الكبرى من الشخصية، خاصة فيما يتعلق بالاجتماعية ووظائف الأنظمة المدعومة بالذكاء الاصطناعي. . على عكس بحثنا، لم يسعَ المؤلفون إلى قياس أحادي البعد، مما أدى إلى نتائج متناقضة (على سبيل المثال، الوعي الذاتي مرتبط بالعواطف السلبية تجاه الذكاء الاصطناعي ولكن أيضًا بتوقعات عالية للوظائف). لذلك، نعتبر أنه من المجدي تطبيق مقياس ATTARI-12 الأحادي البعد في ثقافات أخرى لفهم موقف الناس تجاه الذكاء الاصطناعي بشكل أفضل – وكيف تتشكل هذه المواقف بواسطة سمات شخصية معينة. مع الأخذ في الاعتبار أن مقياسنا الجديد قد سهل قياسًا موثوقًا به وأحادي البعد مع عينات في بلدين، يبدو أنه أداة واعدة لمثل هذه الفحوصات؛ على الرغم من أنه، بالطبع، من الواضح أن مزيدًا من التحقق من صحة أداتنا مرحب به، خاصة بالنسبة لنسخها غير الإنجليزية. علاوة على ذلك، سيتعين على المقارنات بين الثقافات ضمان قياس متسق لسمات الشخصية المعنية. بينما تم تسليط الضوء على كل من العوامل الخمسة الكبرى والثالوث المظلم لثباتهما الثقافي. ، وقد أثارت بعض الدراسات أيضًا شكوكًا حول هذا الادعاء -حتى أن قابلية تطبيق تصنيفات الشخصية المستخدمة قد تشكل تحديًا إضافيًا في هذا الصدد. قد تكون الحلول المحتملة هنا هي استخدام مخزون HEXACO للشخصية الذي يجمع بين كل من الخمسة الكبار وسمات الثلاثي المظلم المنحرفة في تصنيف جديد لشخصية الإنسان. لم يتلق نموذج HEXACO دعمًا قويًا فقط من حيث الثبات الثقافي استخدامه سيمكن الباحثين أيضًا من التحقق من صحة نتائج الدراسة الحالية مع أداة أخرى راسخة جيدًا. بالطبع، هناك سمات شخصية مختلفة تمامًا عن تلك التي تم تضمينها في مسعانا – مثل الاندفاع، والإحساس
السعي، أو الخوف من الفقدان، يمكن أن يكون مفيدًا أيضًا للحصول على فهم شامل للاختلافات بين الأفراد فيما يتعلق بمواقف الناس تجاه الذكاء الاصطناعي.
بالإضافة إلى ذلك، يجب الإشارة إلى أن الأبحاث المستقبلية في هذا الموضوع يمكن أن تستفيد بوضوح من تقييم الجوانب السياقية والدافعية التي تؤثر على آراء المشاركين حول الذكاء الاصطناعي بخلاف خصائص شخصياتهم. على سبيل المثال، تشير الأبحاث الحديثة إلى أن الآراء حول التكنولوجيا المتطورة تتأثر بشكل كبير بالتعرض السابق لوسائط الخيال العلمي. بالإضافة إلى الآراء الفلسفية والمعايير الأخلاقية . علاوة على ذلك، وُجد أن الناس يغيرون مواقفهم بعد عدة لقاءات مع تكنولوجيا الذكاء الاصطناعي , مما يضيف عاملاً آخر إلى المعادلة. معًا، يشير هذا إلى أن الدراسات المستقبلية قد تستفيد من عدة متغيرات عميقة لفك تشابك الحالات والسمات التي تؤثر على المواقف تجاه الذكاء الاصطناعي. على أي حال، مع الأخذ في الاعتبار أن مقياس ATTARI-12 الذي طورناه حديثًا قد سهل قياسًا موثوقًا، أحادي البعد، مع عينات من فئات عمرية وخلفيات ثقافية مختلفة، نقترح أنه يشكل حجر الزاوية المنهجي الواعد لفحوصات مستقبلية لمواقف الذكاء الاصطناعي.

الخاتمة

في مشروعنا، بحثنا في المواقف تجاه الذكاء الاصطناعي كمقياس مركب لأفكار الناس ومشاعرهم ونواياهم السلوكية. سعيًا لتكملة الأبحاث السابقة التي ركزت أكثر على قبول تطبيقات الذكاء الاصطناعي المتخصصة، أنشأنا واستخدمنا مقياسًا جديدًا أحادي البعد: ATTARI-12. نحن واثقون من أن المقياس المطور قد يكون الآن أداة مفيدة للتعامل عمليًا مع الذكاء الاصطناعي كظاهرة على مستوى ماكرو، والتي – على الرغم من احتوائها على مجموعة من البرامج والتطبيقات المختلفة – تتحد من خلال عدة أسس مشتركة. في توقعنا، يمكن أن يتم تطبيق المقياس الذي تم إنشاؤه حتى عند ظهور تقنيات ذكاء اصطناعي جديدة وغير متوقعة، حيث يركز أكثر على المبادئ الأساسية بدلاً من القدرات المحددة. بالنسبة للممارسين والمطورين، قد تكون هذه النظرة مفيدة بشكل خاص، مما يشير إلى أن استجابة الأفراد لتقنية الذكاء الاصطناعي الجديدة (مثل العملاء أو الموظفين) لن تكون مدفوعة فقط بالميزات المحددة لتقنية معينة، ولكن أيضًا بموقف عام تجاه الذكاء الاصطناعي. وبالمثل، نأمل أن تتمكن الجهود السوسيوديموغرافية واسعة النطاق (مثل استبيانات الرأي الوطنية) من الاستفادة من المقياس المقدم.
بعيدًا عن ذلك، أنشأنا روابط مهمة بين المواقف العامة تجاه الذكاء الاصطناعي وسمتين شخصيتين مختارتين. في رأينا، فإن الارتباط الملحوظ مع عقلية المؤامرة لدى المشاركين يحمل أهمية ملحوظة في عالمنا المتزايد التعقيد. إذا لم يتمكن مطورو الذكاء الاصطناعي من إيجاد طرق مناسبة لجعل تقنيتهم تبدو غير ضارة للمراقبين (خصوصًا لأولئك الذين يميلون إلى البحث عن تفسيرات ما بعد الحقائق)، فقد يصبح من الصعب بشكل متزايد إنشاء ابتكارات على نطاق أوسع. ومع ذلك، نؤكد أن المواقف السلبية والاعتراضات ضد تقنية الذكاء الاصطناعي ليست بالضرورة غير مبررة؛ لذا، عند توعية الجمهور بتقنيات الذكاء الاصطناعي، يجب تشجيع الناس على التفكير في كل من المخاطر والفوائد المحتملة. لتقليل خطر ظهور نظريات مؤامرة جديدة، نشجع أيضًا المحترفين في الصناعة على أن يكونوا شفافين قدر الإمكان عند تقديم ابتكاراتهم للجمهور.
أخيرًا، نقترح أن العمل اللاحق ضروري تمامًا لتفصيل مساهمتنا الحالية. بينما نظل مقتنعين بأن قياس المواقف تجاه الذكاء الاصطناعي كمجموعة عامة من القدرات التكنولوجية هو أمر جدير بالاهتمام، يمكن أن تأتي نتائج أكثر ملموسة من تضمين ثنائية نظرية ظهرت مؤخرًا في مجال التفاعل بين الإنسان والآلة. بشكل أكثر تحديدًا، قد يكون من المفيد التمييز بين قدرات الذكاء الاصطناعي التي تتعلق بالوكالة (أي التخطيط والتفكير والعمل) وتلك التي تتعلق بالتجربة (أي الإحساس والشعور). بدوره، قد يتمكن الباحثون من اكتشاف ما إذا كانت سمات المستخدمين المختلفة ترتبط أيضًا بمواقف مختلفة تجاه الذكاء الاصطناعي “الذي يعمل” و”الذي يشعر” – وهي وجهة نظر دقيقة لا تزال تسمح بتداعيات أوسع عبر العديد من السياقات التكنولوجية المختلفة.

توفر البيانات

يمكن العثور على جميع المواد والبيانات التي تم الحصول عليها وأكواد التحليل للدراسات الثلاث المبلغ عنها في مستودع إطار العلوم المفتوحة الخاص بالمشروع (https://osf.io/3j67a/).

الملحق

مقياس المواقف تجاه الذكاء الاصطناعي (ATTARI-12)، النسخة الإنجليزية

التعليمات: في ما يلي، نحن مهتمون بمواقفك تجاه الذكاء الاصطناعي (AI). يمكن للذكاء الاصطناعي تنفيذ مهام تتطلب عادةً الذكاء البشري. إنه يمكّن الآلات من الإحساس والعمل والتعلم والتكيف بطريقة مستقلة تشبه الإنسان. قد يكون الذكاء الاصطناعي جزءًا من جهاز كمبيوتر أو منصة عبر الإنترنت – ولكنه يمكن أيضًا أن يُواجه في مجموعة متنوعة من الأجهزة الأخرى مثل الروبوتات.
قائمة العناصر:
الصياغة الجانب القيمة
1 سيجعل الذكاء الاصطناعي هذا العالم مكانًا أفضل. معرفي إيجابي
2 لدي مشاعر سلبية قوية تجاه . عاطفي سلبي (معكوس)
3 أريد استخدام تقنيات تعتمد على . سلوكي إيجابي
4 لدى الذكاء الاصطناعي المزيد من العيوب مقارنة بالمزايا. معرفي سلبي (معكوس)
5 أتطلع إلى مستقبل التطورات. عاطفي إيجابي
6 يقدم حلولاً للعديد من مشاكل العالم. معرفي إيجابي
7 أفضل التقنيات التي لا تحتوي على . سلوكي سلبي (معكوس)
الصياغة الجانب القيمة
8 أخاف من . عاطفي سلبي (معكوس)
9 أفضل اختيار تقنية تحتوي على بدلاً من واحدة بدونها. سلوكي إيجابي
10 يخلق الذكاء الاصطناعي مشاكل بدلاً من حلها. معرفي سلبي (معكوس)
11 عندما أفكر في ، أشعر في الغالب بمشاعر إيجابية. عاطفي إيجابي
12 أفضل تجنب التقنيات التي تعتمد على . سلوكي سلبي (معكوس)
تم الاستلام: 3 أغسطس 2023؛ تم القبول: 31 يناير 2024
تم النشر عبر الإنترنت: 05 فبراير 2024

