تصورات محترفي تصميم تجربة المستخدم حول الذكاء الاصطناعي التوليدي User Experience Design Professionals’ Perceptions of Generative Artificial Intelligence


DOI: https://doi.org/10.1145/3613904.3642114
تاريخ النشر: 2024-05-11

تصورات محترفي تصميم تجربة المستخدم حول الذكاء الاصطناعي التوليدي

جي لي، إي بي إيه إم، هولنداهانشينغ كاو، جامعة ستانفورد، الولايات المتحدةلورا لين، جوجل، الولايات المتحدةيويانغ هو، نوتيون لابز، الولايات المتحدةرويهاو زو، جامعة كورنيل، الولايات المتحدةعبد الله العلي، مركز الرياضيات والمعلوماتية، هولندا

الملخص

بين المحترفين المبدعين، أثار الذكاء الاصطناعي التوليدي (GenAI) حماسًا بشأن قدراته وخوفًا من العواقب غير المتوقعة. كيف يؤثر GenAI على ممارسة تصميم تجربة المستخدم (UXD)، وهل هذه المخاوف مبررة؟ قمنا بإجراء مقابلات مع 20 مصمم تجربة مستخدم، بتجارب متنوعة ومن شركات مختلفة (من الشركات الناشئة إلى المؤسسات الكبيرة). استفسرنا منهم لتوصيف ممارساتهم، وعينة من مواقفهم واهتماماتهم وتوقعاتهم. وجدنا أن المصممين ذوي الخبرة واثقون من أصالتهم وإبداعهم ومهاراتهم التعاطفية، ويعتبرون دور GenAI مساعدًا. وأكدوا على العوامل البشرية الفريدة مثل “الاستمتاع” و”الوكالة”، حيث يبقى البشر هم الحكام على “محاذاة الذكاء الاصطناعي”. ومع ذلك، يمكن أن تؤثر تدهور المهارات، واستبدال الوظائف، واستنفاد الإبداع سلبًا على المصممين المبتدئين. نناقش الآثار المترتبة على التعاون بين البشر وGenAI، وبشكل خاص حقوق الطبع والنشر والملكية، والإبداع البشري والوكالة، ومعرفة الذكاء الاصطناعي والوصول إليه. من خلال عدسة الذكاء الاصطناعي المسؤول والمشارك، نساهم في فهم أعمق لمخاوف GenAI والفرص المتاحة لتصميم تجربة المستخدم.

مفاهيم CCS: • الحوسبة المتمحورة حول الإنسان الدراسات التجريبية في تفاعل الإنسان مع الحاسوب.
الكلمات والعبارات الرئيسية الإضافية: الذكاء الاصطناعي التوليدي، مصممو تجربة المستخدم، الذكاء الاصطناعي المسؤول، تجربة المستخدم، التعاون بين الإنسان والذكاء الاصطناعي

تنسيق مرجع ACM:

جي لي، هانتشينغ كاو، لورا لين، يويانغ هو، رويهاو زو، وعبد الله العلي. 2024. تصورات محترفي تصميم تجربة المستخدم حول الذكاء الاصطناعي التوليدي. في وقائع مؤتمر CHI لعوامل الإنسان في أنظمة الحوسبة (CHI ’24)، 11-16 مايو 2024، هونولولو، هاواي، الولايات المتحدة الأمريكية. ACM، نيويورك، نيويورك، الولايات المتحدة الأمريكية، 27 صفحة.https://doi.org/10.1145/3613904.3642114

1 المقدمة

يتطور مشهد الذكاء الاصطناعي (AI) بسرعة حيث اكتسبت النماذج الأساسية القابلة للتكيف، المبنية على الشبكات العصبية العميقة والتعلم الذاتي، اعتمادًا واسع النطاق. تشمل هذه النماذج نماذج اللغة الكبيرة المعتمدة على المحولات (مثل GPT-3، BERT)، والنماذج البصرية (مثل DALL-E، فلورنس)، أو النماذج متعددة الوسائط (مثل UniLM). لقد تم استغلال قوة هذه النماذج مؤخرًا، كما يتضح من إطلاق أدوات الذكاء الاصطناعي التوليدي (GenAI)، مثل ChatGPT. دال-إي التشتت المستقر وميدجورني . لقد دفعت هذه النماذج بشكل كبير حدود قدرات الذكاء الاصطناعي، حيث إن أدوات الذكاء الاصطناعي التوليدي قادرة على معالجة كميات كبيرة من البيانات وتعلم الأنماط الإحصائية، مما يمكنها
لتوليد مخرجات متعددة الوسائط جديدة بناءً على مدخلات اللغة الطبيعية (أي، المطالبات). لقد أظهرت القدرة على إنتاج أنواع متنوعة من المحتوى، بما في ذلك النصوص [72]، الصور [82]، الفيديو [42]، والصوت [7]. على سبيل المثال، يُنظر إلى ChatGPT على أنه أفضل روبوت محادثة على الإطلاق [36] مع تسجيل أكثر من مليون شخص لاستخدامه في خمسة أيام فقط [79]. نشر المعجبون المتحمسون أمثلة على إنتاج ChatGPT لرموز الكمبيوتر، والمقالات، والقصائد؛ بينما يكتب آخرون مطالبات غير واقعية لـ Midjourney أو DALL-E 2، تتراوح من طلبات مثل “ارسم بورتريه لرئيسك وهو يداعب نمرًا بأسلوب رامبرانت”، إلى “صور تجمعًا لديناصورات في منتصف العمر تتناول القهوة وتتأمل معنى الحياة”، وستعيد هذه الأدوات تصويرًا دقيقًا بشكل مذهل في لحظة. من ناحية، أثار الاستخدام الواسع لأدوات GenAI حماسًا داخل المجتمع بسبب تقدمها الملحوظ؛ ومن ناحية أخرى، يؤدي ذلك إلى القلق بشأن العواقب غير المتوقعة التي قد تُستحث ضمنيًا وتكون خارج السيطرة البشرية [4]. علاوة على ذلك، تثار المخاوف بشأن التماثل، حيث يمكن أن تنتقل أي نقاط ضعف في هذه النماذج إلى التطبيقات اللاحقة [6].
تتمتع القدرات المتزايدة للذكاء الاصطناعي التوليدي (GenAI) بإمكانية إحداث ثورة في جوانب مختلفة من الحياة الحديثة، مما يؤثر بشكل كبير على الأفراد الذين يكسبون رزقهم من خلال إنشاء المحتوى، بما في ذلك المصممين، وكُتّاب النصوص، والصحفيين، والأساتذة المتفرغين. من بين هؤلاء المحترفين، هناك اعتراف متزايد بالأضرار المحتملة، مثل تلف السمعة، وفقدان الوظائف والاقتصاد، والانتحال، وانتهاك حقوق الطبع والنشر. في سياق الذكاء الاصطناعي التشاركي، من الأهمية بمكان الاعتراف بأن المحترفين المبدعين ومنشئي المحتوى، وليس فقط الخبراء الفنيين، يمتلكون معرفة وخبرة واهتمامات قيمة ضرورية للتطوير المسؤول للذكاء الاصطناعي. هناك العديد من الجهود جارية لاستكشاف تأثير الذكاء الاصطناعي على تجربة المستخدم (مثل تقرير الذكاء الاصطناعي في أبحاث تجربة المستخدم). كخطوة لفهم ومعالجة هذه المخاوف المتزايدة، نستكشف في هذه الورقة التأثير المحتمل لأدوات GenAI على ممارسة تصميم تجربة المستخدم، من خلال إشراك مصممي تجربة المستخدم، وهم محترفون مبدعون مهمون، في محادثاتنا. في تصميم تجربة المستخدم المعاصر، الهدف هو التوافق مع احتياجات المستخدمين وقيم الأعمال. على عكس المهن الإبداعية الأخرى، فإن تصميم تجربة المستخدم ليس عملاً فنياً بحتاً، بل يدور حول تلبية متطلبات المستخدمين النهائيين. وهذا يتطلب ليس فقط تصميم الشكل العام وإحساس المنتج الرقمي، ولكن أيضاً صياغة أنظمة تنقل فعالة، وتسميات، وتصنيف المحتوى لضمان أن المستخدمين يمكنهم بسهولة العثور على المعلومات التي يحتاجونها والوصول إليها. عملية تصميم تجربة المستخدم هي عملية تصميم تكرارية ومركزية حول المستخدم تتطلب من المصممين أن يكونوا مبدعين ومتعاونين، وأن يمتلكوا عقلية حل المشكلات وفهم الأمور، وأن يظهروا تعاطفاً تجاه المستخدمين. يجب أن تمر تصميم تجربة المستخدم الناجحة بعدة تكرارات وغالباً ما ترتبط بمدى قدرة المصممين على فهم وترجمة متطلبات المستخدمين إلى وظائف متCorresponding وجمالية مرضية. مع القدرة على أتمتة المهام التصميمية المتكررة، مثل إنشاء الإطارات السلكية، وتوليد تنويعات التخطيط، أو إنتاج نماذج التصميم، يمكن لأدوات GenAI تسريع عملية تصميم تجربة المستخدم بشكل كبير، مما يسمح لمصممي تجربة المستخدم باستكشاف مجموعة أوسع من خيارات التصميم والتكرار بسرعة لإنشاء تجارب مستخدم أكثر فعالية وجاذبية. بينما تقدم أدوات GenAI العديد من الفرص لمصممي تجربة المستخدم، فإنها تقدم أيضاً تهديدات محتملة تتعلق بفقدان الوظائف، والمخاوف الأخلاقية، ونقص اللمسة الإنسانية، وجودة البيانات والتحيز، وقضايا الملكية الفكرية.