References

  1. Grewal, D., Roggeveen, A. L. & Nordfält, J. The future of retailing. J. Retail. 93(1), 1-6. https://doi.org/10.1016/j.jretai.2016.12.008 (2017).
  2. Vanian, J. Artificial Intelligence will Obliterate These Jobs by 2030 (Fortune, 2019). https://fortune.com/2019/11/19/artificial-intel ligence-will-obliterate-these-jobs-by-2030/.
  3. Ivanov, S., Kuyumdzhiev, M. & Webster, C. Automation fears: Drivers and solutions. Technol. Soc. 63, 101431. https://doi.org/10. 1016/j.techsoc.2020.101431 (2020).
  4. Waytz, A. & Norton, M. I. Botsourcing and outsourcing: Robot, British, Chinese, and German workers are for thinking-not feeling—jobs. Emotion 14, 434-444. https://doi.org/10.1037/a0036054 (2014).
  5. Gherheş, V. Why are we afraid of artificial intelligence (AI)?. Eur. Rev. Appl. Sociol. 11(17), 6-15. https://doi.org/10.1515/eras-2018-0006 (2018).
  6. Liang, Y. & Lee, S. A. Fear of autonomous robots and artificial intelligence: Evidence from national representative data with probability sampling. Int. J. Soc. Robot. 9, 379-384. https://doi.org/10.1007/s12369-017-0401-3 (2017).
  7. Fast, E., & Horvitz, E. Long-term trends in the public perception of artificial intelligence. In Proceedings of the 31st AAAI Conference on Artificial Intelligence pp. 963-969 (AAAI Press, 2016).
  8. Thorp, H. H. ChatGPT is fun, but not an author. Science 379(6630), 313. https://doi.org/10.1126/science.adg7879 (2023).
  9. Lobera, J., Fernández Rodríguez, C. J. & Torres-Albero, C. Privacy, values and machines: Predicting opposition to artificial intelligence. Commun. Stud. 71(3), 448-465. https://doi.org/10.1080/10510974.2020.1736114 (2020).
  10. Laakasuo, M. et al. The dark path to eternal life: Machiavellianism predicts approval of mind upload technology. Pers. Individ. Differ. 177, 110731. https://doi.org/10.1016/j.paid.2021.110731 (2021).
  11. Qu, W., Sun, H. & Ge, Y. The effects of trait anxiety and the big five personality traits on self-driving car acceptance. Transportation. https://doi.org/10.1007/s11116-020-10143-7 (2020).
  12. Kaplan, J. Artificial Intelligence: What Everyone Needs to Know. (Oxford University Press, 2016).
  13. Schepman, A. & Rodway, P. Initial validation of the general attitudes towards Artificial Intelligence Scale. Comput. Hum. Behav. Rep. 1, 100014. https://doi.org/10.1016/j.chbr.2020.100014 (2020).
  14. Sindermann, C. et al. Assessing the attitude towards artificial intelligence: Introduction of a short measure in German, Chinese, and English language. Künstliche Intelligenz 35(1), 109-118. https://doi.org/10.1007/s13218-020-00689-0 (2020).
  15. Wang, Y. Y., & Wang, Y. S. Development and validation of an artificial intelligence anxiety scale: An initial application in predicting motivated learning behavior. Interact. Learn. Environ. https://doi.org/10.1080/10494820.2019.1674887 (2019).
  16. Kieslich, K., Lünich, M. & Marcinkowski, F. The Threats of Artificial Intelligence Scale (TAI): Development, measurement and test over three application domains. Int. J. Soc. Robot. https://doi.org/10.1007/s12369-020-00734-w (2021).
  17. Stein, J.-P., Liebold, B. & Ohler, P. Stay back, clever thing! Linking situational control and human uniqueness concerns to the aversion against autonomous technology. Comput. Hum. Behav. 95, 73-82. https://doi.org/10.1016/j.chb.2019.01.021 (2019).
  18. Rosenberg, M. J., & Hovland, C. I. Attitude Organization and Change: An Analysis of Consistency Among Attitude Components (Yale University Press, 1960).
  19. Zanna, M. P. & Rempel, J. K. Attitudes: A new look at an old concept. In Attitudes: Their Structure, Function, and Consequences (eds Fazio, R. H. & Petty, R. E.) 7-15 (Psychology Press, 2008).
  20. Stein, J.-P., Appel, M., Jost, A. & Ohler, P. Matter over mind? How the acceptance of digital entities depends on their appearance, mental prowess, and the interaction between both. Int. J. Hum.-Comput. Stud. 142, 102463. https://doi.org/10.1016/j.ijhcs.2020. 102463 (2020).
  21. McClure, P. K. “You’re fired”, says the robot. Soc. Sci. Comput. Rev. 36(2), 139-156. https://doi.org/10.1177/0894439317698637 (2017).
  22. Gillath, O. et al. Attachment and trust in artificial intelligence. Comput. Hum. Behav. 115, 106607. https://doi.org/10.1016/j.chb. 2020.106607 (2021).
  23. Young, K. L. & Carpenter, C. Does science fiction affect political fact? Yes and no: A survey experiment on “Killer Robots”. Int. Stud. Q. 62(3), 562-576. https://doi.org/10.1093/isq/sqy028 (2018).
  24. Gnambs, T. & Appel, M. Are robots becoming unpopular? Changes in attitudes towards autonomous robotic systems in Europe. Comput. Hum. Behav. 93, 53-61. https://doi.org/10.1016/j.chb.2018.11.045 (2019).
  25. Horstmann, A. C. & Krämer, N. C. Great expectations? Relation of previous experiences with social robots in real life or in the media and expectancies based on qualitative and quantitative assessment. Front. Psychol. 10, 939. https://doi.org/10.3389/fpsyg. 2019.00939 (2019).
  26. Sundar, S. S., Waddell, T. F. & Jung, E. H. The Hollywood robot syndrome: Media effects on older adults’ attitudes toward robots and adoption intentions. Paper presented at the 11th Annual ACM/IEEE International Conference on Human-Robot Interaction, Christchurch, New Zealand (2016).
  27. Chien, S.-Y., Sycara, K., Liu, J.-S. & Kumru, A. Relation between trust attitudes toward automation, Hofstede’s cultural dimensions, and Big Five personality traits. Proc. Hum. Factors Ergon. Soc. Annu. Meet. 60(1), 841-845. https://doi.org/10.1177/1541931213 601192 (2016).
  28. MacDorman, K. F. & Entezari, S. Individual differences predict sensitivity to the uncanny valley. Interact. Stud. 16(2), 141-172. https://doi.org/10.1075/is.16.2.01mac (2015).
  29. Oksanen, A., Savela, N., Latikka, R. & Koivula, A. Trust toward robots and artificial intelligence: An experimental approach to Human-Technology interactions online. Front. Psychol. 11, 3336. https://doi.org/10.3389/fpsyg.2020.568256 (2020).
  30. Wissing, B. G. & Reinhard, M.-A. Individual differences in risk perception of artificial intelligence. Swiss J. Psychol. 77(4), 149-157. https://doi.org/10.1024/1421-0185/a000214 (2018).
  31. Costa, P. T. & McCrae, R. R. Four ways five factors are basic. Pers. Individ. Differ. 13, 653-665. https://doi.org/10.1016/0191-8869(92)90236-I (1992).
  32. Paulhus, D. L. & Williams, K. M. The Dark Triad of personality: Narcissism, Machiavellianism, and psychopathy. J. Res. Pers. 36, 556-563. https://doi.org/10.1016/S0092-6566(02)00505-6 (2002).
  33. Bruder, M., Haffke, P., Neave, N., Nouripanah, N. & Imhoff, R. Measuring individual differences in generic beliefs in conspiracy theories across cultures: Conspiracy mentality questionnaire. Front. Psychol. 4, 225. https://doi.org/10.3389/fpsyg.2013.00225 (2013).
  34. Imhoff, R. et al. Conspiracy mentality and political orientation across 26 countries. Nat. Hum. Behav. 6(3), 392-403. https://doi. org/10.1038/s41562-021-01258-7 (2022).
  35. Nettle, D. & Penke, L. Personality: Bridging the literatures from human psychology and behavioural ecology. Philos. Trans. R. Soc. B 365(1560), 4043-4050. https://doi.org/10.1098/rstb.2010.0061 (2010).
  36. Poropat, A. E. A meta-analysis of the five-factor model of personality and academic performance. Psychol. Bull. 135(2), 322-338. https://doi.org/10.1037/a0014996 (2009).
  37. Kaplan, A. D., Sanders, T. & Hancock, P. A. The relationship between extroversion and the tendency to anthropomorphize robots: A bayesian analysis. Front. Robot. AI 5, 135. https://doi.org/10.3389/frobt.2018.00135 (2019).
  38. Kaufman, S. B., Yaden, D. B., Hyde, E. & Tsukayama, E. The light vs. Dark triad of personality: Contrasting two very different profiles of human nature. Front. Psychol. 10, 467. https://doi.org/10.3389/fpsyg.2019.00467 (2019).
  39. LeBreton, J. M., Shiverdecker, L. K. & Grimaldi, E. M. The Dark Triad and workplace behavior. Annu. Rev. Organ. Psychol. Organ. Behav. 5(1), 387-414. https://doi.org/10.1146/annurev-orgpsych-032117-104451 (2018).
  40. Feldstein, S. The Global Expansion of AI Surveillance. Carnegie Endowment for International Piece. https://carnegieendowment. org/files/WP-Feldstein-AISurveillance_final1.pdf (2019).
  41. Malabou, C. Morphing Intelligence: From IQ Measurement to Artificial Brains (Columbia University Press, 2019).
  42. Swami, V., Chamorro-Premuzic, T. & Furnham, A. Unanswered questions: A preliminary investigation of personality and individual difference predictors of 9/11 conspiracist beliefs. Appl. Cognit. Psychol. 24, 749-761. https://doi.org/10.1002/acp.1583 (2010).
  43. Pennycook, G. & Rand, D. G. Who falls for fake news? The roles of bullshit receptivity, overclaiming, familiarity, and analytic thinking. J. Pers. 88(2), 185-200. https://doi.org/10.1111/jopy. 12476 (2019).
  44. Shakarian, A. The Artificial Intelligence Conspiracy (2016).
  45. McEvoy, J. Microchips, Magnets and Shedding: Here are 5 (Debunked) Covid Vaccine Conspiracy Theories Spreading Online (Forbes, 2021). https://www.forbes.com/sites/jemimamcevoy/2021/06/03/microchips-and-shedding-here-are-5-debunked-covid-vacci ne-conspiracy-theories-spreading-online/
  46. De Vynck, G., & Lerman, R. Facebook and YouTube Spent a Year Fighting Covid Misinformation. It’s Still Spreading (The Washington Post, 2021). https://www.washingtonpost.com/technology/2021/07/22/facebook-youtube-vaccine-misinformation/.
  47. González, F., Yu, Y., Figueroa, A., López, C., & Aragon, C. Global reactions to the Cambridge Analytica scandal: A cross-language social media study. In Proceedings of the 2019 World Wide Web Conference 799-806. ACM Press. https://doi.org/10.1145/33085 60.3316456 (2019).
  48. Joyce, M. & Kirakowski, J. Measuring attitudes towards the internet: The general internet attitude scale. Int. J. Hum.-Comput. Interact. 31, 506-517. https://doi.org/10.1080/10447318.2015.1064657 (2015).
  49. Bartneck, C., Lütge, C., Wagner, A., & Welsh, S. An Introduction to Ethics in Robotics and AI (Springer, 2021).
  50. German Research Foundation. Statement by an Ethics Committee. https://www.dfg.de/en/research_funding/faq/faq_humanities_ social_science/index.html (2023).
  51. German Psychological Society. Berufsethische Richtlinien [Work Ethical Guidelines]. https://www.dgps.de/fileadmin/user_upload/ PDF/Berufsetische_Richtlinien/BER-Foederation-20230426-Web-1.pdf (2022).
  52. Jobst, L. J., Bader, M. & Moshagen, M. A tutorial on assessing statistical power and determining sample size for structural equation models. Psychol. Methods 28(1), 207-221. https://doi.org/10.1037/met0000423 (2023).
  53. Stöber, J. The Social Desirability Scale-17 (SDS-17): Convergent validity, discriminant validity, and relationship with age. Eur. J. Psychol. Assess. 17(3), 222-232. https://doi.org/10.1027/1015-5759.17.3.222 (2001).
  54. Eid, M., Geiser, C., Koch, T. & Heene, M. Anomalous results in G-factor models: Explanations and alternatives. Psychol. Methods 22, 541-562. https://doi.org/10.1037/met0000083 (2017).
  55. Schermelleh-Engel, K., Moosbrugger, H. & Müller, H. Evaluating the fit of structural equation models: Tests of significance and descriptive goodness-of-fit measures. Methods Psychol. Res. 8(2), 23-74 (2003).
  56. Rodriguez, A., Reise, S. P. & Haviland, M. G. Evaluating bifactor models: Calculating and interpreting statistical indices. Psychol. Methods 21(2), 137-150. https://doi.org/10.1037/met0000045 (2016).
  57. Kennedy, R. et al. The shape of and solutions to the MTurk quality crisis. Polit. Sci. Res. Methods 8(4), 614-629. https://doi.org/10. 1017/psrm. 2020.6 (2020).
  58. Peer, E., Vosgerau, J. & Acquisti, A. Reputation as a sufficient condition for data quality on Amazon Mechanical Turk. Behav. Res. Methods 46(4), 1023-1031. https://doi.org/10.3758/s13428-013-0434-y (2013).
  59. John, O. P., Naumann, L. P. & Soto, C. J. Paradigm shift to the integrative Big-Five trait taxonomy: History, measurement, and conceptual issues. In Handbook of Personality: Theory and Research (eds John, O. P. et al.) 114-158 (Guilford Press, 2008).
  60. Jones, D. N. & Paulhus, D. L. Introducing the Short Dark Triad (SD3): A brief measure of dark personality traits. Assessment 21(1), 28-41. https://doi.org/10.1177/1073191113514105 (2014).
  61. Field, A. Discovering Statistics Using IBM SPSS Statistics (SAGE, 2013).
  62. Baranski, E., Sweeny, K., Gardiner, G. & Funder, D. C. International optimism: Correlates and consequences of dispositional optimism across 61 countries. J. Pers. 89(2), 288-304. https://doi.org/10.1111/jopy. 12582 (2020).
  63. Sharpe, J. P., Martin, N. R. & Roth, K. A. Optimism and the Big Five factors of personality: Beyond neuroticism and extraversion. Pers. Individ. Differ. 51(8), 946-951. https://doi.org/10.1016/j.paid.2011.07.033 (2011).
  64. Steel, G. D., Rinne, T. & Fairweather, J. Personality, nations, and innovation. Cross-Cult. Res. 46(1), 3-30. https://doi.org/10.1177/ 1069397111409124 (2011).
  65. McCarthy, M. H., Wood, J. V. & Holmes, J. G. Dispositional pathways to trust: Self-esteem and agreeableness interact to predict trust and negative emotional disclosure. J. Pers. Soc. Psychol. 113(1), 95-116. https://doi.org/10.1037/pspi0000093 (2017).
  66. Lazer, D. M. J. et al. The science of fake news. Science 359(6380), 1094-1096. https://doi.org/10.1126/science.aao2998 (2018).
  67. Stecula, D. A. & Pickup, M. Social media, cognitive reflection, and conspiracy beliefs. Front. Polit. Sci. 3, 62. https://doi.org/10. 3389/fpos.2021.647957 (2021).
  68. van Prooijen, J. W. Why education predicts decreased belief in conspiracy theories. Appl. Cognit. Psychol. 31(1), 50-58. https:// doi.org/10.1002/acp. 3301 (2016).
  69. Scheibenzuber, C., Hofer, S. & Nistor, N. Designing for fake news literacy training: A problem-based undergraduate online-course. Comput. Hum. Behav. 121, 106796. https://doi.org/10.1016/j.chb.2021.106796 (2021).
  70. Sindermann, C., Schmitt, H. S., Rozgonjuk, D., Elhai, J. D. & Montag, C. The evaluation of fake and true news: On the role of intelligence, personality, interpersonal trust, ideological attitudes, and news consumption. Heliyon 7(3), e06503. https://doi.org/ 10.1016/j.heliyon.2021.e06503 (2021).
  71. Park, J. & Woo, S. E. Who likes artificial intelligence? Personality predictors of attitudes toward artificial intelligence. J. Psychol. 156(1), 68-94. https://doi.org/10.1080/00223980.2021.2012109 (2022).
  72. McCrae, R. R. & Costa, P. T. Jr. Personality trait structure as a human universal. Am. Psychol. 52(5), 509-516. https://doi.org/10. 1037/0003-066X.52.5.509 (1997).
  73. Rogoza, R. et al. Structure of Dark Triad Dirty Dozen across eight world regions. Assessment 28(4), 1125-1135. https://doi.org/ 10.1177/1073191120922611 (2020).
  74. Gurven, M., von Rueden, C., Massenkoff, M., Kaplan, H. & Lero Vie, M. How universal is the Big Five? Testing the five-factor model of personality variation among forager-farmers in the Bolivian Amazon. J. Pers. Soc. Psychol. 104(2), 354-370. https://doi. org/10.1037/a0030841 (2013).
  75. Lee, K. & Ashton, M. C. Psychometric properties of the HEXACO-100. Assessment 25, 543-556. https://doi.org/10.1177/10731 91116659134 (2018).
  76. Thielmann, I. et al. The HEXACO-100 across 16 languages: A Large-Scale test of measurement invariance. J. Pers. Assess. 102(5), 714-726. https://doi.org/10.1080/00223891.2019.1614011 (2019).
  77. Mara, M. & Appel, M. Science fiction reduces the eeriness of android robots: A field experiment. Comput. Hum. Behav. 48, 156-162. https://doi.org/10.1016/j.chb.2015.01.007 (2015).
  78. Laakasuo, M. et al. What makes people approve or condemn mind upload technology? Untangling the effects of sexual disgust, purity and science fiction familiarity. Palgrave Commun. https://doi.org/10.1057/s41599-018-0124-6 (2018).
  79. Latikka, R., Savela, N., Koivula, A. & Oksanen, A. Attitudes toward robots as equipment and coworkers and the impact of robot autonomy level. Int. J. Soc. Robot. 13(7), 1747-1759. https://doi.org/10.1007/s12369-020-00743-9 (2021).
  80. Appel, M., Izydorczyk, D., Weber, S., Mara, M. & Lischetzke, T. The uncanny of mind in a machine: Humanoid robots as tools, agents, and experiencers. Comput. Hum. Behav. 102, 274-286. https://doi.org/10.1016/j.chb.2019.07.031 (2020).
  81. Gray, K. & Wegner, D. M. Feeling robots and human zombies: Mind perception and the uncanny valley. Cognition 125, 125-130. https://doi.org/10.1016/j.cognition.2012.06.007 (2012).
  82. Grundke, A., Stein, J.-P. & Appel, M. Mind-reading machines: Distinct user responses to thought-detecting and emotion-detecting robots. Technol. Mind Behav. https://doi.org/10.17605/OSF.IO/U52KM (2021).
  83. Yuan, K. H. & Bentler, P. M. Three likelihood-based methods for mean and covariance structure analysis with non-normal missing data. Sociol. Methodol. 30(1), 165-200. https://doi.org/10.1111/0081-1750.00078 (2000).
  84. Satorra, A. & Bentler, P. M. Ensuring positiveness of the scaled difference chi-square test statistic. Psychometrika 75, 243-248. https://doi.org/10.1007/s11336-009-9135-y (2010).