2 أسئلة البحث والمساهمات

للاستفادة من قوة الذكاء الاصطناعي التوليدي في دعم مصممي تجربة المستخدم ولضمان الكشف المسؤول عن المزيد من التقدمات في الذكاء الاصطناعي، فإن هدفنا ثلاثي الأبعاد. أولاً، نهدف إلى الحصول على نظرة عامة على ممارسات تصميم تجربة المستخدم الحالية، بما في ذلك الأدوات المستخدمة، وقيودها، بالإضافة إلى سير العمل والتحديات الموجودة. إن تحديد الممارسات والتحديات الحالية يوفر قاعدة يمكن من خلالها تقييم تأثير وفائدة أدوات الذكاء الاصطناعي التوليدي. كما يساعد ذلك في وضع الذكاء الاصطناعي التوليدي في سياق سير العمل الحالي. على سبيل المثال، غالبًا ما تعمل الشركات الكبيرة والشركات الصغيرة بشكل مختلف من حيث سير عمل تصميم تجربة المستخدم. إن فهم هذه السياقات أمر بالغ الأهمية لتقييم كيفية عمل أدوات الذكاء الاصطناعي التوليدي.
قد تتناسب أو تغير هذه التدفقات المختلفة. ثانياً، نسعى لفهم تصورات مصممي تجربة المستخدم تجاه الذكاء الاصطناعي التوليدي – نعتزم استكشاف كيف يمكن لمصممي تجربة المستخدم الاستفادة من إمكانيات الذكاء الاصطناعي التوليدي لتجاوز القيود الحالية في الأدوات والتحديات في التدفقات، مما يسمح لهم في النهاية بخلق تجارب مستخدم مؤثرة وذات مغزى الآن وفي المستقبل. أخيراً، نسعى لفهم كيف وبأي شكل قد يؤثر الذكاء الاصطناعي التوليدي على ممارسة تصميم تجربة المستخدم، وما إذا كانت المخاوف التي قد تنشأ مبررة. لهذا الغرض، نطرح الأسئلة البحثية التالية:
  • RQ1: كيف يرى مصممو تجربة المستخدم أدوات الذكاء الاصطناعي التوليدي، وما هو رأيهم في إمكانية دمج هذه الأدوات في سير العمل الحالي لديهم؟
  • RQ2: ما الفرص والمخاطر التي يتصورها مصممو تجربة المستخدم لمستقبل التعاون بين الإنسان والذكاء الاصطناعي في ممارسة تصميم تجربة المستخدم؟
للإجابة على هذه الأسئلة، أجرينا مقابلات متعمقة فردية مع 20 مصمم تجربة مستخدم لديهم سنوات متنوعة من الخبرة ويعملون حاليًا في شركات تتراوح من الشركات الناشئة إلى الشركات الكبيرة التي تضم أكثر من 10,000 موظف. تقع هذه الشركات بشكل رئيسي في أربع دول عبر أوروبا والولايات المتحدة. قمنا بمقابلة المشاركين لتوصيف ممارساتهم، والاستفسار عن مواقفهم واهتماماتهم وتوقعاتهم.
يساهم عملنا في فهم أعمق لممارسات تصميم تجربة المستخدم الحالية، والتحديات بالإضافة إلى مخاوف وفرص الذكاء الاصطناعي التوليدي لتصميم تجربة المستخدم. نجد أن مصممي تجربة المستخدم ذوي الخبرة واثقون من مهاراتهم من حيث الأصالة والإبداع ومهارات التعاطف مع المستخدم. يعتقدون أن الذكاء الاصطناعي التوليدي يمكن أن يعمل كأداة مساعدة للمساعدة في المهام المتكررة والعامة، ولتعزيز الإنتاجية – ومع ذلك، فقط لأولئك الذين هم بالفعل مصممون مهرة. وأكدوا على العوامل البشرية الفريدة مثل “الاستمتاع” و”الوكالة”، حيث سيظل البشر هم الحكم النهائي على “محاذاة الذكاء الاصطناعي” على الرغم من التوقعات بشأن قدرات الذكاء الاصطناعي الناشئة في المستقبل. ومع ذلك، أعربوا عن مخاوف جدية بشأن تدهور المهارات والبطالة، خاصة بالنسبة للمصممين المبتدئين، الذين قد يفقدون الفرص لدراسة التصميم بشكل منهجي، وقد ينتهي بهم الأمر بدلاً من ذلك إلى أن يتم تدريبهم كمساعدين. علاوة على ذلك، أكدوا على أهمية أن تولي المجتمع اهتمامًا كافيًا للمشكلات المحتملة التي قد تنشأ مع مثل هذه التقدمات في الذكاء الاصطناعي التوليدي، مثل قضايا حقوق الطبع والنشر والملكية، والآثار السلبية على الإبداع البشري (“استنفاد الإبداع”) من خلال سرعة إنتاج الذكاء الاصطناعي والتجانس، والمخاوف بشأن تعزيز معرفة الذكاء الاصطناعي وضمان الوصول المتساوي للبقاء ذي صلة. في النهاية، بينما قد تكون ممارسة تصميم تجربة المستخدم في حالة تغير مستمر، نعتقد أن تأثير الذكاء الاصطناعي التوليدي على هذه المهنة يتطلب تكيفًا فوريًا. نناقش تداعيات هذه النتائج، من خلال عدسة ضمان تطوير الذكاء الاصطناعي المسؤول والتعاون بين الإنسان والذكاء الاصطناعي ضمن ممارسة تصميم تجربة المستخدم.

3.1 تأثير الذكاء الاصطناعي على العمل (الإبداعي)

تدرس مجموعة رئيسية من الأدبيات ذات الصلة تأثير الذكاء الاصطناعي على العمل. يقترح فيلتون وآخرون [27] أن أعلى المهن المعرضة لنمذجة اللغة تشمل المسوقين عبر الهاتف ومجموعة متنوعة من المعلمين في التعليم العالي مثل معلمي اللغة الإنجليزية والأدب، ومعلمي اللغات الأجنبية والأدب، ومعلمي التاريخ. يقترح إلوندو وآخرون [23] أن حوالي من قوة العمل الأمريكية قد تتأثر على الأقل من مهام عملهم بسبب إدخال نماذج اللغة الكبيرة، بينما قد يرى حوالي من العمال على الأقل من مهامهم متأثرة. يصف سليمان وآخرون [89] التأثير الاجتماعي المحدد الذي قد تحدثه أنظمة الذكاء الاصطناعي التوليدي على المجتمع – الإطار يحدد سبع فئات من التأثير الاجتماعي: التحيز، الصور النمطية، والأضرار التمثيلية؛ القيم الثقافية والمحتوى الحساس؛ الأداء المتباين؛ الخصوصية وحماية البيانات؛ التكاليف المالية؛ التكاليف البيئية؛ وتكاليف العمل في تعديل البيانات والمحتوى. أظهرت أعمال أخرى بشكل تجريبي المكاسب الكبيرة في الإنتاجية من خلال استخدام أدوات الذكاء الاصطناعي الحديثة مثل ChatGPT [71، 77].
ناقشت عدد من الأعمال تأثيرات الذكاء الاصطناعي على الأعمال الإبداعية، والتي ترتبط ارتباطًا وثيقًا بتصميم تجربة المستخدم والبحث [32]. يتناول بيرسون [76] دور تقنيات الذكاء الاصطناعي الإبداعية الحديثة كمصدر إلهام محتمل للفنانين والمبدعين، مسلطًا الضوء على قدرتها على تمكين أشكال جديدة من إنشاء الصور، وتأليف الموسيقى، والرسوم المتحركة، وإنتاج الفيديو. يستكشف كولكارني وآخرون [55] استخدام نماذج النص إلى الصورة (TTIs) في التصميم التعاوني من خلال إشراك 14 مصممًا غير محترف. يكشفون أن TTIs تسهل الاستكشاف السريع لمساحات التصميم وتدعم التعاون السلس، حيث تعمل النصوص كأدوات تصميم عاكسة تسهل الاستكشاف والتكرار والتفكير في التصميم الثنائي. يستكشف نينغ وآخرون [69] مساحة تصميم أدوات دعم الإبداع المدفوعة بتقنية الذكاء الاصطناعي (AI-CSTs)، مسلطين الضوء على تأثير AI-CSTs على سير العمل، وإمكاناتها كمنشئين مشاركين، واستراتيجيات التعامل مع أخطاء الذكاء الاصطناعي، مما يوفر رؤى حول متطلبات تصميم وتكنولوجيا AI-CSTs. يفحص إبستين وآخرون [25] تأثير أدوات الذكاء الاصطناعي التوليدي على الممارسات الفنية التقليدية، معالجين أسئلة حول التأليف والملكية والتحول المحتمل للعمل الإبداعي والتوظيف. درس تشانغ وآخرون [15] إدراك الفنانين الذين استخدموا نماذج الذكاء الاصطناعي النص إلى الصورة لأعمالهم الفنية، ووجدوا أن 1) يحتفظ الفنانون بالنص الموجه والصورة الناتجة يمكن اعتبارها بشكل جماعي كشكل من أشكال التعبير الفني (النصوص كفن)، و2) تم تطوير قوالب النصوص (نصوص تحتوي على “فراغات” لملئها بكلماتهم الخاصة) لإنشاء أنماط فنية توليدية. اقترح فينشون وآخرون [98] أربعة مستقبلات محتملة لعلاقة الذكاء الاصطناعي بالإنسان في العمل الإبداعي، وهي “التعاون الإبداعي بالذكاء الاصطناعي”، “العضوي”، “السرقة الأدبية 3.0″، و”الإغلاق”. وبشكل أكثر ارتباطًا بالعمل الحالي، ناقش إيني وآخرون [47] مخاوف وتوقعات المهنيين الإبداعيين بشأن الذكاء الاصطناعي التوليدي، مستنتجين أن المهنيين الإبداعيين يجب أن يفهموا بشكل أفضل، ويتعاملوا مع، ويتكيفوا مع، بالإضافة إلى استغلال الذكاء الاصطناعي. قدم لونغ وماجيركو [63] تعريفًا ملموسًا لمحو الأمية في الذكاء الاصطناعي من خلال تحديد 16 كفاءة أساسية (مثل التعرف على الذكاء الاصطناعي، ودور الإنسان في الذكاء الاصطناعي، والأخلاقيات) التي يحتاجها البشر للتفاعل بفعالية مع الذكاء الاصطناعي وتقييمه بشكل نقدي، بالإضافة إلى 15 اعتبارًا تصميميًا موصى بها تعزز فهم المستخدم للذكاء الاصطناعي (مثل القابلية للتفسير، وتعزيز الشفافية، وانخفاض الحواجز للدخول). تكملًا لما سبق، نركز ونقوم بتحليل متعمق مخصص لفهم كيف يؤثر الذكاء الاصطناعي على مهنة معينة مرتبطة ارتباطًا وثيقًا بمجتمع التفاعل بين الإنسان والحاسوب – ممارسة تصميم تجربة المستخدم.