مساهمات المؤلفين

J.-P.S.: التصور، المنهجية، التحقيق، التحليل الرسمي، تنسيق البيانات، الكتابة (المسودة الأصلية)، الكتابة (المراجعة والتحرير).T.M.: التصور، المنهجية، التحقيق، تنسيق البيانات، الكتابة (المراجعة والتحرير).T.G.: المنهجية، التحقيق، التحليل الرسمي، تنسيق البيانات، الكتابة (المراجعة والتحرير).F.H.: التصور، المنهجية، التحقيق، الكتابة (المراجعة والتحرير).M.A.: التصور، المنهجية، التحليل الرسمي، تنسيق البيانات، الكتابة (المسودة الأصلية)، الكتابة (المراجعة والتحرير)، الإشراف.

التمويل

تم تمكين وتنظيم تمويل الوصول المفتوح بواسطة مشروع DEAL.

المصالح المتنافسة

يعلن المؤلفون عدم وجود مصالح متنافسة.

معلومات إضافية

المعلومات التكميلية تحتوي النسخة عبر الإنترنت على مواد تكميلية متاحة على https://doi.org/ 10.1038/s41598-024-53335-2.
يجب توجيه المراسلات وطلبات المواد إلى J.-P.S.
معلومات إعادة الطبع والتصاريح متاحة على www.nature.com/reprints.
ملاحظة الناشر تظل Springer Nature محايدة فيما يتعلق بالمطالبات القضائية في الخرائط المنشورة والانتماءات المؤسسية.
الوصول المفتوح هذه المقالة مرخصة بموجب رخصة المشاع الإبداعي 4.0 الدولية، التي تسمح بالاستخدام والمشاركة والتكيف والتوزيع وإعادة الإنتاج في أي وسيلة أو تنسيق، طالما أنك تعطي الائتمان المناسب للمؤلفين الأصليين والمصدر، وتوفر رابطًا لرخصة المشاع الإبداعي، وتوضح ما إذا تم إجراء تغييرات. الصور أو المواد الأخرى من طرف ثالث في هذه المقالة مشمولة في رخصة المشاع الإبداعي للمقالة، ما لم يُشار إلى خلاف ذلك في سطر ائتمان للمادة. إذا لم تكن المادة مشمولة في رخصة المشاع الإبداعي للمقالة وكان استخدامك المقصود غير مسموح به بموجب اللوائح القانونية أو يتجاوز الاستخدام المسموح به، ستحتاج إلى الحصول على إذن مباشرة من صاحب حقوق الطبع والنشر. لعرض نسخة من هذه الرخصة، قم بزيارة http://creativecommons.org/licenses/by/4.0/.
© المؤلفون 2024

  1. قسم علم نفس الإعلام، معهد أبحاث الإعلام، جامعة كمنيتز للتكنولوجيا، طريق ثورينجر 11، 09126 كمنيتز، ألمانيا. علم نفس الاتصال والإعلام الجديد، معهد الإنسان-الكمبيوتر-الإعلام، جامعة فيورزبورغ، فيورزبورغ، ألمانيا. معهد لايبنيز لمسارات التعليم، بامبرغ، ألمانيا. البريد الإلكتروني: jpstein@phil.tu-chemnitz.de

Journal: Scientific Reports, Volume: 14, Issue: 1
DOI: https://doi.org/10.1038/s41598-024-53335-2
PMID: https://pubmed.ncbi.nlm.nih.gov/38316898
Publication Date: 2024-02-05

OPEN

Attitudes towards AI: measurement and associations with personality

Jan-Philipp Stein , Tanja Messingschlager , Timo Gnambs , Fabian Hutmacher & Markus Appel

Artificial intelligence (AI) has become an integral part of many contemporary technologies, such as social media platforms, smart devices, and global logistics systems. At the same time, research on the public acceptance of AI shows that many people feel quite apprehensive about the potential of such technologies-an observation that has been connected to both demographic and sociocultural user variables (e.g., age, previous media exposure). Yet, due to divergent and often ad-hoc measurements of AI-related attitudes, the current body of evidence remains inconclusive. Likewise, it is still unclear if attitudes towards AI are also affected by users’ personality traits. In response to these research gaps, we offer a two-fold contribution. First, we present a novel, psychologically informed questionnaire (ATTARI-12) that captures attitudes towards AI as a single construct, independent of specific contexts or applications. Having observed good reliability and validity for our new measure across two studies ( ), we examine several personality traits-the Big Five, the Dark Triad, and conspiracy mentality-as potential predictors of AI-related attitudes in a third study ( ). We find that agreeableness and younger age predict a more positive view towards artificially intelligent technology, whereas the susceptibility to conspiracy beliefs connects to a more negative attitude. Our findings are discussed considering potential limitations and future directions for research and practice.
Artificial intelligence (AI) promises to make human life much more comfortable: It may foster intercultural collaboration via sophisticated translating tools, guide online customers towards the products they are most likely going to buy, or carry out jobs that feel tedious to human workers . At the same time, the proliferation of AI-based technologies is met with serious reservations. It is argued, for instance, that AI could lead to the downsizing of human jobs , the creation of new intelligent weaponry , or a growing lack of control over the emerging technologies . Similarly, the recent and much-publicized presentation of new text- and image-creating AI (such as ChatGPT and Midjourney) has raised concerns about artistic license, academic fraud, and the value of human creativity .
Crucially, individuals may differ regarding their evaluation of such chances and risks, and in turn, hold different attitudes towards AI. However, we note that scholars interested in these differences often used a basic ad-hoc approach to assess participants’ AI attitudes , or focused only on highly domain-specific sub-types of . Against this background, the current paper offers a twofold contribution. First, we present a novel questionnaire on AI-related attitudes-the ATTARI-12, which incorporates the psychological trichotomy of cognition, emotion, and behavior to facilitate a comprehensive yet economic measurement-and scrutinize its validity across two studies. Subsequently, we connect participants’ attitudes (as measured by our new scale) to several fundamental personality traits, namely the Big Five, the Dark Triad, and conspiracy mentality.

Research objective 1: measuring attitudes towards AI

Broadly speaking, the term artificial intelligence (AI) is used to describe technologies that can execute tasks one may expect to require human intelligence-i.e., technologies that possess a certain degree of autonomy, the capacity for learning and adapting, and the ability to handle large amounts of data . As AI becomes deeply embedded in many technical systems and, thus, people’s daily lives, it emerges as an important task for numerous scientific disciplines (e.g. psychology, communication science, computer science, philosophy) to
better understand users’ response to-and acceptance of-artificially intelligent technology. Of course, in order to achieve this goal, it is essential to employ theoretically sound measures of high psychometric quality when assessing AI-related attitudes. Furthermore, researchers need to decide whether they want to examine only a specific type of application (such as self-driving cars or intelligent robots) or focus on AI as an abstract technological concept that can be applied to many different settings. While both approaches have undeniable merit, it may be argued that the comparability of different research efforts-especially those situated at the intersection of different disciplines-clearly benefits from the latter, i.e., an empirical focus on attitudes towards AI as a set of technological capabilities instead of concrete use cases.
In line with this argument, scientific efforts have produced several measures to assess attitudes towards AI as a more overarching concept. These include the General Attitudes Towards Artificial Intelligence Scale (GAAIS ), the Attitudes Towards Artificial Intelligence Scale (ATAI ), the AI Anxiety Scale (AIAS ), the Threats of Artificial Intelligence Scale (TAI ), and the Concerns About Autonomous Technology questionnaire . Upon thorough examination, however, we believe that none of these measures offer an entirely satisfactory option for at least one of five reasons. First, many of the above-mentioned measures only inquire participants about negative impressions and concerns regarding AI, leaving aside the possibility of distinctly positive attitudes . Second, the measured attitudes towards AI are often subdivided into several factors , which complicates handling and interpreting these scales from both a theoretical and a practical perspective. For instance, the GAAIS presents two sub-scales called “acceptance” and “fear,” even though it can be argued that these merely cover different poles of the same spectrum. Relatedly, the AIAS scale offers a four-factor solution with the (statistically induced) dimensions “learning,” “job replacement,” “sociotechnical blindness,” and “AI configuration”-clusters whose practical merit may be related to more specific use cases. Third, while some scales have a relatively low number of items (e.g., five items ), other scales are comparably long (e.g., 21 items ), which may complicate their use in research settings that need to include attitudes towards AI as one concept among multiple others. Fourth, for one of the reviewed scales , the low number of items resulted in a notable lack of reliability. Fifth, none of the available scales acknowledge the cognitive, affective, and behavioral facets of attitudes in their designs .
Hence, based on the identified lack of a one-dimensional, psychometrically sound, yet economic measure that captures both positive and negative aspects of the attitude towards AI, our first aim was to develop a scale that overcomes these limitations. For the sake of broad interdisciplinary applicability, we explicitly focused the creation of our scale on the perception of AI as a general set of technological capabilities, independent of its physical embodiment or use context. This perspective was not least informed by recent research, which indicated that people’s evaluation of digital ‘minds’ is likely to change once visual factors come into play -further highlighting the merit of addressing AI as a disembodied concept.
Thus far, most studies on public AI acceptance have explored the role of demographic and sociocultural variables, revealing that concerns about AI seem to be more prevalent among women, the elderly, ethnic minorities, and individuals with lower levels of education . Furthermore, it was demonstrated that egalitarian values and distrust in science , attachment anxiety , and the exposure to dystopian science fiction constitute meaningful predictors of negative attitudes towards AI. Indeed, many of these findings are echoed by a large body of human-robot interaction literature, which also emphasizes the crucial impact of demographic variables, cultural norms, and media use on the public acceptance of robotic machinery . Yet, considering that studies from this discipline rarely elucidate whether the reported effects arise from the perception of robotic bodies or robotic minds (i.e., AI), we suggest that they should only be considered as secondary evidence for the current topic.
Apart from considerable insight into the sociocultural predictors of (dis-)liking AI technology, much less is known about the impact of users’ personality factors on their AI-related attitudes. To our knowledge, only a handful of scientific studies have actually addressed this issue, and most of them have yielded rather limited results-either by focusing on very specific types of or by measuring reactions to physically embodied AI (e.g., robots), which makes it much harder to pinpoint the cause of the effect . While another recent study indeed explored attitudes towards AI as a technical concept regardless of its specific use or embodiment , this effort focused exclusively on negative attitudes, thus leaving out a substantial part (i.e., the positive side) of how users think and feel about artificially intelligent systems.
Therefore, building upon the reviewed work, our second objective was to scrutinize several central personality dimensions as predictors for people’s AI-related attitudes, including the well-known Big Five and Dark Triad of personality . In addition to that, we include the rather novel construct of conspiracy mentality in our work , considering its high contemporary relevance in an increasingly digital world .

Overview of studies and predictions

Addressing the outlined research propositions, we present three studies (see Table 1 for a brief overview of main study characteristics). All materials, obtained data, and analysis codes for these three studies can be found in our project’s Open Science Framework repository (https://osf.io/3j67a/). Informed consent was collected from all participants before they took part in our research efforts.
Studies 1 and 2 mainly served to investigate the reliability and validity of our new attitudes towards AI scale, the ATTARI-12. Hence, these studies were mainly guided by several analytical steps examining the statistical properties of our measure, including its convergent validity, re-test reliability, and susceptibility to social desirability bias. In Study 3, we then set out to connect participants’ attitude towards AI (as measured by the developed questionnaire) to fundamental personality traits and demographic factors. For this third and final study, we will detail all theoretical considerations and hypotheses in the following.
Study 1 Study 2 Study 3
Sample type MTurk (US) Student sample MTurk (US)
Study focus Scale development: Factorial validity, reliability (internal consistency), construct validity Scale development: Reliability (re-test reliability), construct validity Personality predictors (Big Five, Dark Triad, conspiracy mentality)
Study language English German English
Final sample size 490 166 (at Time 1), 163 at Time 2, 150 at both times 298
Measurements 1 2 (four to five week follow-up) 1
Table 1. Overview of the conducted studies.
Faced with a lack of previous findings regarding personality predictors of AI-related attitudes, we deemed it important to anchor our investigation in a broader view at human personality. This perspective led us to first include the Big Five model , which has remained the most widely used taxonomy of human personality for several decades. As the name suggests, the Big Five consist of five fundamental personality dimensions-openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism-which are presumed to cover a substantial portion of the dispositional variance between individuals . Adhering to this notion, we developed a first set of hypotheses about the individual contribution of each Big Five trait to people’s AI-related attitudes.
For openness to experience-which can be defined as the tendency to be adventurous, intellectually curious, and imaginative-we expected a clear positive relationship to AI-related attitudes. Since AI technologies promise many new possibilities for human society, people who are open to experience should feel more excited and curious about the respective innovations (this is mirrored by recent findings on self-driving cars ). We hypothesize:
H1 The higher a person’s openness for experience, the more positive attitudes they hold towards AI.
Conscientiousness can be described as the inclination to be diligent, efficient, and careful, and to act in a disciplined or even perfectionist manner. Considering that it may be difficult for humans to understand the inner workings of an AI system or to anticipate its behavior, it seems likely that conscientious people would express more negative views about AI, a technological concept that might make the world less comprehensible for them .
H2 The higher a person’s conscientiousness, the more negative attitudes they hold towards AI.
The third of the Big Five traits, extraversion, encompasses the tendency to be out-going, talkative, and gregarious; in turn, it is negatively related to apprehensive and restrained behavior. Based on this definition, it comes as no surprise that extroverted individuals reported fewer concerns about autonomous technologies in previous studies . Thus, we hypothesize:
H3 The higher people’s extraversion, the more positive attitudes they hold towards AI.
High agreeableness manifests itself in the proclivity to show warm, cooperative, and kind-hearted behavior. Regarding the topic at hand, previous research suggests that agreeableness might be (moderately) correlated with positive views about automation or specific types of . In line with this, we assume:
H4 The higher a person’s agreeableness, the more positive attitudes they hold towards AI.
The last Big Five trait, neuroticism, is connected to self-conscious and shy behavior. People with high scores in this trait tend to be more vulnerable to external stressors and have trouble controlling spontaneous impulses. MacDorman and Entezari discovered that neurotic individuals reported significantly stronger eeriness after they had encountered an autonomous android than those with lower scores in the trait. Similar findings emerged with regards to people’s views on self-driving cars and narrow AI applications . Based on this evidence as well as the generally anxious nature of neurotic individuals, our hypothesis is as follows:
H5 The higher a person’s neuroticism, the more negative attitudes they hold towards AI.
Although the Big Five are considered a relatively comprehensive set of human personality traits, research has yielded several other concepts that serve to make sense of interpersonal differences (and how they relate to attitudes and behaviors). Importantly, tapping into more negatively connotated aspects of human nature, Paulhus and Williams introduced the Dark Triad of personality, a taxonomy that gathers three malevolent character traits: Machiavellianism, psychopathy, and narcissism. Due to their antisocial qualities, the Dark Triad have been frequently connected to deviant behaviors, interpersonal problems, and increased difficulties in the workplace . Moreover, they have become an important part of exploring interpersonal differences in attitude formation, not least including views on modern-day technology . In consequence, our second set of hypotheses revolves around the potential influence of these rather vindictive personality traits.
Machiavellianism is a personality dimension that encompasses manipulative, callous, and amoral qualities. For our research topic, we contemplated two opposing notions as to how this trait could relate to AI attitudes. On the one hand, recent literature suggests that Machiavellian individuals might feel less concerned about amoral uses of AI and instead focus on its utilitarian value, which may reflect in more positive attitudes . On the other hand, the prospect of AI-driven surveillance remains a prominent topic in the public discourse -and we deemed it likely that people high in Machiavellianism would be wary of this, wanting for their more deviant actions to remain undetected. Weighing both arguments, we ultimately settled for the latter and hypothesized:
H6 The higher a person’s Machiavellianism, the more negative attitudes they hold towards AI.
A person scoring high on psychopathy is likely prone to thrill-seeking behavior and may experience only little anxiety. Unlike the deliberate manipulations typical for Machiavellian people, the psychopathic personality trait involves much more impulsive antisocial tendencies. Based on this, we came to assume a positive relationship between psychopathy and attitudes towards AI—not least considering that our dependent variable would also involve people’s emotional reactions towards AI, which should turn out less fearful among those high in psychopathy. We assumed:
H7 The higher a person’s psychopathy, the more positive attitudes they hold towards AI.
Broadly speaking, narcissists tend to show egocentric, proud, and unempathetic behavior, which is often accompanied by feelings of grandeur and entitlement. Whereas some qualitative research suggests that intelligent computers might deal a severe “blow to our narcissism” ( , p. 145), certain possibilities offered by AI could also appear quite attractive to people who score high in this trait-such as the futuristic notion of inserting aspects of oneself into a computer algorithm, to be preserved for all eternity. Surprisingly enough, inquiring participants about this very idea did not reveal a significant relation to narcissism in a recent study . Then again, it should be noted that highly narcissistic individuals also tend to show open and extroverted behavior , which would suggest a more positive relationship with AI-related attitudes depending on our other hypotheses. In summary, we settled on the following, cautiously positive assumption:
H8 The higher a person’s narcissism, the more positive attitudes they hold towards AI.
Lastly, we proceeded to the exploration of a dispositional variable on a notably smaller scale of abstraction, which appeared promising to us from a contemporary perspective: people’s conspiracy mentality. Strongly related to general distrust in people, institutions, and whole political systems, this personality trait focuses on the susceptibility to conspiracy theories, i.e., explanations about famous events that defy common sense or publicly presented facts . Psychological research has shown that people who believe in one conspiracy theory are typically more likely to also believe in another, indicating a stable trait-like tendency to subscribe to an overly skeptical and distrusting way of thinking . Further emphasizing this angle, recent research has suggested that the tendency to accept epistemically suspect information might constitute a single dimension that further connects to difficulties in analytic thinking, an overestimation of one’s own knowledge, as well as receptivity for ‘pseudo-profound bullshit .
Although they might not be quite as well-known as other prominent conspiracy theories (e.g., concerning the attacks), there are in fact several conspiracist beliefs about the use of sophisticated technology and AI. The respective theories range from the idea that those in power want to replace certain individuals with intelligent machines to the belief that COVID-19 vaccinations are actually injections of sophisticated nanochips to establish a global surveillance system . While these notions might seem somewhat obscure to many, the triumph of social media as a news source has presented proponents of conspiracy theories with a powerful platform to disseminate their ideas . In addition to that, the well-documented (mis-)uses of big data-e.g., in the Cambridge Analytica scandal-has led many people to at least think about how computer algorithms might be used for sinister purposes . Taking into account all of these observations, we anticipated that conspiracy mentality relates to more negative attitudes towards AI:
H9 The higher a person’s level of conspiracy mentality, the more negative attitudes they hold towards AI.
Concluding the theoretical preparation of our study, we devised hypotheses on two basic demographic variables that were previously connected to AI-related attitudes in scientific publications. Specifically, it has been found that women typically hold more negative attitudes towards AI than men, which may, among other causes, be explained by societal barriers that limit women’s access to (and interest in) technical domains . Likewise, studies have hinted at the higher apprehensiveness towards AI-powered technology among the elderly, potentially due to a lack of understanding and accessibility . In the expectation to replicate these prior findings, we proposed:
H10 Women hold more negative attitudes towards AI than men.
H11 Older individuals hold more negative attitudes towards AI than younger individuals.

Study 1

The goal underlying Study 1 was to develop a psychometrically sound scale that was (a) one-dimensional, (b) incorporated items reflecting the three bases or facets of attitudes (cognitive, affective, behavioral) that have guided attitude research in psychology for the last decades , and (c) consisted of positively and negatively worded items to ensure that attitudes were measured on a full spectrum between aversion and enthusiasm (and agreement bias did not confound the results systematically). Based on theory and existing attitude scales in applied fields original items were generated. Item generation was guided by the goals of developing an equal number of negatively and positively worded items and an equal number of items representing the three attitude facets (cognitive, affective, behavioral). Adhering to prior AI theory and existing measurement approaches , the items were preceded by an instruction that introduced the phenomenon of artificial intelligence to participants, in order to reduce semantic ambiguities about the attitude target. This instruction is an integral part of the measurement.
In a second step, the authors discussed the items, and excluded items that were ambiguous in content, too specific in terms of attitude target, or involved too much linguistic overlap with other items. The remaining scale consisted of twelve items, with each of the three psychological facets of human attitudes-cognitive, affective, and behavioral-being represented by two positively and two negatively worded items. Nevertheless, all twelve items were expected to represent one general factor: People’s attitude towards AI. The full scale (ATTARI-12) can be found in the Appendix.
To gain insight into the validity of our new scale, Study 1 further included items on more specific AI applications. We expected that general attitudes towards AI as measured by the ATTARI-12 would be positively associated with specific attitudes towards electronic personal assistants (e.g., Alexa) and attitudes towards robots. Moreover, we assessed participants’ tendency to give socially desirable answers to establish whether our novel instrument would be affected by social desirability bias. Based on our careful wording during the construction of the scale, we expected that people’s attitude towards AI (per the ATTARI-12) would not be substantially related to social desirability. The hypotheses and planned data analysis steps for Study 1 were preregistered at https:// aspredicted.org/8B5_GHZ (see also Supplement S1).

Method

Ethics statement

In Germany, institutional ethics approval is not required for psychological research as long as it does not involve issues regulated by law . All reported studies (Study 1, Study 2, and Study 3) were conducted in full accordance with the Declaration of Helsinki, as well as the ethical guidelines provided by the German Psychological Society (DGPs ). Of course, this also included obtaining informed consent from all participants before they were able to take part in this study.

Participants and procedure

Power analyses with semPower (Version 2.0.1 ) suggested a necessary sample size of at least 500 respondents for the planned study. Therefore, at least 600 participants were aspired in order to have a proper data basis for our analyses-while still retaining a buffer for potential exclusions. As such, a total of 601 participants were recruited from the US-American MTurk participant pool. To ensure satisfactory data quality, we set our MTurk request to at least 500 previously completed tasks (also known as HITs), as well as > 98% HIT approval rate. Each participant was rewarded for their participation, which lasted around three to five minutes.
Following preregistered exclusion criteria (i.e., completion time, failing at least one of two attention checks), 111 participants were excluded from our analyses. More specifically, 31 participants completed the questionnaire in less than participants did not describe the study as instructed, and one participant reported a year of birth that differed from the reported age by more than three years. Thus, the final sample consisted of 490 participants ( 212 female, 273 male, 5 other or no answer). The participants were between 19 and 72 years old ( years, ). For additional demographic information, please consult Supplement S2. After participants had given their informed consent, they completed a first attention check before proceeding to the ATTARI-12 questionnaire. Following this, we assessed their attitudes towards electronic personal assistants and robots, before presenting a measure of socially desirable answering. Lastly, participants had to summarize the study’s topic as an additional attention check and respond to several sociodemographic questions.

Measures

Attitudes towards artificial intelligence. We applied the newly created ATTARI-12, using a five-point answer format to capture participants’ responses ( strongly disagree, strongly agree ). Descriptive measures and the reliability of the scale are reported in the Results section.
Attitudes towards personal voice assistants (PVA). Participants indicated their attitudes towards PVAs on three semantic differential items (with five gradation points) that were created for the purpose of this study (e.g., “hate it-love it”; Cronbach’s ; see Supplement S3).
Attitudes towards robots. People’s general assessment of robots was measured with three items that were previously used in robot acceptance research . The items were answered on a four-point scale (Cronbach’s , ; e.g., “Robots are a good thing for societies because they help people”, with totally disagree to totally agree, see Supplement S3).
Social desirability. Participants’ tendency to give socially desirable answers was measured with the Social Desirability Scale . In this measure, participants have to answer whether 16 socially desirable or undesirable actions match their own behavior on a dichotomous scale (“true” or “false”). The number of socially desirable responses is added up for each participant, resulting in a range between 0 and 16 (Cronbach’s , ).