3.2 التعاون بين الإنسان والذكاء الاصطناعي لتمكين الإنسان

تمت مناقشة العلاقة بين الذكاء البشري والذكاء الآلي منذ الأيام الأولى للتفاعل بين الإنسان والحاسوب [44، 60]. نظرًا للأداء المتزايد وانتشار الذكاء الاصطناعي في المجتمع، تم جذب الكثير من الانتباه لدراسة وتصميم التعاون الفعال والمسؤول بين الإنسان والذكاء الاصطناعي في السنوات الأخيرة. اقترح أمريشي وآخرون [2] إرشادات تصميم للتفاعل بين الإنسان والذكاء الاصطناعي. من خلال عدة جولات من التقييم مع الممارسين، قاموا بتكثيف أكثر من 150 توصية تصميم مرتبطة بالذكاء الاصطناعي إلى 18 جانبًا تتعلق بأربع مراحل زمنية مختلفة، واختبروا إرشاداتهم ضد 20 منتجًا مدمجًا بالذكاء الاصطناعي من فئات وميزات متنوعة (مثل التجارة الإلكترونية، والبحث على الويب). استقصى يانغ وآخرون [103] ما إذا كان، ولماذا، وكيف يمارس التفاعل بين الإنسان والذكاء الاصطناعي صعوبات فريدة في التصميم. من خلال دمج الأبحاث السابقة مع تصميمهم وتجاربهم التعليمية، حددوا عدم اليقين في قدرة الذكاء الاصطناعي وتعقيد مخرجات الذكاء الاصطناعي كمصدرين من مصادر التحديات الفريدة في تصميم الذكاء الاصطناعي. قبل أكثر من عقدين، قدم هورفيتز [44] مجموعة من المبادئ لبناء واجهات مستخدم مختلطة المبادرة (UIs) التي تمكن المستخدمين والوكالات الذكية من التعاون بكفاءة. لا تزال المبادئ المقترحة قابلة للتطبيق على التعاون بين الإنسان والذكاء الاصطناعي اليوم، مثل “توفير آليات للتعاون الفعال بين الوكيل والمستخدم لتحسين النتائج” و”استخدام سلوكيات اجتماعية مناسبة لتفاعل الوكيل والمستخدم”. كما يستقصي ليهمان [58] التفاعلات المختلطة المبادرة بين الإنسان والذكاء الاصطناعي والعمل التعاوني مع الأنظمة التوليدية. يركز المؤلف على تصميم وتقييم النماذج الأولية الوظيفية من خلال تجارب قائمة على الويب، مستكشفًا مفاهيم مثل المبادرة، والنية، والتحكم، موضحًا أن مستويات المبادرة والتحكم التي توفرها واجهات المستخدم تؤثر على الإحساس بالتأليف عند كتابة النص.
استخدم كيم وآخرون [52] طرق حسابية لتصنيف عشرة أدوار مختلفة للذكاء الاصطناعي شائعة في حياتنا اليومية وقارنوا تصورات الناس العاديين عنها باستخدام بيانات استطلاع عبر الإنترنت، على سبيل المثال، يُعتبر الذكاء الاصطناعي له أدوار مثل الأدوات (منخفضة في كل من مشاركة الإنسان واستقلالية الذكاء الاصطناعي)، والخدم (مشاركة إنسانية عالية واستقلالية ذكاء اصطناعي منخفضة)، والمساعدين (مشاركة إنسانية منخفضة واستقلالية ذكاء اصطناعي عالية)، والوسطاء (عالية في كلا البعدين). وجدوا أن الناس يقيمون الوسطاء الذكاء الاصطناعي بشكل إيجابي أكثر، والأدوات الذكاء الاصطناعي بشكل أقل. ذات صلة، درس سكوت وآخرون [84] ما إذا كان الناس العاديون يرون الذكاء الاصطناعي على أنه “واعي”، ووجدوا توترات ديناميكية بين الإنكار والتخمين، والتفكير والشعور، والتفاعل والخبرة، والسيطرة والاستقلال، والجمود والعفوية. استكشف كابل وبريتون [14] المجال الناشئ للذكاء الاصطناعي المتمركز حول الإنسان (HCAI) من خلال مراجعة الأدبيات الموجودة حول هذا الموضوع. يحدد عملهم مجموعات HCAI المعروفة، ويسلط الضوء على المجالات الناشئة، بما في ذلك التفاعل مع الذكاء الاصطناعي والأخلاقيات في الذكاء الاصطناعي. بالإضافة إلى ذلك، يقترحون تعريفًا جديدًا لـ HCAI ويشجعون على مزيد من التعاون بين أبحاث الذكاء الاصطناعي وHCI بينما يقترحون مفاهيم جديدة لـ HCAI. بالمثل، استكشف شنييدرمان [86] مفهوم HCAI كنهج تحويلي لتصميم أنظمة الذكاء الاصطناعي، مؤكدًا على أهمية وضع البشر في قلب التفكير التصميمي، مقدمًا ثلاث أفكار رئيسية: (1) إطار عمل HCAI ثنائي الأبعاد يوازن بين السيطرة البشرية والأتمتة، (2) تحول من تقليد البشر إلى تمكين الناس، و(3) هيكل حوكمة من ثلاثة مستويات لإنشاء أنظمة HCAI موثوقة وآمنة وجديرة بالثقة. تقترح هذه الأفكار إعادة صياغة في مناقشات التصميم مع إمكانية تحقيق فوائد أكبر للأفراد والمجتمعات والمجتمع، على الرغم من أنها ستتطلب التحقق والتنقيح من خلال مزيد من البحث والتطوير. في الواقع، تهدف مثل هذه الطرق HCAI [85] في النهاية إلى إنتاج تصاميم موثوقة وآمنة وجديرة بالثقة (RST)، وتعزيز الكفاءة الذاتية البشرية، والإبداع، والمسؤولية بينما تحسن الأداء بشكل كبير. راجع بيرهان وآخرون [5] الطرق والممارسات التشاركية ضمن خط أنابيب الذكاء الاصطناعي وتعلم الآلة وأقروا أن الجمهور الأوسع بخلاف الخبراء الفنيين لديه معرفة وخبرة واهتمامات ضرورية لتصميم وتطوير ونشر الذكاء الاصطناعي. تضع هذه الأعمال السابقة الأساس لضمان، على كل من مستوى البحث والسياسة، ضمان التعاون المسؤول بين البشر والذكاء الاصطناعي الذي يمكّن في النهاية الناس والممارسات. نستند إلى هذه الأعمال للمساعدة في تفسير نتائجنا حول كيفية تصور مصممي تجربة المستخدم للتعاون الحالي والمستقبلي بين البشر والذكاء الاصطناعي.

3.3 استخدام الذكاء الاصطناعي في تصميم وتجربة المستخدم

دراسة واجهة المستخدم (UI) وتجربة المستخدم (UX) المقابلة، وكيفية بناء أدوات لدعم تصميم وتجربة المستخدم بشكل أفضل كانت موضوعًا محوريًا في مجتمع HCI [12، 32، 53]. مع إظهار الذكاء الاصطناعي بشكل متزايد إمكانياته في تحويل ممارسات العمل، بدأ الباحثون في دراسة ومناقشة آثار الذكاء الاصطناعي على تصميم وتجربة المستخدم [51، 90]. حتى الآن، أظهرت الأبحاث الإمكانات الهائلة للذكاء الاصطناعي في تصميم تجربة المستخدم، مع نماذج أولية وتطبيقات متاحة في العمليات الرئيسية، بما في ذلك فهم سياق الاستخدام، وكشف متطلبات المستخدم، والمساعدة في حل المشكلات، وتقييم التصميم، والمساعدة في تطوير الحلول [1، 90]. بشكل عام، يعتقد معظم الناس أن الذكاء الاصطناعي من المحتمل أن يؤدي إلى أدوات دعم الإبداع المعززة بالذكاء الاصطناعي بدلاً من تحول كامل [61، 98]. استقصى ثولاندر وآخرون [92] كيف يمكن أن تلعب التعلم الآلي التوليدي ونماذج اللغة الكبيرة دورًا في عمليات التصميم الإبداعي من الفكرة، والنمذجة الأولية المبكرة، والرسم، موضحين الفائدة العملية والقيود للنظام في عمليات توليد الأفكار التصميمية، بالإضافة إلى كيفية تشكيل تفاعل المستخدم، والنقاش الأوسع حول الذكاء الاصطناعي، لتوقعات المستخدم حول قدرات وإمكانات الذكاء الاصطناعي. وجد فنغ وآخرون [28] أن ممارسي تجربة المستخدم يمكن أن يكونوا أكثر تراجعًا عند تصميم النماذج الأولية حيث يمكن أن يوفر الذكاء الاصطناعي الدعم. قدموا بالتفصيل كيف يتواصل ممارسو تجربة المستخدم مع الذكاء الاصطناعي ويستخدمون الذكاء الاصطناعي كمواد تصميم لعملهم اليومي. اقترحت أعمال أخرى أيضًا نماذج أولية محددة لدعم تجارب المستخدم بشكل أفضل [38، 99]. تشير تقييمات مثل هذه الأنظمة إلى إمكانيات الذكاء الاصطناعي في تعزيز الإبداع البشري، حيث غالبًا ما يجلب الذكاء الاصطناعي منظورًا مميزًا يفتح آفاقًا جديدة للتعبير الفني [16]، ويعزز تبادل الأفكار بين المستخدمين والذكاء الاصطناعي [56]. استكشف لين وآخرون [62] تصميم
مساحة أنظمة التعاون الإبداعي المختلط حيث يمكن للبشر وأنظمة الذكاء الاصطناعي التواصل مع بعضهم البعض حول النوايا الإبداعية.
في الوقت نفسه، كشفت الأبحاث أيضًا عن تحديات دمج الذكاء الاصطناعي في عملية تصميم تجربة المستخدم. أظهر غماينر وآخرون [35] أن المصممين يواجهون العديد من التحديات في تعلم كيفية التعاون بشكل فعال مع أنظمة الذكاء الاصطناعي التوليدية الحالية، بما في ذلك التحديات في فهم وتعديل مخرجات الذكاء الاصطناعي وفي التواصل حول أهداف تصميمهم. على سبيل المثال، عند توقع أدوات الذكاء الاصطناعي أن تقدم لهم أمثلة عمل ذات صلة بالمشروع، أو تظهر وعيًا أكبر بالسياق للمشكلة المحددة المطروحة. أظهر زانغ وآخرون [106] المزيد من القيود على الذكاء الاصطناعي في التصميم المعماري، حيث “يفشل أداة الذكاء الاصطناعي في فهم المصطلحات الخاصة بالمجال المعماري” و”تولد صورًا غير واقعية غير مناسبة لأغراض البناء”. تظهر أعمال أخرى تحديات مماثلة في هندسة المطالبات عند استخدام الذكاء الاصطناعي في تصميم المنتجات [43]. من ناحية أخرى، وجد عباس وآخرون [1] أنه بينما لم يكن لدى الغالبية العظمى من مصممي تجربة المستخدم خبرة في التعلم الآلي كأداة تصميم، فإن لديها إمكانات كبيرة لتحسين الإنتاجية. عرّف ثورينغ وآخرون [93] أجندة بحثية لتصميم مدعوم بالذكاء الاصطناعي، مع تسليط الضوء على المجالات الرئيسية التي تشمل إنشاء مطالبات موجهة بشكل أفضل، وواجهات أفضل لتداخل المخرجات، وجعل الذكاء الاصطناعي يفكر خارج صندوق الفضاء الكامن (أي، الاستقراء). فيما يتعلق بإمكانات الذكاء الاصطناعي في أبحاث تجربة المستخدم، اقترح يانغ وآخرون [102] منهجية لمحاكاة تجربة المستخدم باستخدام تقنية التصميم المدعومة بالذكاء الاصطناعي في تصميم التطبيقات المحمولة. استغل هاملينين وآخرون [37] نموذج GPT-3 من OpenAI لتوليد استجابات استبيانات مفتوحة حول تجارب المستخدمين في ألعاب الفيديو، حيث وجدوا أن GPT-3 يمكن أن ينتج حسابات موثوقة لتجارب HCI يصعب تمييزها عن الاستجابات البشرية الحقيقية. استغل بارك وآخرون [74] GPT-3 لمحاكاة مستخدمين اصطناعيين ومحادثات حول نماذج أولية لمنصات الحوسبة الاجتماعية، حيث وجدوا أن الاستجابات المولدة يصعب تمييزها عن سلوك المجتمع الفعلي. على النقيض من الغالبية العظمى من الأدبيات التي تركز على تقييم استخدام الذكاء الاصطناعي في تصميم وتجربة المستخدم في إعدادات أو تقنيات محددة، تسهم أعمالنا في فهم أعمق للممارسات الحالية، واعتماد الذكاء الاصطناعي، وتصورات ممارسي تجربة المستخدم بشأن تقنيات الذكاء الاصطناعي التوليدية الناشئة، مما يضع الأساس لتصور كيف قد تتطور تجربة المستخدم كمهنة في المستقبل مع القدرات المتزايدة باستمرار للذكاء الاصطناعي.

4 الطريقة

لقد استخدمنا منهجية المقابلات شبه المنظمة لهذه الدراسة [64]. في البداية، وزعنا استبيان فحص بهدف تجنيد مصممي تجربة المستخدم الذين لديهم خلفيات تعليمية وتجارب صناعية متنوعة، بما في ذلك سنوات الخبرة، والعمل في أحجام شركات مختلفة، والإقامة في قارتين مختلفتين على الأقل (وهي أوروبا وأمريكا الشمالية). تم توزيع استبيان الفحص عبر شبكات التواصل الاجتماعي (مثل لينكد إن) وقوائم البريد (بما في ذلك مدارس التصميم)، مستهدفين مجتمعات تصميم تجربة المستخدم. تلقينا 48 استجابة، تمت مراجعة جميعها. بينما استوفى جميع المجيبين معايير التجنيد الخاصة بنا، اختار 20 مشاركًا فقط المشاركة. كانت هذه المجموعة من المشاركين متوازنة بشكل جيد من حيث متطلبات التجنيد لدينا. شارك هؤلاء الـ 20 مشاركًا معنا في مقابلات فردية عبر الإنترنت لمدة 60 دقيقة بين أبريل ويوليو 2023. اعتُبر عدد 20 مشاركًا كافيًا نظرًا لنقاط التشبع التي تتجاوز 17 مقابلة في الأبحاث السوسيولوجية [41] ومعايير حجم العينة المحلية في مجال التفاعل بين الإنسان والحاسوب (عادةً ما تكون المقابلات عن بُعد تم إجراء كل مقابلة بواسطة محاور (المؤلف)، بينما كان باحث ثانٍ (أيضًا مؤلف) حاضرًا لتدوين الملاحظات. تم تسجيل جميع جلسات المقابلة صوتيًا أيضًا. كانت عملية المقابلة، وجمع البيانات، وتخزينها تلتزم بدقة باللائحة العامة لحماية البيانات (GDPR) وسياسة الأخلاقيات وإرشادات حماية البيانات التي وضعتها مؤسسة البحث الخاصة بالمؤلف الأخير.
خصائص مصممي تجربة المستخدم
خبرة العمل >5 سنوات ب1، ب5، ب6، ب8، ب10، ب11، ب12، ب14، ب16، ب17، ب18
3-5 سنوات ب2، ب7
<3 سنوات P3، P4، P9، P13، P15، P19، P20
حجم الشركة >10,000 P5، P6، P7، P8، P9، P11، P14، P15، P16، P17
1,000-10,000 ب1، ب2، ب10، ب13، ب18
500-1,000 P20
<50 P3، P4، P12، P19
الجدول 1. سنوات خبرة العمل وحجم الشركة (عدد الموظفين) لمشاركي تصميم تجربة المستخدم لدينا (P1-P20)
عروض فيديو GenAI تشمل أدوات الذكاء الاصطناعي تاريخ النشر على يوتيوب
تصميم واجهة موبايل ميدجورني (V4)، ملحق استقرار فوتوشوب (0.7.0) وChatGPT (نسخة 30 يناير 2023) 20 فبراير 2023
صياغة اقتراح تصميم (استخدمت فقط جزء من Notion AI: نوتيون AI (2.21) 15 مارس 2023
إنشاء مكونات واجهة المستخدم برومبت2ديزاين (الإصدار 3) 6 مايو 2023
تنظيم ملاحظات الأفكار ميرو إيه آي (بيتا) 8 مارس 2023
الجدول 2. عروض الفيديو الخاصة بـ GenAI المستخدمة في المقابلة

4.1 المشاركون

لضمان إمكانية تتبع النتائج، قمنا بتسمية المشاركين العشرين بـ P1-P20. من بين المشاركين العشرين، كان هناك 12 أنثى و8 ذكور. كان ثلاثة من المشاركين ضمن الفئة العمرية 18-25، بينما كان 12 في الفئة 26-35. وكان هناك خمسة مشاركين إضافيين ينتمون إلى فئة العمر. كان لدى 11 منهم أكثر من خمس سنوات من الخبرة كمصممي تجربة المستخدم، وكان لدى اثنين منهم من ثلاث إلى خمس سنوات، وكان لدى سبعة منهم أقل من ثلاث سنوات من الخبرة. بالنسبة لخبرتهم في استخدام أدوات الذكاء الاصطناعي التوليدي، استخدم جميع المشاركين العشرين ChatGPT بانتظام، وثمانية منهم استخدموا GPT-4 (المعروف باسم ChatGPT Plus) بانتظام. جرب ثلاثة عشر مشاركًا Midjourney واستخدموه أحيانًا للإلهام. كما ذكر سبعة مشاركين استخدامهم العرضي لـ Notion AI. جرب أربعة Miro AI، واستخدم اثنان Bard.
كان المشاركون العشرون من 15 شركة مختلفة، حيث كان عشرة مشاركين موجودين في أوروبا، بما في ذلك هولندا، السويد، رومانيا، المجر، وكان المشاركون العشرة الآخرون موجودين في الولايات المتحدة. علاوة على ذلك، كان عشرة مشاركين يعملون في مؤسسات تضم أكثر من 10,000 موظف (ضمن شركات التكنولوجيا الكبرى المعروفة). الشركات)، عمل خمسة في مؤسسات معروفة دولياً تضم من 1,000 إلى 10,000 موظف، وكان واحد يعمل في شركة تضم من 500 إلى 1,000 موظف، وعمل أربعة في شركات صغيرة أو شركات ناشئة تضم أقل من 50 موظفاً. لضمان خصوصية البيانات، نضمن أن مكان عمل المشاركين لا يمكن تتبعه إليهم (الجدول 1).