Results

Our analysis of factorial validity was guided by the assumption that the items represent a single construct: People’s attitude towards AI. To this end, we compared different, increasingly less restrictive, models in a confirmatory factor analysis: The first model (a) specified a single factor and, thus, assumed that individual differences in item responses could be explained by a single, general attitude construct. However, this assumption is often too strong in practice because specific content facets or item wording might lead to minor multidimensionality. Therefore, Model (b) estimated a bifactor S-1 structure that, in addition to the general factor, included two orthogonal specific factors for the cognitive or affective items. These specific factors captured the unique variance resulting from the two content domains that were not accounted by the general attitude factor. Following Eid and colleagues , no specific factor was specified for the behavioral items which, thus, acted as reference domain. Because the ATTARI-12 included positively and negatively worded items, Model (c) explored potential wording effects by evaluating a bifactor S-1 model that included an orthogonal specific factor for the negatively worded items. Finally, we combined Models (b) and (c) to study the joint effects of content and method effects.
The goodness of fit of these models are summarized in Table 2. Using established recommendations for the interpretation of these fit indices , the single factor model exhibited an inferior fit to the data. Also, acknowledging the different content facets (cognitive, affective, behavioral) did not improve model fit. However, the ATTARI-12 exhibited non-negligible wording effects as demonstrated by Model (c). This model showed a good fit to the data with a comparative fit index (CFI) of 0.98 , a root mean squared error (RMSEA) of 0.03 , and a standardized root mean residual (SRMR) of 0.03 . Despite the observed multidimensionality, all items showed good to excellent associations with the general factor as evidenced by rather high factor loadings falling between 0.48 and 0.88 (see Table 3). To examine the consequence of the multidimensional measurement structure in more detail, we examined how much of the systematic variance in the ATTARI-12 was explained by the general
CFI RMSEA SRMR AIC BIC Comp
(a) Single factor model 434.77 54 0.85 0.15 0.08 14,564 14,664
Bifactor S-1 models with one global factor and orthogonal specific factors for …
(b) Content facets 327.39 46 0.89 0.14 0.07 14,412 1456 (a) 102.95 8 <0.001
(c) Item wording 109.64 48 0.98 0.06 0.03 14,086 14,212 (a) 268.35 6 <0.001
(d) Content facets and item wording 93.91 40 0.98 0.06 0.03 14,076 14,236 (b) 191.45 6 <0.001
(c) 16.07 8 0.041
Table 2. Goodness of fit for competing confirmatory factor models for the ATTARI-12 (Study 1; US-American MTurk Panelists). Robust test statistic , degrees of freedom, CFI comparative fit index, RMSEA root mean squared error of approximation, SRMR standardized root mean residual, AIC Akaike’s information criterion, BIC Bayesian information criterion, Comp. Comparison model, Scaled chi square difference test statistic .
Item Loading on general factor Loading on specific factor Residual variance
1 0.79 (0.04)/0.84 0.26
2 0.66 (0.05) /0.59 0.61 (0.06) /0.54 0.46
3 0.81 (0.04) /0.77 0.45
4 0.66 (0.05) /0.57 0.54 (0.06) /0.48 0.58
5 0.91 (0.04) /0.87 0.27
6 0.64 (0.05) /0.65 0.55
7 0.77 (0.05) /0.64 0.56 (0.06) /0.46 0.56
8 0.55 (0.05) /0.48 0.52 (0.06) /0.45 0.75
9 0.71 (0.05) / 0.65 0.69
10 0.67 (0.05) / 0.59 0.70 (0.06) / 0.61 0.37
11 0.95 (0.04) / 0.88 0.27
12 0.84 (0.05) / 0.69 0.59 (0.06) / 0.48 0.43
Table 3. Factor loading pattern for the ATTARI-12 (Study 1; US-American MTurk panelists). Presented are unstandardized factor loadings (with standard errors in parenthesis)/standardized factor loadings and residual variances. The latent factors were identified by constraining the latent factor variances to 1 .
factor or the specific wording factor (cf. ). While the wording factor explained about of the variance, most of the common variance ( ) was reflected by the general factor. As a result, the hierarchical omega reliability for the general factor was but only for the specific factor. These results emphasize that all twelve items comprised an essentially unidimensional scale operationalizing a strong general factor reflecting the overarching attitude towards AI. Based on the analyses, no items needed to be excluded from the scale due to subpar psychometric properties.
Descriptive and correlational results on the composite measure of the ATTARI-12 are shown in Table 4. The reliability in terms of internal consistency was excellent, Cronbach’s . The distribution approximated a normal distribution, skewness ( ) and kurtosis ( ) were within the boundaries expected for psychometrically tested scales (also see Fig. 1). We acknowledge, however, that the distribution was somewhat leftskewed. As expected, the measure was positively correlated with attitudes towards electronic personal assistants ( ) and with attitudes towards robots ( ). ATTARI-12 scores were not substantially related to social desirability bias ( ).
In sum, the one-dimensional ATTARI-12 measured attitudes towards AI in a reliable way, and the indicators of convergent (attitudes towards specific AI applications) and divergent (social desirability) construct validity corroborate the validity of the scale.
1 2 3 4 5
Cronbach’s M Skew Kurt
1 ATTARI-12 0.93 3.66 0.83 -0.63 0.22
2 Social Desirability 0.85 8.74 4.24 -0.25 -0.87 0.037
3 Attitude Towards PVAs 0.94 3.72 1.09 -0.81 0.02 0.599*** 0.125**
4 Attitude Towards Robots 0.77 3.24 0.57 -0.75 0.81 0.676*** -0.021 0.435***
5 Age 39.78 11.06 0.81 -0.08 – 0.039 0.090* -0.023 -0.036
6 Gender -0.087 0.042 0.099* -0.058
Table 4. Descriptive statistics and zero-order correlations (Study 1; US-American MTurk panelists). for analyses that involve gender, for all other analyses. . Gender was dummy-coded ( male; female). PVAs = Personal Voice Assistants.
Figure 1. Distributions of the ATTARI-12 Results in Study 1 and Study 2 (at Time 1 and Time 2).

Study 2

The aim of the second study was to develop a German-language version of the scale and to gain further insight into the scale’s reliability and validity. In particular, we examined the re-test reliability of the scale, by administering the items at two points in time. We further assessed the extent to which the participants (undergraduates) wished to work with (or without) AI in their future careers. We expected a positive association between the latter variable and attitudes towards AI.

Method

This study consisted of two parts, administered online with an average delay of 31.40 days ( ; range 26-36 days). Students of a social science major at the University of Würzburg, Germany participated for course credit. According to calculations with Power software-based on an expected small to moderate correlation of power, and a minimum sample size of 123 participants was required. Yet, as we expected dropouts from Time 1 to Time 2, a starting sample of 180 participants was targeted. Eventually, 166 participants completed the survey at Time 1, and a total of 150 students provided data at both points in time. This final sample of 150 participants with complete data had an average age of 21.21 years ( ; range: 18 to 41 years) and consisted predominantly of women ( 113 female, 36 male, 1 other). Nearly all of the participants were German native speakers (see Supplement S4 for detailed descriptive data and Supplement S8 for the German-language version of the scale). Importantly, since participants for Study 2 were recruited among German students (in contrast to the use of English-language survey panel members in Study 1), we are confident that our two validation studies were based on entirely different samples.
At time 1, participants answered the ATTARI-12, as well as four items that measured their interest in a career involving AI technology (two items were reverse-coded, e.g., “I would prefer a position in which AI plays no role”). The items went with a 5 -point scale ( strongly disagree, strongly agree), Cronbach’s (see Supplement S5). In the resulting index, higher scores indicated higher aspirations for AI-related careers. At the study’s second measurement time, the ATTARI-12 had to be answered once again. Furthermore, the study was initially designed to also include a measure of social desirability at this point ( , in its German translation); however, the respective assessment did not achieve satisfactory reliability (Cronbach’s ) and was, therefore, excluded from our analysis.

Results

The main results are shown in Table 5. The reliability in terms of internal consistency was very good for the German-language version applied in this study, T1 Cronbach’s ; T2 Cronbach’s . The distribution approximated a normal distribution, skewness (skewness ; skewness ) and kurtosis (kurtosis ; kurtosis ) were within the boundaries expected for psychometrically sound scales (Fig. 1). Importantly, the test-re-test association was large, , supporting the reliability of the scale. Associations with the AI-career measure were significant and ranged around . In sum, the one-dimensional ATTARI-12 (German version) showed good psychometric properties.

Study 3

Having found empirical support for the validity of our newly created measure, we proceeded with our second main research goal: Exploring potential associations between personality traits and attitudes towards AI. To ensure transparency, we decided to preregister Study 3 before data collection, including all hypotheses and planned analyses (https://aspredicted.org/VRU_BJI).

Method

Participants

An a priori calculation of minimum sample size by means of Power software (assuming a small to moderate effect of power, , and eleven predictors in a hierarchical linear regression) resulted in a lower threshold of 221 participants. Since we intended to screen our sample for several data quality indicators and, thus, desired some leeway for potential exclusions, we recruited 353 US-American participants via the Amazon MTurk participant pool (age: years, ; 112 female, 239 male, 2 other; payment ). Specifically, we used the following MTurk criteria to ensure high data quality : (a) at least 100 approved HITs; (b) HIT approval rate . Moreover, the web service “IPhub.info” was used to check whether users’ individual IP address correctly indicated the United States as a current location.
ATTARI-12 T1 ATTARI-12 T2
Cronbach’s M SD Skew Kurt
ATTARI-12 T1 0.90 3.51 0.64 – 0.58 0.07
ATTARI-12 T2 0.89 3.49 0.63 -0.48 -0.24 0.804
Career-AI T1 0.85 3.11 0.92 -0.08 -0.78 0.635 0.563
Table 5. Descriptives and zero-order correlations (Study 2; German University Students). for both variables assessed at Time 1 (T1). for ATTARI-12 assessed at Time 2 (T2). for the correlations between T1- and T2-variables. All correlations are significant at (two-tailed). The value printed in bold font represents the focal re-test reliability.
Examining the collected data, several measures were taken to exclude MTurk workers whose answers indicated careless and inattentive responding. First, we asked all participants to indicate their age and year of birth on two separate pages of the survey; if these answers deviated more than two years from each other, the participant was excluded ( ). Second, as a specific check against bot workers, participants were asked to name a type of vegetable (“eggplant”) that was depicted on a large-scale photograph. Apart from minor typos, all answers that did not resemble the correct answer in English language led to the removal of the data . Third, participants were asked to choose the correct study topic (“artificial intelligence”) in a multiple-choice item, a task that was not fulfilled correctly by another participants.
In addition, we initially considered removing all participants who had filled in the survey in less than four minutes, a threshold we had measured as the minimum time for attentive responding. Looking at our obtained data, this would have led to the additional exclusion of 81 participants. However, as we noticed that a lot of MTurk workers had finished the questionnaire in slightly less than four minutes, we decided to ease our initial exclusion criterion to a minimum duration of three minutes-leading to the exclusion of participants. To make sure that this deviation from our preregistered analysis plan would not substantially change our results, we repeated all planned analyses with the initial four-minute criterion. Doing so, we found no noteworthy statistical differences (the results of this additional analysis are presented in Supplement S6).
Thus, in summary, our final sample consisted of 298 participants with a mean age of 39.29 years ( ; range from 22 to 73 years). Gender balance was slightly skewed towards men ( 102 female, 195 male, 1 other). Regarding ethnicity, most participants identified themselves as White ( ), followed by Asian ( ), Black ( ), and Hispanic ( ). Level of education was relatively high, with most participants having obtained either a bachelor’s degree ( ), a master’s degree ( ) or a Ph.D. ( ). Lastly, we note that our sample was quite balanced in terms of political orientation. For a complete overview of the obtained descriptive data, please consult Supplement S7.

Measures

All items of the following measures were presented on 5 -point scales ( strongly disagree; strongly agree).
Attitudes towards AI. To measure the main outcome variable, we administered the ATTARI-12 scale in its English version. Our data analysis showed that the scale again yielded excellent internal consistency this time around (Cronbach’s ).
Big Five. The Big Five were assessed with the Big Five Inventory , which consists of 44 items (openness: ten items, e.g., “I am a someone who is curious about many different things”; conscientiousness: nine items, e.g., “… does a thorough job”; extraversion: 8 items, e.g., “…is talkative”; agreeableness: nine items, e.g., “…is considerate and kind to almost everyone”; neuroticism: 8 items, e.g., “…can be tense”). We observed very good to excellent internal consistency for all five measured dimensions (openness: Cronbach’s ; conscientiousness: Cronbach’s ; extraversion: Cronbach’s ; agreeableness: Cronbach’s ; neuroticism: Cronbach’s ).

Dark Triad

We assessed participants’ Dark Triad personality traits with the Short Dark Triad scale (SD3 ). This instrument includes nine items on Machiavellianism (e.g., “It’s not wise to tell your secrets”), six items on psychopathy (e.g., “People often say I’m out of control.”), and nine items on narcissism (e.g., “Many group activities tend to be dull without me.”). Reliability analyses suggested good to very good internal consistencies for all three scales (Machiavellianism: Cronbach’s ; psychopathy: Cronbach’s ; narcissism: Cronbach’s ).