4.2 أسئلة المقابلة والإجراءات

تم إعداد أسئلة المقابلة لجمع الإجابات على أسئلة البحث المقدمة في القسم 1، والتي تنقسم إلى أربعة أجزاء. يبدأ الجزء الأول بأسئلة عامة تهدف إلى جمع معلومات حول خلفيات المشاركين، وتجاربهم العملية، والتحديات الشائعة في تصميم تجربة المستخدم، والقيود التي يرونها في أدواتهم الحالية. يتكون الجزء الثاني من مجموعة من الأسئلة التي تركز على فهم تجارب المشاركين الحالية ومعرفتهم بأدوات الذكاء الاصطناعي التوليدي، بالإضافة إلى مواقفهم تجاه استخدام الذكاء الاصطناعي التوليدي في عمل تصميم تجربة المستخدم. في الجزء الثالث، يتم عرض مجموعة من مقاطع الفيديو على المشاركين.
الشكل 1. أسفرت التحليل عن ستة مواضيع رئيسية، حيث احتوى كل موضوع على اثنين إلى أربعة مواضيع فرعية.
يتكون من أربع عروض توضيحية من يوتيوب باستخدام أدوات GenAI المختلفة لتصميم تجربة المستخدم، وتوليد الأفكار، وتجميع بيانات البحث، وأعمال تطوير الواجهة الأمامية (الجدول 2). استخدمنا هذه العروض التوضيحية بشكل أساسي كأدوات لاستكشاف آراء المشاركين حول هذه الأدوات بشكل عام وكيف ترتبط بممارساتهم، وليس لتقييم ما إذا كانوا قد استخدموا هذه الأدوات المحددة بأنفسهم. تتناول الأسئلة اللاحقة انطباعاتهم عن أدوات GenAI المعروضة في الفيديوهات، والأدوار التي يتصورونها للذكاء الاصطناعي في التعاون بين الإنسان والذكاء الاصطناعي، واهتماماتهم بشأن الأخلاقيات والملكية. الجزء 4 يختتم المقابلة بأسئلة تطلب مزيدًا من التأملات، والنصائح أو التوصيات لمصممي تجربة المستخدم المهتمين بدمج GenAI في عملهم وتحديد أكثر الوظائف الابتكارية التي يرغبون بها في GenAI الحالية والمستقبلية. القائمة الكاملة لأسئلة المقابلة متاحة في المواد التكميلية A.
استغرق عملية المقابلة حوالي 60 دقيقة وشملت الخطوات التالية:
  • الخطوة 1 (5 دقائق): مقدمة يقدمها ميسر المقابلة لضمان فهم المشاركين لغرض المقابلة، وأنهم قد قرأوا وفهموا ووقعوا على نموذج الموافقة المستنيرة؛
  • الخطوة 2 ( أسئلة عامة حول الخلفية، سير العمل، وممارسة تجربة المستخدم؛
  • الخطوة 3 (20 دقيقة): المواقف والمخاوف والتوقعات من أنظمة الذكاء الاصطناعي العام؛
  • الخطوة 4 (20 دقيقة): تصور دور الذكاء الاصطناعي التوليدي في التعاون بين الإنسان والذكاء الاصطناعي (بما في ذلك مشاهدة عرض فيديو لعينة من أدوات الذكاء الاصطناعي التوليدي الحالية);
  • الخطوة 5 (5 دقائق): أسئلة ختامية.
الشكل 2. ثماني فئات من الأدوات المستخدمة بشكل شائع في ممارسات تصميم تجربة المستخدم
الشكل 3. الأنشطة التي يمكن ولا يمكن إكمالها بواسطة GenAI في سير عمل تصميم تجربة المستخدم

4.3 تحليل البيانات

استخدمنا ميرو للتعاون عن بُعد في تحليل البيانات. تم نسخ التسجيلات الصوتية للمقابلات. قمنا بتحليل البيانات وفقًا لنهج تحليل موضوعي هجين استنتاجي/استقرائي [29، 101]: أولاً، تم ترميز النصوص بواسطة خمسة باحثين باستخدام نهج الترميز الاستنتاجي، حيث حدد كل باحث مواضيع في النصوص بشكل مستقل وفقًا لأربع فئات في دليل الترميز الخاص بنا، الذي تم إنشاؤه بناءً على أسئلة البحث ودليل المقابلة. بعد مرحلة الترميز، عقد الباحثون الخمسة ورشة عمل لمدة نصف يوم لمقارنة ومناقشة
تم تحديد الموضوعات وتحويل البيانات المشفرة إلى 415 بطاقة بيان رقمية على ميرو. تتكون كل بطاقة بيان من الاقتباس الأصلي من المشارك، ورقم تعريف المشارك (P1-P20)، وبيان أو ملخص من جملة واحدة للاقتباس على البطاقة. بدلاً من حساب موثوقية المقيّمين الإحصائية (IRR) للتحليل، تم التوصل إلى توافق بين الباحثين الخمسة من خلال الاجتماعات اليومية وورش العمل والنقاشات. نظرًا لأن الفئات في دليل الترميز كانت لا تزال واسعة جدًا، اعتمدنا بعد ذلك نهج الترميز الاستقرائي لاستخراج الموضوعات الفرعية من بطاقات البيان باستخدام تقنية رسم المخططات التوافقية، ولضمان دمج الفئات الأوسع. للمساعدة في ضمان أن تحليلنا يلبي معايير الموثوقية، نصف نهجنا الهجين بالتفصيل في المواد التكميلية ب. أسفر التحليل عن ستة موضوعات رئيسية: (1) القيود في أدوات تصميم تجربة المستخدم الحالية: (أ) انتقال التصميم المعيق؛ (ب) الأدوات المجزأة؛ و(ج) نقص الدعم المتقدم في التصميم. (2) التحديات في تصميم تجربة المستخدم الحالي: (أ) إدارة توقعات أصحاب المصلحة؛ (ب) فهم المستخدمين؛ (ج) تلبية احتياجات الأعمال والمستخدمين؛ و(د) نقص المعرفة في المجال. (3) مواقف مصممي تجربة المستخدم تجاه الذكاء الاصطناعي التوليدي: (أ) الانطباع العام عن الفيديوهات؛ و(ب) أوجه القصور في أدوات الذكاء الاصطناعي التوليدي في الفيديوهات. (4) التعاون بين الإنسان والذكاء الاصطناعي: (أ) التهديدات والتعاون مع الذكاء الاصطناعي التوليدي؛ (ب) الأعمال القابلة للاستبدال بالذكاء الاصطناعي التوليدي؛ و(ج) الأعمال غير القابلة للاستبدال بالذكاء الاصطناعي التوليدي. (5) المخاوف الأخلاقية والملكية: (أ) الأصالة، حقوق الطبع والنشر، والملكية؛ (ب) تدهور المهارات، والبطالة؛ (ج) الخصوصية، الشفافية، والوصول المتساوي إلى الذكاء الاصطناعي التوليدي. (6) مستقبل الذكاء الاصطناعي التوليدي في تصميم تجربة المستخدم: (أ) تصور التعاون مع الذكاء الاصطناعي التوليدي؛ (ب) تعزيز معرفة الذكاء الاصطناعي وإمكانية الوصول للمصممين؛ (ج) الدور المرغوب للذكاء الاصطناعي في تحسين عمليات التصميم؛ و(د) تحديات تعريف مقاييس النتائج الناتجة عن الذكاء الاصطناعي.

5 نتائج

توضح هذه القسم أدوات وعمليات تصميم تجربة المستخدم الحالية، وتقدم عرضًا تفصيليًا للنتائج تحت المواضيع الستة المحددة في تحليل البيانات (القسم 4.3 والشكل 1). يعرض الشكل 3 الأنشطة في عملية تصميم تجربة المستخدم التي ذكرها مصممو تجربة المستخدم في دراستنا والتي يمكن ولا يمكن إكمالها بواسطة الذكاء الاصطناعي العام.

5.1 أدوات تصميم تجربة المستخدم الحالية وسير العمل

فهم القيود والتحديات التي تواجه سير العمل الحالي في تصميم تجربة المستخدم أمر حاسم لإنشاء قاعدة لتقييم تأثير وفائدة أدوات الذكاء الاصطناعي التوليدي. هذه المعرفة لا تُعلم فقط كيف يمكن للذكاء الاصطناعي التوليدي معالجة القضايا الحالية وتعزيز سير العمل، ولكنها أيضًا تساعد في وضع هذه الأدوات في سياق بيئات تشغيلية مختلفة، من الشركات الكبرى إلى الشركات الصغيرة. هذا أمر أساسي لتقييم التكامل والتعديلات المحتملة التي قد تجلبها أدوات الذكاء الاصطناعي التوليدي إلى هذه السير المتنوعة، حيث قد لا تناسب مقاس واحد الجميع. علاوة على ذلك، تلعب هذه الفهم دورًا رئيسيًا في تحديد توقعات واقعية لاعتماد أدوات الذكاء الاصطناعي التوليدي، مما يوفر رؤى حول الجوانب التي من المحتمل أن يتم تبنيها أو مقاومتها من قبل مصممي تجربة المستخدم.
قمنا بتجميع ثماني فئات من الأدوات المستخدمة بشكل متكرر التي تدعم مصممي تجربة المستخدم في عملهم اليومي (الشكل 2)، بالإضافة إلى ثماني خطوات نموذجية ضمن ممارسة تصميم تجربة المستخدم – بدءًا من جمع المتطلبات من أصحاب المصلحة في الأعمال والمستخدمين، إلى التفكير الإبداعي، والنمذجة، واختبار المستخدم، والتكرار، وتمكين تحقيق الحلول مع فريق التطوير. توضح الجدول 3 الأنشطة الرئيسية والأدوات المستخدمة في كل خطوة. نلاحظ أن Figma تم ذكرها من قبل جميع المشاركين كأداة تصميم لدعم النمذجة عالية الدقة وتسليمها إلى فرق التطوير. كما تم ذكر Miro بشكل متكرر للعصف الذهني وتجميع بيانات أبحاث المستخدم.
من حيث سير العمل، لاحظنا أن الشركات الكبيرة (التي تضم أكثر من 10,000 موظف) غالبًا ما تكون لديها تقسيمات واضحة للأدوار بين مصممي تجربة المستخدم ومديري المنتجات وباحثي تجربة المستخدم. على النقيض من ذلك، فإن الشركات الصغيرة، مثل الشركات الناشئة التي تضم أقل من 50 موظفًا، غالبًا ما تفتقر إلى مثل هذه التقسيمات الواضحة. يتحمل مصممو تجربة المستخدم في الشركات الصغيرة مسؤولية إجراء أبحاث المستخدم وقد يقضون وقتًا أطول في إقناع أصحاب المصلحة بإعطاء الأولوية لتجربة المستخدم. علاوة على ذلك، فإن سير العمل
خطوات الأنشطة الرئيسية أدوات
1 متطلبات العمل
– تحليل متطلبات أصحاب المصلحة
– عقد اجتماع انطلاق مع أصحاب المصلحة لتوحيد التوقعات والأهداف التجارية والتحديات التقنية.
عصف ذهني
2 متطلبات المستخدم
– تحديد أهداف البحث وتحديد الميزات اللازمة لتحقيق تلك الأهداف.
– دمج الرؤى من أبحاث المستخدمين لإنشاء شخصيات وخرائط تعاطف.
– الغوص في دوافع المستخدم ونقاط الألم والتحديات لإبلاغ قرارات التصميم.
العصف الذهني وبحث المستخدم
٣
– تحويل رؤى البحث إلى أفكار تصميم من خلال الرسومات.
– إجراء مراجعات داخلية لصقل الأفكار مع فريق التصميم.
– تنظيم ورش عمل مع أصحاب المصلحة لتوحيد حلول التصميم.
العصف الذهني، التوثيق والعرض
4 النمذجة
– قم بإنشاء نماذج تفاعلية تمثل بصريًا ووظيفيًا الحلول التصميمية المقترحة.
– تأكد من أن النماذج الأولية تلتقط تفاعلات المستخدم، وسير العمل، ومسارات التنقل.
تصميم
٥
– إجراء اختبارات المستخدم لتقييم النماذج الأولية.
– قم بتطبيق اختبار المستخدم عبر الإنترنت (أو خارج الإنترنت) المراقب (أو غير المراقب) لإرشاد المشاركين من خلال مهام محددة مسبقًا.
الاتصال وبحوث المستخدمين
6 تكرار
– تحليل نتائج اختبار المستخدم، وتحديد المجالات التي تحتاج إلى تحسين.
– قم بتكرار التصميم بناءً على الرؤى المكتسبة
التصميم وغير الرقمي
7 حل – التعاون مع أصحاب المصلحة لضمان أن التصميم يلبي احتياجات المستخدم ويتماشى مع أهداف العمل. التوثيق والعرض
8 تسليم التصميم
– توفير أصول التصميم والوثائق لفريق التطوير وإرشادهم لضمان اتساق التصميم ودقته عبر أحجام الشاشات المختلفة.
– معالجة القيود الفنية وإجراء تعديلات على واجهة المستخدم لتحسين واجهة المستخدم للمنتج النهائي.
التوثيق، العرض والتصميم
الجدول 3. سير عمل تصميم تجربة المستخدم النموذجي في عام 2023
تميل العمليات في الشركات الصغيرة إلى أن تكون أقصر من تلك الموجودة في الشركات الكبيرة بسبب قلة الأشخاص المعنيين وانخفاض تكاليف التواصل. ومع ذلك، قد تتجاهل الشركات الصغيرة أو تبسط بعض الخطوات بسبب قيود الموارد، مثل الميزانية المحدودة لتقييمات المستخدمين الشاملة. أوضح P6، الذي لديه خبرة في العمل في كل من الشركات الصغيرة والشركات الكبيرة، “في الشركات الكبيرة، تكون أدوارنا [مصممي تجربة المستخدم] محددة جيدًا. على الرغم من أننا نواجه أيضًا مشهدًا تجاريًا معقدًا وأصحاب مصلحة لديهم آراء مختلفة، لدينا متعاونون مثل باحثي تجربة المستخدم والمنتج.
يجب على المديرين التعامل مع تلك الجوانب، ويمكننا التركيز أكثر على التصميم. في الشركات الصغيرة، يجب علينا [مصممي تجربة المستخدم] تحمل أدوار الباحثين ومديري المنتجات والمصممين.

5.2 الموضوع 1: قيود أدوات تصميم تجربة المستخدم الحالية

5.2.1 القيد 1: تسليم التصميم إلى فرق التطوير. أشار المشاركون إلى أن تسليم مواصفات التصميم والأصول والوثائق إلى فريق التطوير هو عملية تعاني من الاحتكاك. لا تدعم أدوات التصميم الحالية انتقالًا سلسًا من التصميم إلى التطوير بصيغة يمكن الوصول إليها وفهمها من قبل المطورين: “[نحتاج] إلى المزيد من البوابات بين المصممين والمطورين، لتحسين سير العمل في النماذج الأولية من خلال تمكين نماذج Figma من التغذية مباشرة في سير عمل ترميز التطوير، لتقليل الاحتكاك في التواصل بين فرق التصميم والتطوير.”[P14]
5.2.2 القيد 2: الأدوات المجزأة. أكد المشاركون أن أدوات تصميم تجربة المستخدم الحالية متفرقة. كل أداة لها نقاط قوة أو وظائف محددة، لكن الانتقال بسلاسة بينها يمثل تحديًا: “نستخدم ميرو للعصف الذهني وإنشاء نماذج أولية منخفضة الدقة، وفigma للنماذج عالية الدقة. ومع ذلك، فإن نقل الإطارات السلكية من ميرو إلى Figma ليس بالأمر السهل.” [P10] أشار بعض المشاركين بشكل خاص إلى أن أدوات التصميم الحالية تفتقر إلى الدعم لأبحاث المستخدم: “شيء ذو قيمة سيكون القدرة على إجراء اختبارات قابلية الاستخدام وتوليد البيانات داخل أداة التصميم نفسها.” [P17].
5.2.3 القيود 3: نقص الدعم المتقدم في التصميم. أعرب المشاركون عن عدم رضاهم عن أدوات التصميم الحالية بسبب نقص الدعم المتقدم للرسوم المتحركة المعقدة، والتصاميم متعددة المنصات، وتصميم الجرافيك، وأنظمة التصميم الأكثر ذكاءً: “هناك أشياء من نوع الرسوم المتحركة الدقيقة التي لا تحصل عليها من Figma، مثل الجداول الزمنية التي ستستخدمها في After Effects أو Principle لضبط الرسوم المتحركة الخاصة بك.” [P17] “تفتقر أنظمة التصميم إلى الذكاء. سأكون ممتنًا لنظام تصميم يوصي بمكونات تصميم مناسبة ويضبط هذه المكونات تلقائيًا لتناسب الشاشات المختلفة.” [P5].