Conspiracy mentality

The Conspiracy Mentality Questionnaire ( ) offers a measure of people’s general susceptibility to conspiracy theories and beliefs. It consists of five items that encapsulate an inherent skepticism about the workings of society, governments, and secret organizations (e.g., “I think that many important things happen in the world, which the public is never informed about.”; “I think that events which superficially seem to lack a connection are often the result of secret activities.”). With our data, we observed a very good Cronbach’s of 0.86 for the CMQ.

Results

Table 6 summarizes descriptive information for and zero-order correlations between all study variables. On average, participants’ AI-related attitudes were slightly positive, considering that a value of 3 indicates the neutral midpoint of our 5 -point scale, which was assembled from equal numbers of negative (inversed) and positive items.
With our main data analysis slated to involve multiple regression, we first made sure that all necessary assumptions were met : Residuals were independent and normally distributed, and neither multicollinearity nor heteroskedascity issues could be found. Moreover, satisfying Cook’s distance values revealed that no influential cases were biasing our model. As such, we proceeded with hierarchical linear regression as the core procedure of our data analysis. Using participants’ ATTARI score as the criterion, we first entered their age and gender as predictors (Step 1), before adding the Big Five (Step 2), the Dark Triad (Step 3), and conspiracy mentality (Step 4) to an extended regression model. Table 7 presents the main calculations and coefficients of this hierarchical regression analysis. As can be seen here, the first step of the regression resulted in an insignificant equation, , 294) , with an of 0.02 . In contrast to this, the second step of the procedure (including the Big
Variable M SD 1 2 3 4 5 6 7 8 9 10 11 12
1 Age 39.29 11.08
2 Gender -0.23***
3 ATTARI-12 3.60 0.81 -0.12* 0.04
4 Openness to Experience 3.56 0.74 0.02 0.01 0.14*
5 Conscientiousness 3.97 0.72 0.25*** – 0.12* 0.08
6 Extraversion 2.77 1.00 0.04 0.07 0.04 0.27*** 0.25***
7 Agreeableness 3.77 0.73 -0.15** 0.22*** 0.16**
8 Neuroticism 2.64 1.01 -0.13* -0.10 -0.07 -0.14* -0.60*** -0.44 -0.53***
9 Machiavellianism 2.88 0.81 -0.19** 0.17** -0.11 -0.05 -0.25*** 0.07 -0.50*** 0.25***
10 Psychopathy 2.19 0.77 -0.25*** -0.12* -0.02 -0.36*** -0.50*** 0.14*
11 Narcissism 2.50 0.74 -0.15** -0.06 0.23** -0.03 -0.18** -0.16** 0.62***
12 Conspiracy Mentality 3.36 0.94 -0.08 -0.06 0.06 -0.07 0.07 0.15*
Table 6. Descriptive statistics and correlations for study variables (Study 3; US-American MTurk Panelists).
Note. Gender coded with ” 0 ” female, ” 1 ” male.
Step 1 Step 2 Step 3 Step 4
Age – 0.12 -1.96 – 0.15* -2.52 -0.16** – 2.68 – 0.17** -2.86
Gender 0.01 0.24 0.06 0.95 0.07 1.15 0.04 0.68
Openness to experience 0.09 1.53 0.10 1.67 0.11 1.85
Conscientiousness – 0.01 0.09 -0.03 -0.35 -0.01 -0.15
Extraversion 0.01 0.08 0.08 0.93 0.07 0.85
Agreeableness 3.95 0.25** 3.12 3.34
Neuroticism 0.08 1.01 0.07 0.81 0.10 1.24
Machiavellianism 0.04 0.48 0.09 1.08
Psychopathy -0.06 -0.64 -0.05 -0.49
Narcissism -0.09 – 1.02 -0.05 -0.60
Conspiracy mentality -0.22*** -3.68
0.02 (0.08***) (0.01) (0.04***)
Table 7. Hierarchical regression predicting participants’ attitudes towards AI (Study 3). , . Gender coded with ” 0 ” female, ” 1 ” male.
Five) yielded a significant result, . Proceeding with Step 3, we found that adding the Dark Triad traits did not lead to a significant increase of . Entering participants’ conspiracy mentality as a final predictor, however, Step 4 resulted in a solution with significantly higher explained variance, . In the following, we will take a closer look at the predictive value of each entered predictor.

Predictive value of the Big Five

In response to hypotheses H1 through H5, we first directed our attention to the Big Five personality dimensions. Based on the second regression step, which was designed to include these variables, it was found that only agreeableness significantly predicted participants’ attitudes towards AI ( ), with a positive beta coefficient of 0.29. This positive association also persisted after entering other personality traits in Steps 3 and 4 of the regression. Hence, we conclude that higher agreeableness was related to more positive views about AI technology in our sample-providing empirical support for Hypothesis 4 . Out of the other Big Five dimensions, openness to experience slightly missed the threshold of statistical significance in the final regression step, , so that further exploration of the respective hypothesis may be warranted before rejecting it entirely. The remaining Big Five traits conscientiousness (H2), extraversion (H3), and neuroticism (H5), however, remained clearly insignificant as predictors.

Predictive value of the Dark Triad

Proceeding to the dark personality traits Machiavellianism, psychopathy, and narcissism (as entered during Step 3 ), we note that none of the three predictors approached the conventional threshold of statistical significance. As such, we cannot accept hypotheses H6 through H8; according to our data, more malevolent character traits were not predictive of people’s general attitudes about AI.

Predictive value of conspiracy mentality

Concluding our investigation of dispositional influences, we focused on participants’ conspiracy mentality as it was added during the regression’s final step. We found that this variable emerged as another meaningful predictor, . In support of Hypothesis 9, we therefore report that a stronger conspiracy mentality was related to more negative views towards artificially intelligent technology.

Predictive value of age and gender

Lastly, the impact of participants’ age and gender on their AI attitudes was explored. To this end, we focused on our final regression model with all predictors entered into the equation. While the influence of gender remained insignificant, it was examined that higher age was significantly related to lower scores in the ATTARI-12 scale, i.e., to more aversive cognitions, feelings, and behavioral intentions towards AI ( ). In light of this, our data successfully replicated previous findings regarding this sociodemographic variable, resulting in a positive answer to H 11 . Meanwhile, we reject H 10 on the influence of gender.

General discussion

In a world shaped by autonomous technologies that are supposed to make human life safer, healthier, and more convenient, it is important to understand how people evaluate the very notion of artificially intelligent technology-and to identify factors that account for notable interindividual variance in this regard. Thus, the current project set out to explore the role of fundamental personality traits as predictors for people’s AI-related attitudes. In order to build our research upon a valid and reliable measurement of the outcome in question, we first developed a novel questionnaire-the ATTARI-12-and confirmed its psychometric quality across two studies. Distinguishing our instrument from extant alternatives, we note that our one-dimensional scale assesses attitudes towards AI on a full spectrum between aversion and enthusiasm. Furthermore, the ATTARI-12 items incorporate the classic trichotomy of human attitude (cognition, emotion, behavior), thus emerging as a conceptually sound way to measure people’s evaluation of AI. Lastly, since the instruction of the ATTARI-12 does not focus on specific use cases but rather on a broader understanding of AI as a set of technological abilities, we believe it may be suitable to be used across many different disciplines and research settings.
Utilizing our newly developed measure, we proceeded to our second research aim: Investigating potential connections between individuals’ AI-related attitudes and their personality traits. Specifically, we focused on two central taxonomies from the field of personality psychology (the Big Five, the Dark Triad), as well as conspiracy mentality as a trait of high contemporary relevance. By these means, we found significant effects for two of the explored dispositional predictors: Agreeableness (one of the Big Five traits) was significantly related to more positive attitudes about AI, whereas stronger conspiracy mentality predicted the opposite. We believe that both of these findings may offer intriguing implications for researchers, developers, and users of autonomous technology.
A personality dimension that typically goes along with an optimistic outlook at life , agreeableness has been found to predict people’s acceptance of innovations in many different domains . Psychologically speaking, this makes perfect sense: In line with their own benevolent nature, agreeable individuals tend to perceive others more positively as well-and may therefore come to think more about the opportunities than the risks of a new invention when forming their opinion. Since our specific measurement of participants’ attitudes requested them to consider both positive and negative aspects of AI, this tendency to focus more on the upside might have played an important role for the observed effect. Furthermore, we suppose that the trusting nature that often characterizes agreeable individuals offers another reasonable explanation for our result. Faced with a complex concept such as AI, which may appear obscure or downright incomprehensible to the layperson user, it arguably becomes all the more important for people to understand the intentions of the decision makers and stakeholders behind it. While the technological companies that develop AI systems might not always inspire the necessary confidence with their actions (e.g., excessive data collection, convoluted company policies), having a stronger inclination to trust others will likely compensate for this-presenting another reason as to why agreeableness might have emerged as a significant predictor in the current study.
A surprisingly similar argument may be put forward when addressing the second dispositional factor that achieved notable significance in our analyses, i.e., people’s conspiracy mentality. By definition, this personality trait is also anchored in perceptions of trust (or rather, a lack thereof), a notion that is echoed by our findings: The stronger participants expressed their skepticism about governments, organizations, and related news coverage (as indicated by items such as “I think many very important things happen in the world, which the public is never informed about”), the more negatively did their attitudes about AI turn out. In our reading, this further illustrates how transparency and perceived trustworthiness crucially affect the public acceptance of AI; just as having an agreeable, credulous character predicted more favorable attitudes, the disposition to suspect sinister forces on the global stage related to more negative views about autonomous technology.
On a societal level, we believe that our identification of conspiracy beliefs as a pitfall for AI-related attitudes is a particularly timely result, as it connects to the on-going debate about fake news in modern society. In the past few years, many people have started turning to social media as a source for news and education, which has paved the way for an unusual proliferation of misinformation . Virtual spaces, in which like-minded individuals mutually confirm their attitudes and suspicious about different issues (so-called echo chambers; e.g., discussion forums or text messaging groups) have become highly prevalent, and, by these means, turned into a cause for concern among media scholars. Indeed, research suggests that using platforms such as Facebook, Twitter, or YouTube not only provides an immensely effective way to disseminate postfactual beliefs but may even increase people’s susceptibility to conspiracy theories in the first place . Taken together with the fact that it is mostly complex and ambiguous topics for which people seek out postfactual information , negative views about intelligent
technology may proliferate most easily in the online realm. Arguably, the much-observed social media claim that COVID-19 vaccines secretly inject AI nanotechnology into unwilling patients illustrates this perfectly.
Based on the presented arguments, it stands to reason that the successful mass-adoption of intelligent technologies may also depend on whether postfactual theories about AI can be publicly refuted. In order to do so, it might be key to increase the public’s trust, not only in the concept of AI itself, but also in the organizations and systems employing it. For instance, developers may set out to educate users about the capabilities and limits of autonomous technologies in a comprehensible way, whereas companies might strive for more accessible policies and disclosures. At the same time, we note that overcoming deep-rooted fears and suspicions about AI will likely remain a great challenge in the future; in our expectation, the sheer sophistication and complexity of intelligent computers will continue to offer fertile ground for conspiracy theorists. Also, it cannot be ruled out that attempts to make AI more transparent in the future could also backfire, as conspiracy theorists might interpret these attempts as actual proof for a conspiracy in the first place. More so, the fact that popular movies and TV shows frequently present AI-based technology in a dangerous or creepy manner (e.g., in dystopian science fiction media ) may make it even more difficult to establish sufficient levels of trust-especially among those who tend to have a more skeptical mentality to begin with.
Apart from the interesting implications offered by the two significant personality predictors, several other examined traits showed no significant association with participants’ AI-related attitudes. Specifically, we found that three of the Big Five (conscientiousness, extraversion, and neuroticism) and all Dark Triad variables fell short in explaining notable variance in the obtained ATTARI-12 scores. Taken together with the abovementioned results, this indicates that user personality may have valuable yet limited potential to explain public acceptance of AI technology. In their nature as overarching meta-level traits, certain Big Five dimensions might simply be too abstract to address the nuances that characterize people’s experiences with-and attitudes towards-AI technology. For example, being a more extraverted person might involve aspects that both increase acceptance of AI (e.g., by being less apprehensive and restrained in general) and decrease it (e.g., by valuing genuine social connection to other humans, which may be challenged by new AI technology). Similarly, the more deviant personality traits included in the Dark Triad might be connected to both positive and negative views towards AI, thus preventing a significant prediction in a specific direction. Just as the many possibilities brought by new intelligent technology could be seen as useful tools to act manipulatively or enhance the self-suggesting a positive link to Machiavellianism and narcissism-they may also raise concerns among people scoring high in these traits, as AI might make it harder to act out devious impulses in an undetected or well-accepted manner.