5.3 الموضوع 2: التحديات في تصميم تجربة المستخدم الحالي

5.3.1 التحدي 1: إدارة توقعات أصحاب المصلحة والحصول على دعمهم لتجربة المستخدم.. التحدي الأكثر ذكرًا هو أن مصممي تجربة المستخدم غالبًا ما يقضون أكثر من نصف وقتهم في التواصل والتعاون مع مجموعة متنوعة من أصحاب المصلحة. يشمل ذلك العمل مع البائعين الخارجيين الذين يستغرقون وقتًا للتكيف مع ثقافة العمل وأسلوبه (P13)، وقضاء وقت مفرط في الحفاظ على توافق أصحاب المصلحة، والتنسيق عبر المناطق الزمنية، وإقناعهم بضرورة تجربة المستخدم والحصول على دعم من أصحاب المصلحة على مستوى الإدارة (P1، P8، P12). بالإضافة إلى ذلك، يجب عليهم موازنة توقعات أصحاب المصلحة والعملاء مع احتياجات المستخدمين للتوصل إلى حل مثالي (P18). ذكر P12، “يجب أن أقنع الآخرين بأن تجربة المستخدم ضرورية وأقوم بالكثير من العمل مع المطورين لضمان تنفيذ التصميم بشكل مناسب. تأخذ بعض المنظمات تجربة المستخدم بشكل خفيف كأنها مهمة تجميلية وتمنح المزيد من الدعم للواجهة الخلفية. لذا، نحتاج إلى تعليمهم باستمرار حول أهمية تعزيز تجربة المستخدم.” تعتمد هذه العمليات الإدارية، والتواصل مع أصحاب المصلحة، ومحاذاة التوقعات بشكل كبير على التواصل البشري الحيوي، وهي منطقة لا تدعمها تقنيات GenAI حاليًا.
5.3.2 التحدي 2: فهم المستخدمين النهائيين. ذكر المشاركون أن إجراء أبحاث لفهم احتياجات المستخدمين يمثل تحديًا من حيث اكتشاف النقاط المؤلمة الصحيحة للمستخدمين وبناء المنتجات الصحيحة: “أعتقد أن التحدي يكمن دائمًا في محاولة فهم من أصمم له، وما هي احتياجاتهم، وأهدافهم، ومشاكلهم، وما إذا كان المنتج أو الخدمة التي أعمل عليها تتناسب مع المشهد الأوسع.”[P17]. يكمن مثل هذا التحدي في قلب
تصميم تجربة المستخدم والبحث [12]. بينما تهدف بعض الأساليب إلى محاكاة المستخدمين الاصطناعيين والمحادثات (على سبيل المثال، استخدام GPT-3 [74])، من غير المحتمل أن تحل هذه حاليًا محل الفهم العميق للمستخدمين النهائيين من قبل المصممين.
5.3.3 التحدي 3: تصميم حلول تلبي كل من قيمة الأعمال واحتياجات المستخدمين. أشار المشاركون إلى أن التوصل إلى حلول تصميم جيدة توازن بين احتياجات العملاء، واحتياجات المستخدمين، وقيمة الأعمال، والجدة البصرية يمثل تحديًا: “مجال تجربة المستخدم ناضج تمامًا، مع وجود مراجع تصميم وأنظمة تصميم قائمة تساعد في العمل. التحدي هو التوصل إلى أفكار جديدة باستمرار وإيجاد توازن عند النظر في كل من مراجع التصميم المعتمدة، وتوقعات أصحاب المصلحة الداخليين، واحتياجات المستخدمين، ودمجها في التصميم الجديد.”[P18] يمكن أن تحمل تلبية احتياجات الأعمال تداعيات على الأصالة، وحقوق الطبع والنشر، والملكية، وكلها تؤثر على التبني (انظر، القسم 5.6.1).
5.3.4 التحدي 4: نقص المعرفة والموارد في المجال. ذكر المشاركون أن نقص الخبرة في المجال والموارد للوصول إلى المعرفة يمثل تحديًا كبيرًا: “أحد التحديات لمصممي تجربة المستخدم هو اكتساب معرفة واسعة في المجال، مثل إنشاء تصاميم مخصصة لمجالات متخصصة مثل الطب، والسيارات، والتكنولوجيا المالية، وما إلى ذلك.”[P1]; “في شركة ناشئة، غالبًا ما نواجه قيودًا في الموارد تحد من وصولنا إلى الخبراء، وإجراء أبحاث المستخدم أو اختبارات القابلية للاستخدام على نطاق أوسع، وهو أمر حاسم لضمان جدوى وجاذبية المنتجات.”[P6]. نعتقد أن هذه قد تكون طريقًا واعدًا يمكن أن تمكّن فيه أدوات GenAI، وخاصة نماذج اللغة الكبيرة المعتمدة على النص مثل ChatGPT، المصممين. على الرغم من الهلاوس الحالية للنموذج (انظر، [46])، يمكن للمصممين الوصول الفوري والواسع والسهل إلى المعلومات والموارد الخاصة بالمجال.

5.4 الموضوع 3: مواقف مصممي تجربة المستخدم تجاه GenAI

اتفق جميع المشاركين على أن أدوات GenAI يمكن أن تكون قوية ومريحة في دعم المهام العامة مثل توليد الأفكار في مراحل التصميم المبكرة، وكتابة تجربة المستخدم الأساسية، والترميز الأساسي. توفر هذه الأدوات نقطة انطلاق للمصممين البشريين للبناء عليها، مما يساعدهم على التغلب على الخوف من البدء من الصفر. بينما كان العديد من المشاركين متفائلين بشأن إمكانيات أدوات GenAI، متخيلين هذه الأدوات كوسيلة لتعزيز كفاءة التصميم، أعرب بعض الآخرين عن مشاعر مختلطة تجاه GenAI. من ناحية، اعترفوا بأن الذكاء الاصطناعي يمكن أن يساعد في المهام المتكررة، لكنهم أكدوا أن تصميم تجربة المستخدم هو بطبيعته موجه نحو المستخدم. موثوقية التصاميم التي ينتجها الذكاء الاصطناعي مشكوك فيها، حيث قد لا تتماشى دائمًا مع احتياجات وأهداف المستخدمين. كان الاعتماد بشكل مفرط على الذكاء الاصطناعي لتصميم المنتجات أو الواجهات دون مدخلات بشرية يُعتبر مخاطرة. أكد جميع المشاركين على أن الجودة الحالية لمخرجات الذكاء الاصطناعي، مثل دقة النص، ودقة الصورة، وتفاصيل التصميم والوظائف، لا تزال تتطلب مدخلات ومراجعة بشرية كبيرة، مما قد لا يوفر الوقت بالضرورة.
اعتقد جميع المشاركين أن المصممين البشريين يمتلكون قدرات فريدة، حيث يظهرون تعاطفًا تجاه المستخدمين ويستمدون المتعة من عملية التصميم، حتى عند بدء توليد الأفكار من الصفر: “يمكن أن تساعد أدوات GenAI اليوم في توليد أفكار أساسية لمساعدتنا في ملء القماش الفارغ، مما يساعد في التغلب على الخوف من ‘القماش الفارغ’. إنه مشابه عندما نحتاج إلى كتابة تقارير؛ يمكنك البدء بهيكل أساسي تم إنشاؤه بواسطة ChatGPT ثم البناء عليه. ومع ذلك، أخشى أنه على المدى الطويل، قد أصبح أكثر من محرر، وأفقد تدريجيًا قوتي التصميمية لملء القماش الفارغ وأن أكون متعاطفًا تجاه المستخدمين.”[P8].
5.4.1 العيوب المتصورة لأدوات GenAI التي تم عرضها في العرض التقديمي. كانت هناك عيبان يُعتبران رئيسيين بين المشاركين:
(1) العملية والكفاءة. أشار المشاركون إلى أن أدوات GenAI المعروضة في العروض التقديمية عامة جدًا، ولا يزال يتطلب عملية التصميم قدرًا كبيرًا من العمل اليدوي. يشمل ذلك مهامًا مثل إعادة رسم الرموز والأزرار وإدخال نصوص دقيقة. “تم تدريب المصممين على التفكير بصريًا. تتضمن عملية التصميم لدينا
تجربة وخطأ على دفاتر الرسم. تظهر التصاميم من العديد من تجارب الرسم، وليس من النصوص. [في العرض التقديمي]، يُطلب منك التبديل بين أدوات GenAI المختلفة، وإضافات Photoshop، وأداء قدر كبير من العمل اليدوي، بما في ذلك إعادة الرسم وإعادة توليد الصور عالية الدقة. أشك في ما إذا كانت هذه الأدوات GenAI تعزز حقًا سير العمل في التصميم.”[P12]
(2) الدقة والموثوقية. تساءل المشاركون عن الدقة العامة والموثوقية للنتائج التي تنتجها أدوات الذكاء الاصطناعي. يعتقدون أن المدخلات البشرية ضرورية للتحقق من النتائج التي ينتجها الذكاء الاصطناعي. شارك P16: “بالنسبة للتصميم البصري، لا تزال أدوات GenAI محدودة جدًا في التقاط الفروق الدقيقة والتفاصيل المعقدة. فيما يتعلق بتوليد الأفكار، قد تفشل أدوات الذكاء الاصطناعي مثل Miro AI في تمثيل رؤى أصحاب المصلحة الذين يمتلكون خبرة واسعة، لنقل، 20 عامًا في مجال متخصص مثل الرعاية الصحية.” [P16]. وصفت P19 تجربتها السلبية في استخدام أدوات GenAI: “بالنسبة لـ Midjourney، وجدت نفسي غالبًا أقضي ساعات في تعديل نصوصي، ومع ذلك لم أتمكن من تحقيق النتائج المتوقعة. أما بالنسبة لـ Miro، فإن البطاقات تتجمع فقط بناءً على الكلمات الرئيسية، مما يتطلب جهدًا يدويًا لإنشاء تجمعات متماسكة.” [P19]. علاوة على ذلك، أعرب P3 عن شكوكه بشأن موثوقية قدرة الذكاء الاصطناعي على توليد الشفرات: “لقد حاولت فعليًا أن أجعل ChatGPT يولد شفرات لبعض الميزات أو الوظائف، لكنه لم ينتج النتائج التي كنت أبحث عنها.”[P3].