Limitations and future directions

We would like to point out important limitations that need to be considered when interpreting the presented results of our project, especially the third and final study that served to address our main hypotheses. Although we recruited a relatively diverse sample in terms of age, gender, and socioeconomic background-and put several measures into place to ensure high data quality-our data on the association between AI attitudes and user personality are based on only one group of participants. Moreover, given that the sample of our third study was recruited via the same method as used in Study 1 (MTurk), we cannot rule out that certain participants might have taken part in both research efforts (even though, based on the very large participant pool of the chosen panel, we deem it highly unlikely that duplicate responses occurred). As such, replication efforts with different samples are encouraged in order to consolidate the yielded evidence. Not least, this concerns the recruitment of samples with a more balanced gender distribution, as our own recruitment yielded a notable majority of male participants. Also, considering that socio-economic and educational factors exert notable impact on people’s opinions towards technology (not least regarding AI; see ), shifting focus to different backgrounds should be most enlightening regarding the generalizability of our work. We believe that this might be especially important when focusing further on the dimension of conspiracy mentality-keeping in mind that a lack of education, media literacy, and analytic thinking ability have all been shown to predict the susceptibility to post-factual information .
Along these lines, we suggest that studies with samples from different cultures should be carried out in order to establish whether our findings are consistent across national borders. Most recently, a study focusing on a sample of South Korean participants also reported positive associations between attitudes towards AI and the Big Five dimension agreeableness, especially regarding the sociality and functionality of AI-powered systems . In contrast to our research, however, the authors did not pursue a one-dimensional measurement, thus yielding ambivalent findings (e.g., conscientiousness related to negative emotions towards AI but also to expectations of high functionality). Hence, we consider it worthwhile to apply the one-dimensional ATTARI-12 in other cultures in order to better understand people’s stance towards AI—and how this attitude is shaped by certain personality traits. Keeping in mind that our newly developed scale facilitated a reliable, one-dimensional measurement with samples in two countries, it appears as a promising tool for such examinations; even though, of course, further validation of our instrument is clearly welcome, especially for its non-English versions. Moreover, intercultural comparisons will have to ensure a consistent measurement of the personality traits in question. While both the Big Five and the Dark Triad have been highlighted for their cultural invariance , some studies have also raised doubts about this claim -so that the applicability of the used personality taxonomies might pose an additional challenge in this regard. A potential solution here would be to use the HEXACO inventory of personality , which combines both the Big Five and the Dark Triad’s deviant traits into a novel taxonomy of human personality. Not only has the HEXACO model received strong support in terms of cultural invariance , using it would also enable scholars to cross-validate the findings of the current study with another well-established instrument. Of course, completely different personality traits than the ones included in our endeavor-e.g., impulsivity, sensation
seeking, or fear of missing out-could also be helpful to gain a thorough understanding of interindividual differences regarding people’s attitudes towards AI.
Additionally, it should be noted that future research into this topic can clearly benefit from assessing situational and motivational aspects that affect participants’ views on AI beyond their personality characteristics. For instance, recent research suggests that views on sophisticated technology are strongly modulated by prior exposure to science fiction media , as well as philosophical views and moral norms . Furthermore, people have been found to change their attitudes after several encounters with AI technology , adding another factor to the equation. Taken together, this suggests that future studies might tap into several profound covariates to disentangle the states and traits affecting attitudes towards AI. In any case, keeping in mind that our newly developed ATTARI-12 scale facilitated a reliable, one-dimensional measurement with samples of different age ranges and cultural backgrounds, we suggest that it constitutes a promising methodological cornerstone for future examinations of AI attitudes.

Conclusion

In our project, we investigated attitudes towards AI as a composite measure of people’s thoughts, feelings, and behavioral intentions. Striving to complement previous research that focused more on the acceptance of specialized AI applications, we created and utilized a new unidimensional measure: the ATTARI-12. We are confident that the developed measure may now serve as a useful tool to practically address AI as a macro-level phenomenon, which-despite encompassing a host of different programs and applications-is united by several shared fundamentals. In our expectation, the created scale can still be administered even when new and unforeseen AI technologies emerge, as it focuses more on underlying principles than specific capabilities. For practitioners and developers, this perspective may be particularly insightful, suggesting that individuals’ responses to new AI technology (e.g., as customers or employees) will not only be driven by the specific features of a certain technology, but also by a general attitude towards AI. Similarly, we hope that large-scale sociodemographic efforts (e.g., national opinion compasses) might be able to benefit from the provided measure.
Apart from that, we established significant connections between general AI attitudes and two selected personality traits. In our opinion, it is especially the observed association with participants’ conspiracy mentality that holds notable relevance in our increasingly complex world. If AI developers cannot find suitable ways to make their technology appear innocuous to observers (in particular to those who tend to seek out post-factual explanations), it might become increasingly challenging to establish innovations on a larger scale. However, we emphasize that negative attitudes and objections against AI technology are not necessarily unjustified; hence, when educating the public on AI technologies, people should be encouraged to reflect upon both potential risks and benefits. To reduce the risk for new conspiracy theories, we further encourage industry professionals to stay as transparent as possible when introducing their innovations to the public.
Lastly, we suggest that follow-up work is all but needed to elaborate upon our current contribution. While we remain convinced that measuring attitudes towards AI as a general set of technological abilities is meritorious, more tangible results could stem from including a theoretical dichotomy that has recently emerged in the field of human-machine interaction . More specifically, it might be worthwhile to distinguish between AI abilities that relate to agency (i.e., planning, thinking, and acting) and those that relate to experience (i.e., sensing and feeling). In turn, scholars may be able to find out whether different user traits also relate to different attitudes towards “acting” and “feeling” AI-a nuanced perspective that would still allow for broader implications across many different technological contexts.

Data availability

All materials, obtained data, and analysis codes for the three reported studies can be found in the project’s Open Science Framework repository (https://osf.io/3j67a/).

Appendix

Attitudes towards artificial intelligence Scale (ATTARI-12), English version

Instruction: In the following, we are interested in your attitudes towards artificial intelligence (AI). AI can execute tasks that typically require human intelligence. It enables machines to sense, act, learn, and adapt in an autonomous, human-like way. AI may be part of a computer or online platform-but it can also be encountered in various other hardware devices such as robots.
Item List:
Wording Facet Valence
1 AI will make this world a better place. Cognitive Positive
2 I have strong negative emotions about . Affective Negative (reversecoded)
3 I want to use technologies that rely on . Behavioral Positive
4 AI has more disadvantages than advantages. Cognitive Negative (reversecoded)
5 I look forward to future developments. Affective Positive
6 offers solutions to many world problems. Cognitive Positive
7 I prefer technologies that do not feature . Behavioral Negative (reversecoded)
Wording Facet Valence
8 I am afraid of . Affective Negative (reversecoded)
9 I would rather choose a technology with than one without it. Behavioral Positive
10 AI creates problems rather than solving them. Cognitive Negative (reversecoded)
11 When I think about , I have mostly positive feelings. Affective Positive
12 I would rather avoid technologies that are based on . Behavioral Negative (reversecoded)
Received: 3 August 2023; Accepted: 31 January 2024
Published online: 05 February 2024

References

  1. Grewal, D., Roggeveen, A. L. & Nordfält, J. The future of retailing. J. Retail. 93(1), 1-6. https://doi.org/10.1016/j.jretai.2016.12.008 (2017).
  2. Vanian, J. Artificial Intelligence will Obliterate These Jobs by 2030 (Fortune, 2019). https://fortune.com/2019/11/19/artificial-intel ligence-will-obliterate-these-jobs-by-2030/.
  3. Ivanov, S., Kuyumdzhiev, M. & Webster, C. Automation fears: Drivers and solutions. Technol. Soc. 63, 101431. https://doi.org/10. 1016/j.techsoc.2020.101431 (2020).
  4. Waytz, A. & Norton, M. I. Botsourcing and outsourcing: Robot, British, Chinese, and German workers are for thinking-not feeling—jobs. Emotion 14, 434-444. https://doi.org/10.1037/a0036054 (2014).
  5. Gherheş, V. Why are we afraid of artificial intelligence (AI)?. Eur. Rev. Appl. Sociol. 11(17), 6-15. https://doi.org/10.1515/eras-2018-0006 (2018).
  6. Liang, Y. & Lee, S. A. Fear of autonomous robots and artificial intelligence: Evidence from national representative data with probability sampling. Int. J. Soc. Robot. 9, 379-384. https://doi.org/10.1007/s12369-017-0401-3 (2017).
  7. Fast, E., & Horvitz, E. Long-term trends in the public perception of artificial intelligence. In Proceedings of the 31st AAAI Conference on Artificial Intelligence pp. 963-969 (AAAI Press, 2016).
  8. Thorp, H. H. ChatGPT is fun, but not an author. Science 379(6630), 313. https://doi.org/10.1126/science.adg7879 (2023).
  9. Lobera, J., Fernández Rodríguez, C. J. & Torres-Albero, C. Privacy, values and machines: Predicting opposition to artificial intelligence. Commun. Stud. 71(3), 448-465. https://doi.org/10.1080/10510974.2020.1736114 (2020).
  10. Laakasuo, M. et al. The dark path to eternal life: Machiavellianism predicts approval of mind upload technology. Pers. Individ. Differ. 177, 110731. https://doi.org/10.1016/j.paid.2021.110731 (2021).
  11. Qu, W., Sun, H. & Ge, Y. The effects of trait anxiety and the big five personality traits on self-driving car acceptance. Transportation. https://doi.org/10.1007/s11116-020-10143-7 (2020).
  12. Kaplan, J. Artificial Intelligence: What Everyone Needs to Know. (Oxford University Press, 2016).
  13. Schepman, A. & Rodway, P. Initial validation of the general attitudes towards Artificial Intelligence Scale. Comput. Hum. Behav. Rep. 1, 100014. https://doi.org/10.1016/j.chbr.2020.100014 (2020).
  14. Sindermann, C. et al. Assessing the attitude towards artificial intelligence: Introduction of a short measure in German, Chinese, and English language. Künstliche Intelligenz 35(1), 109-118. https://doi.org/10.1007/s13218-020-00689-0 (2020).
  15. Wang, Y. Y., & Wang, Y. S. Development and validation of an artificial intelligence anxiety scale: An initial application in predicting motivated learning behavior. Interact. Learn. Environ. https://doi.org/10.1080/10494820.2019.1674887 (2019).
  16. Kieslich, K., Lünich, M. & Marcinkowski, F. The Threats of Artificial Intelligence Scale (TAI): Development, measurement and test over three application domains. Int. J. Soc. Robot. https://doi.org/10.1007/s12369-020-00734-w (2021).
  17. Stein, J.-P., Liebold, B. & Ohler, P. Stay back, clever thing! Linking situational control and human uniqueness concerns to the aversion against autonomous technology. Comput. Hum. Behav. 95, 73-82. https://doi.org/10.1016/j.chb.2019.01.021 (2019).
  18. Rosenberg, M. J., & Hovland, C. I. Attitude Organization and Change: An Analysis of Consistency Among Attitude Components (Yale University Press, 1960).
  19. Zanna, M. P. & Rempel, J. K. Attitudes: A new look at an old concept. In Attitudes: Their Structure, Function, and Consequences (eds Fazio, R. H. & Petty, R. E.) 7-15 (Psychology Press, 2008).
  20. Stein, J.-P., Appel, M., Jost, A. & Ohler, P. Matter over mind? How the acceptance of digital entities depends on their appearance, mental prowess, and the interaction between both. Int. J. Hum.-Comput. Stud. 142, 102463. https://doi.org/10.1016/j.ijhcs.2020. 102463 (2020).
  21. McClure, P. K. “You’re fired”, says the robot. Soc. Sci. Comput. Rev. 36(2), 139-156. https://doi.org/10.1177/0894439317698637 (2017).
  22. Gillath, O. et al. Attachment and trust in artificial intelligence. Comput. Hum. Behav. 115, 106607. https://doi.org/10.1016/j.chb. 2020.106607 (2021).
  23. Young, K. L. & Carpenter, C. Does science fiction affect political fact? Yes and no: A survey experiment on “Killer Robots”. Int. Stud. Q. 62(3), 562-576. https://doi.org/10.1093/isq/sqy028 (2018).
  24. Gnambs, T. & Appel, M. Are robots becoming unpopular? Changes in attitudes towards autonomous robotic systems in Europe. Comput. Hum. Behav. 93, 53-61. https://doi.org/10.1016/j.chb.2018.11.045 (2019).
  25. Horstmann, A. C. & Krämer, N. C. Great expectations? Relation of previous experiences with social robots in real life or in the media and expectancies based on qualitative and quantitative assessment. Front. Psychol. 10, 939. https://doi.org/10.3389/fpsyg. 2019.00939 (2019).
  26. Sundar, S. S., Waddell, T. F. & Jung, E. H. The Hollywood robot syndrome: Media effects on older adults’ attitudes toward robots and adoption intentions. Paper presented at the 11th Annual ACM/IEEE International Conference on Human-Robot Interaction, Christchurch, New Zealand (2016).
  27. Chien, S.-Y., Sycara, K., Liu, J.-S. & Kumru, A. Relation between trust attitudes toward automation, Hofstede’s cultural dimensions, and Big Five personality traits. Proc. Hum. Factors Ergon. Soc. Annu. Meet. 60(1), 841-845. https://doi.org/10.1177/1541931213 601192 (2016).
  28. MacDorman, K. F. & Entezari, S. Individual differences predict sensitivity to the uncanny valley. Interact. Stud. 16(2), 141-172. https://doi.org/10.1075/is.16.2.01mac (2015).
  29. Oksanen, A., Savela, N., Latikka, R. & Koivula, A. Trust toward robots and artificial intelligence: An experimental approach to Human-Technology interactions online. Front. Psychol. 11, 3336. https://doi.org/10.3389/fpsyg.2020.568256 (2020).
  30. Wissing, B. G. & Reinhard, M.-A. Individual differences in risk perception of artificial intelligence. Swiss J. Psychol. 77(4), 149-157. https://doi.org/10.1024/1421-0185/a000214 (2018).
  31. Costa, P. T. & McCrae, R. R. Four ways five factors are basic. Pers. Individ. Differ. 13, 653-665. https://doi.org/10.1016/0191-8869(92)90236-I (1992).
  32. Paulhus, D. L. & Williams, K. M. The Dark Triad of personality: Narcissism, Machiavellianism, and psychopathy. J. Res. Pers. 36, 556-563. https://doi.org/10.1016/S0092-6566(02)00505-6 (2002).
  33. Bruder, M., Haffke, P., Neave, N., Nouripanah, N. & Imhoff, R. Measuring individual differences in generic beliefs in conspiracy theories across cultures: Conspiracy mentality questionnaire. Front. Psychol. 4, 225. https://doi.org/10.3389/fpsyg.2013.00225 (2013).
  34. Imhoff, R. et al. Conspiracy mentality and political orientation across 26 countries. Nat. Hum. Behav. 6(3), 392-403. https://doi. org/10.1038/s41562-021-01258-7 (2022).
  35. Nettle, D. & Penke, L. Personality: Bridging the literatures from human psychology and behavioural ecology. Philos. Trans. R. Soc. B 365(1560), 4043-4050. https://doi.org/10.1098/rstb.2010.0061 (2010).
  36. Poropat, A. E. A meta-analysis of the five-factor model of personality and academic performance. Psychol. Bull. 135(2), 322-338. https://doi.org/10.1037/a0014996 (2009).
  37. Kaplan, A. D., Sanders, T. & Hancock, P. A. The relationship between extroversion and the tendency to anthropomorphize robots: A bayesian analysis. Front. Robot. AI 5, 135. https://doi.org/10.3389/frobt.2018.00135 (2019).
  38. Kaufman, S. B., Yaden, D. B., Hyde, E. & Tsukayama, E. The light vs. Dark triad of personality: Contrasting two very different profiles of human nature. Front. Psychol. 10, 467. https://doi.org/10.3389/fpsyg.2019.00467 (2019).
  39. LeBreton, J. M., Shiverdecker, L. K. & Grimaldi, E. M. The Dark Triad and workplace behavior. Annu. Rev. Organ. Psychol. Organ. Behav. 5(1), 387-414. https://doi.org/10.1146/annurev-orgpsych-032117-104451 (2018).
  40. Feldstein, S. The Global Expansion of AI Surveillance. Carnegie Endowment for International Piece. https://carnegieendowment. org/files/WP-Feldstein-AISurveillance_final1.pdf (2019).
  41. Malabou, C. Morphing Intelligence: From IQ Measurement to Artificial Brains (Columbia University Press, 2019).
  42. Swami, V., Chamorro-Premuzic, T. & Furnham, A. Unanswered questions: A preliminary investigation of personality and individual difference predictors of 9/11 conspiracist beliefs. Appl. Cognit. Psychol. 24, 749-761. https://doi.org/10.1002/acp.1583 (2010).
  43. Pennycook, G. & Rand, D. G. Who falls for fake news? The roles of bullshit receptivity, overclaiming, familiarity, and analytic thinking. J. Pers. 88(2), 185-200. https://doi.org/10.1111/jopy. 12476 (2019).
  44. Shakarian, A. The Artificial Intelligence Conspiracy (2016).
  45. McEvoy, J. Microchips, Magnets and Shedding: Here are 5 (Debunked) Covid Vaccine Conspiracy Theories Spreading Online (Forbes, 2021). https://www.forbes.com/sites/jemimamcevoy/2021/06/03/microchips-and-shedding-here-are-5-debunked-covid-vacci ne-conspiracy-theories-spreading-online/
  46. De Vynck, G., & Lerman, R. Facebook and YouTube Spent a Year Fighting Covid Misinformation. It’s Still Spreading (The Washington Post, 2021). https://www.washingtonpost.com/technology/2021/07/22/facebook-youtube-vaccine-misinformation/.
  47. González, F., Yu, Y., Figueroa, A., López, C., & Aragon, C. Global reactions to the Cambridge Analytica scandal: A cross-language social media study. In Proceedings of the 2019 World Wide Web Conference 799-806. ACM Press. https://doi.org/10.1145/33085 60.3316456 (2019).
  48. Joyce, M. & Kirakowski, J. Measuring attitudes towards the internet: The general internet attitude scale. Int. J. Hum.-Comput. Interact. 31, 506-517. https://doi.org/10.1080/10447318.2015.1064657 (2015).
  49. Bartneck, C., Lütge, C., Wagner, A., & Welsh, S. An Introduction to Ethics in Robotics and AI (Springer, 2021).
  50. German Research Foundation. Statement by an Ethics Committee. https://www.dfg.de/en/research_funding/faq/faq_humanities_ social_science/index.html (2023).
  51. German Psychological Society. Berufsethische Richtlinien [Work Ethical Guidelines]. https://www.dgps.de/fileadmin/user_upload/ PDF/Berufsetische_Richtlinien/BER-Foederation-20230426-Web-1.pdf (2022).
  52. Jobst, L. J., Bader, M. & Moshagen, M. A tutorial on assessing statistical power and determining sample size for structural equation models. Psychol. Methods 28(1), 207-221. https://doi.org/10.1037/met0000423 (2023).
  53. Stöber, J. The Social Desirability Scale-17 (SDS-17): Convergent validity, discriminant validity, and relationship with age. Eur. J. Psychol. Assess. 17(3), 222-232. https://doi.org/10.1027/1015-5759.17.3.222 (2001).
  54. Eid, M., Geiser, C., Koch, T. & Heene, M. Anomalous results in G-factor models: Explanations and alternatives. Psychol. Methods 22, 541-562. https://doi.org/10.1037/met0000083 (2017).
  55. Schermelleh-Engel, K., Moosbrugger, H. & Müller, H. Evaluating the fit of structural equation models: Tests of significance and descriptive goodness-of-fit measures. Methods Psychol. Res. 8(2), 23-74 (2003).
  56. Rodriguez, A., Reise, S. P. & Haviland, M. G. Evaluating bifactor models: Calculating and interpreting statistical indices. Psychol. Methods 21(2), 137-150. https://doi.org/10.1037/met0000045 (2016).
  57. Kennedy, R. et al. The shape of and solutions to the MTurk quality crisis. Polit. Sci. Res. Methods 8(4), 614-629. https://doi.org/10. 1017/psrm. 2020.6 (2020).
  58. Peer, E., Vosgerau, J. & Acquisti, A. Reputation as a sufficient condition for data quality on Amazon Mechanical Turk. Behav. Res. Methods 46(4), 1023-1031. https://doi.org/10.3758/s13428-013-0434-y (2013).
  59. John, O. P., Naumann, L. P. & Soto, C. J. Paradigm shift to the integrative Big-Five trait taxonomy: History, measurement, and conceptual issues. In Handbook of Personality: Theory and Research (eds John, O. P. et al.) 114-158 (Guilford Press, 2008).
  60. Jones, D. N. & Paulhus, D. L. Introducing the Short Dark Triad (SD3): A brief measure of dark personality traits. Assessment 21(1), 28-41. https://doi.org/10.1177/1073191113514105 (2014).
  61. Field, A. Discovering Statistics Using IBM SPSS Statistics (SAGE, 2013).
  62. Baranski, E., Sweeny, K., Gardiner, G. & Funder, D. C. International optimism: Correlates and consequences of dispositional optimism across 61 countries. J. Pers. 89(2), 288-304. https://doi.org/10.1111/jopy. 12582 (2020).
  63. Sharpe, J. P., Martin, N. R. & Roth, K. A. Optimism and the Big Five factors of personality: Beyond neuroticism and extraversion. Pers. Individ. Differ. 51(8), 946-951. https://doi.org/10.1016/j.paid.2011.07.033 (2011).
  64. Steel, G. D., Rinne, T. & Fairweather, J. Personality, nations, and innovation. Cross-Cult. Res. 46(1), 3-30. https://doi.org/10.1177/ 1069397111409124 (2011).
  65. McCarthy, M. H., Wood, J. V. & Holmes, J. G. Dispositional pathways to trust: Self-esteem and agreeableness interact to predict trust and negative emotional disclosure. J. Pers. Soc. Psychol. 113(1), 95-116. https://doi.org/10.1037/pspi0000093 (2017).
  66. Lazer, D. M. J. et al. The science of fake news. Science 359(6380), 1094-1096. https://doi.org/10.1126/science.aao2998 (2018).
  67. Stecula, D. A. & Pickup, M. Social media, cognitive reflection, and conspiracy beliefs. Front. Polit. Sci. 3, 62. https://doi.org/10. 3389/fpos.2021.647957 (2021).
  68. van Prooijen, J. W. Why education predicts decreased belief in conspiracy theories. Appl. Cognit. Psychol. 31(1), 50-58. https:// doi.org/10.1002/acp. 3301 (2016).
  69. Scheibenzuber, C., Hofer, S. & Nistor, N. Designing for fake news literacy training: A problem-based undergraduate online-course. Comput. Hum. Behav. 121, 106796. https://doi.org/10.1016/j.chb.2021.106796 (2021).
  70. Sindermann, C., Schmitt, H. S., Rozgonjuk, D., Elhai, J. D. & Montag, C. The evaluation of fake and true news: On the role of intelligence, personality, interpersonal trust, ideological attitudes, and news consumption. Heliyon 7(3), e06503. https://doi.org/ 10.1016/j.heliyon.2021.e06503 (2021).
  71. Park, J. & Woo, S. E. Who likes artificial intelligence? Personality predictors of attitudes toward artificial intelligence. J. Psychol. 156(1), 68-94. https://doi.org/10.1080/00223980.2021.2012109 (2022).
  72. McCrae, R. R. & Costa, P. T. Jr. Personality trait structure as a human universal. Am. Psychol. 52(5), 509-516. https://doi.org/10. 1037/0003-066X.52.5.509 (1997).
  73. Rogoza, R. et al. Structure of Dark Triad Dirty Dozen across eight world regions. Assessment 28(4), 1125-1135. https://doi.org/ 10.1177/1073191120922611 (2020).
  74. Gurven, M., von Rueden, C., Massenkoff, M., Kaplan, H. & Lero Vie, M. How universal is the Big Five? Testing the five-factor model of personality variation among forager-farmers in the Bolivian Amazon. J. Pers. Soc. Psychol. 104(2), 354-370. https://doi. org/10.1037/a0030841 (2013).
  75. Lee, K. & Ashton, M. C. Psychometric properties of the HEXACO-100. Assessment 25, 543-556. https://doi.org/10.1177/10731 91116659134 (2018).
  76. Thielmann, I. et al. The HEXACO-100 across 16 languages: A Large-Scale test of measurement invariance. J. Pers. Assess. 102(5), 714-726. https://doi.org/10.1080/00223891.2019.1614011 (2019).
  77. Mara, M. & Appel, M. Science fiction reduces the eeriness of android robots: A field experiment. Comput. Hum. Behav. 48, 156-162. https://doi.org/10.1016/j.chb.2015.01.007 (2015).
  78. Laakasuo, M. et al. What makes people approve or condemn mind upload technology? Untangling the effects of sexual disgust, purity and science fiction familiarity. Palgrave Commun. https://doi.org/10.1057/s41599-018-0124-6 (2018).
  79. Latikka, R., Savela, N., Koivula, A. & Oksanen, A. Attitudes toward robots as equipment and coworkers and the impact of robot autonomy level. Int. J. Soc. Robot. 13(7), 1747-1759. https://doi.org/10.1007/s12369-020-00743-9 (2021).
  80. Appel, M., Izydorczyk, D., Weber, S., Mara, M. & Lischetzke, T. The uncanny of mind in a machine: Humanoid robots as tools, agents, and experiencers. Comput. Hum. Behav. 102, 274-286. https://doi.org/10.1016/j.chb.2019.07.031 (2020).
  81. Gray, K. & Wegner, D. M. Feeling robots and human zombies: Mind perception and the uncanny valley. Cognition 125, 125-130. https://doi.org/10.1016/j.cognition.2012.06.007 (2012).
  82. Grundke, A., Stein, J.-P. & Appel, M. Mind-reading machines: Distinct user responses to thought-detecting and emotion-detecting robots. Technol. Mind Behav. https://doi.org/10.17605/OSF.IO/U52KM (2021).
  83. Yuan, K. H. & Bentler, P. M. Three likelihood-based methods for mean and covariance structure analysis with non-normal missing data. Sociol. Methodol. 30(1), 165-200. https://doi.org/10.1111/0081-1750.00078 (2000).
  84. Satorra, A. & Bentler, P. M. Ensuring positiveness of the scaled difference chi-square test statistic. Psychometrika 75, 243-248. https://doi.org/10.1007/s11336-009-9135-y (2010).

Author contributions

J.-P.S.: Conceptualization, Methodology, Investigation, Formal Analysis, Data Curation, Writing (original draft), Writing (reviewing and editing).T.M.: Conceptualization, Methodology, Investigation, Data Curation, Writing (reviewing and editing).T.G.: Methodology, Investigation, Formal Analysis, Data Curation, Writing (reviewing and editing).F.H.: Conceptualization, Methodology, Investigation, Writing (reviewing and editing).M.A.: Conceptualization, Methodology, Formal Analysis, Data Curation, Writing (original draft), Writing (reviewing and editing), Supervision.

Funding

Open Access funding enabled and organized by Projekt DEAL.

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/s41598-024-53335-2.
Correspondence and requests for materials should be addressed to J.-P.S.
Reprints and permissions information is available at 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 licence, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence 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 licence, visit http://creativecommons.org/licenses/by/4.0/.
© The Author(s) 2024

  1. Department of Media Psychology, Institute for Media Research, Chemnitz University of Technology, Thüringer Weg 11, 09126 Chemnitz, Germany. Psychology of Communication and New Media, Human-Computer-Media Institute, University of Würzburg, Würzburg, Germany. Leibniz Institute for Educational Trajectories, Bamberg, Germany. email: jpstein@phil.tu-chemnitz.de