5.5 الموضوع 4: التعاون بين البشر والذكاء الاصطناعي

وصف جميع المشاركين أدوات GenAI بأنها مساعدون أو معاونون وأوصوا بأن “الزيادة” و”التحسين” هما كلمتان جيدتان لوصف المساعدة التي تقدمها GenAI في تصميم تجربة المستخدم.
5.5.1 تصورات التهديدات والتعاون مع GenAI. أعرب المشاركون عن مستويات متفاوتة من القلق بشأن التهديدات المحتملة لـ GenAI. شعر البعض أنه إذا ظلت GenAI أداة عامة تحت سيطرتهم، فلن تكون مصدر قلق، بينما كان الآخرون قلقين بشأن قدرتها على محاكاة تفكير المصممين. شملت التوصيات التعامل مع GenAI كعضو في الفريق، مع التأكيد من بعض المشاركين على أهمية احتفاظ البشر بسلطة اتخاذ القرار النهائية في التعاون بين البشر والذكاء الاصطناعي. علق P18: “إذا وصلت التكنولوجيا في النهاية إلى نقطة يمكنها فيها محاكاة مبررات المصممين بدقة، فقد أبدأ في القلق بشأنها.” [P18]. أوصى P19 بمعاملة GenAI كعضو في الفريق: “أتصور الذكاء الاصطناعي كجزء من الفريق، يساهم في العملية العامة، ولكن ليس مسؤولاً وحده عن النتيجة النهائية.” [P19]. أكد P8 على أنه يجب أن يكون للبشر السلطة النهائية في اتخاذ القرار في التعاون بين البشر والذكاء الاصطناعي: “يجب أن يعتمد الأمر على الناس لاختيار الجزء الذي أريد فيه مساعدة الذكاء الاصطناعي، بغرض تسريع المهام. لا أحب كلمة ‘استبدال’. أستمتع كثيرًا بتصميم واجهة المستخدم وتجربة المستخدم… أريد أن يكون الذكاء الاصطناعي متدربي. يقوم البشر بالأجزاء الإبداعية والمعقدة، ويملأ متدربو الذكاء الاصطناعي الألوان أو يرسمون الخطوط.” [P8].
5.5.2 يمكن استبدال العمل العام والمتكرر بـ GenAI. شارك جميع المشاركين الاعتقاد بأن GenAI لديها القدرة على استبدال المهام المتكررة، مثل إنشاء نصوص “لوريم إيبسوم” لنماذج التصميم، وتلخيص موجزات التصميم والوثائق الفنية، وصياغة مكونات تصميم متكررة، وإنشاء تصميمات استجابة لشاشات مختلفة، وتوثيق عناصر التصميم المرئية تلقائيًا في أنظمة التصميم، وإنشاء نماذج أولية منخفضة الدقة، وتوليد تنويعات تصميمية للإلهام (انظر الشكل 3: “يمكن لـ GenAI”). كما قال P8، “أتصور الذكاء الاصطناعي كأداة إنتاجية تساعدني بسرعة في المهام المحددة. على سبيل المثال، يمكن للذكاء الاصطناعي تلخيص PRDs [وثائق متطلبات المنتج] أو شرح الخلفيات الفنية بلغة بسيطة. حاليًا، تتطلب هذه الجوانب جهدًا كبيرًا من مديري المنتجات والمهندسين لنقلها لنا كمصممين.” [P8]. أعرب P12 عن أمله في تصميم استجابة مدعوم بالذكاء الاصطناعي، “آمل أن تساعد GenAI في مهام مثل التصميم الاستجابي لشاشات مختلفة واقتراح تدفقات المستخدم عبر الصفحات، وليس فقط على صفحة واحدة.” [P12] أعرب العديد من المشاركين الآخرين عن رغبتهم في دعم GenAI في مهام متنوعة، بما في ذلك توفير الإلهام التصميمي (P2 وP4 وP8)، بالإضافة إلى تسهيل الكتابة الفعالة لتجربة المستخدم (P18 وP19).
شارك P19: “في شركتنا الناشئة، ليس لدينا دور مخصص لكتابة تجربة المستخدم. غالبًا ما يستخدم مصممونا ChatGPT لتقييم ملاءمة محتوى تجربة المستخدم في تصميمنا.” [P19].
5.5.3 لا يمكن استبدال عمل تصميم تجربة المستخدم بـ GenAI. أوضح العديد من المشاركين أن الكثير من عمل تصميم تجربة المستخدم لا يمكن ببساطة استبداله بالذكاء الاصطناعي، بما في ذلك العمل الذي يتطلب التواصل بين أصحاب المصلحة البشريين، والتعاطف مع احتياجات المستخدمين، وتصميم وظائف تركز على المستخدم. وأكدوا على أهمية كون البشر هم المبدعون الأصليون والمحققون لنتائج الذكاء الاصطناعي (انظر الشكل 3: “لا يمكن لـ GenAI”).
(1) مدخلات المستخدمين والتواصل أو التعاون بين البشر. أعرب المشاركون بثقة عن أنهم لا يرون حاليًا GenAI كمنافس، حيث يعمل المصممون البشر كوسطاء لاحتياجات المستخدمين ويستمرون في كونهم صانعي القرار النهائيين ووسطاء توافق الذكاء الاصطناعي. وأكدوا أن GenAI لا يمكن أن تحل محل العمل الذي يتطلب مدخلات المستخدم، والتعاون بين البشر: “قد تكون GenAI قادرة على تولي المهام الأساسية والمتكررة والمباشرة، لكنها لا يمكن أن تحل محل تصميم الخدمة والجهود التعاونية التي تتطلب تفكيرًا أكثر نظامية وتجريديًا.” [P6]. أشار P14 أيضًا إلى أن “تصميم تجربة المستخدم ليس مجرد جمالية، بل أيضًا قابلية الاستخدام والوصول إلى العناصر. المستخدمون الحقيقيون ضروريون لتحسين التصميمات من خلال البحث الفعلي واختبار المستخدمين.” [P14]. اعتقد P16 أن التواصل البشري لا يمكن استبداله بالذكاء الاصطناعي، قائلاً “أقدر الأفكار من زملائي البشر، وتجاربهم، وأفكارهم، والمعرفة التي يجلبونها إلى الطاولة. لا يجب أن تكون هذه مثالية، لكنها تساعدنا في رسم الاتجاهات معًا. لا أعتقد أن الذكاء الاصطناعي يمكنه أداء هذا الجانب من العمل.” [P16].
(2) الإبداع البشري واتخاذ القرار. أشار بعض المشاركين إلى أن الإبداع البشري والأصالة هما صفات بشرية قوية يصعب استبدالها بالذكاء الاصطناعي. تعتمد نتائج الذكاء الاصطناعي بشكل كبير على البيانات السابقة، مما قد يؤدي إلى نتائج متكررة بسبب محدودية مجموعة البيانات. لا يزال المصممون البشر هم القوة الدافعة القادرة على فهم سياقات التصميم بعمق، والتعاطف مع المستخدمين، والابتكار بناءً على أحدث المعرفة: “يجب ألا يتم استبدال العمل الإبداعي الذي يستمتع البشر بالقيام به بالذكاء الاصطناعي. الذكاء الاصطناعي هو نموذج مدرب يعتمد على الماضي، بينما الإبداع البشري والأصالة أكثر توجهاً نحو المستقبل.” [P8]; “يمكن دعم الإبداع بواسطة الذكاء الاصطناعي، حيث يمكنه إعادة مزج وتجديد بناءً على البيانات الإبداعية المدخلة من البشر، لكن البشر يمتلكون الأصالة.” [P4].
أكد العديد من المشاركين أن اتخاذ القرار النهائي لا يمكن تفويضه إلى GenAI. يُعتبر التحقق البشري ضروريًا لكل مرحلة من مراحل التصميم ومشاريع التطوير: “يجب استخدام الذكاء الاصطناعي بشكل متعمد، ومراقبته، والتحقق منه من قبل البشر. هم في الأساس نماذج لغوية كبيرة، وهي أجهزة كمبيوتر تم تدريبها على التحدث مثل البشر. جودة المحتوى والبيانات المقدمة من الذكاء الاصطناعي ليست بالضرورة مدعومة ببيانات بحث فعلية تم التحقق منها من قبل باحثين بشريين.” [P8].

5.6 الموضوع 5: المخاوف الأخلاقية والملكية

تُعتبر حقوق الطبع والنشر، والخصوصية، والتحيزات في البيانات والشفافية، وتدهور المهارات والبطالة، والفجوة الاجتماعية، والوصول المتساوي من المخاوف الأخلاقية التي تم ذكرها بشكل متكرر بين جميع المشاركين.
5.6.1 الأصالة وحقوق الطبع والنشر والملكية. أكد المشاركون على أهمية منح الفضل للمبدعين الأصليين للأعمال الفنية أو أنماط الفن المستخدمة في عمل تجربة المستخدم. اقترح P3 وP4 أن تقنية البلوكشين قد تكون مفيدة في ضمان نسب الملكية المباشرة للمبدعين الأصليين، بغض النظر عن مدى تعديل العمل الأصلي. قال P4: “يجب أن تظل الملكية قابلة للتتبع بغض النظر عن مدى تعديل أو اشتقاق أو إعادة إنتاج مخرجات الذكاء الاصطناعي من الأعمال الأصلية. ربما يمكن أن توفر تقنية البلوكشين حلاً هنا.” أعربت P14 عن مخاوفها بشأن سرقة التصميم، “من وجهة نظر السرية، ما البيانات التي يمكن وضعها في أدوات GenAI، وكيف يمكننا التأكد من أننا نستخدم مصادر قانونية وأننا لا ننتهك لوائح GDPR. هناك بالفعل حالات يقوم فيها المصممون بنسخ أعمال الآخرين لبناء محافظهم الخاصة. يمكن أن تجعل GenAI السرقة في التصميم أسوأ بكثير.” [P14].
توسيعًا لموضوع الملكية، اعتقد بعض المشاركين أنه يجب أن يكون البشر هم المالكون، مع كون أدوات GenAI تعمل كمعينات داعمة للبشر في تحقيق النتائج النهائية. علق P11: “الذكاء الاصطناعي ليس إنسانًا؛ الملكية تعود للبشر. الذكاء الاصطناعي أداة، مثل Figma. لا أعتقد أنه يجب أن تُنسب الملكية للتكنولوجيا أو الأدوات. إذا كنت أنت من يقوم بتحسين، وصياغة الطلب، والمساهمة بالإبداع والأفكار للذكاء الاصطناعي من أجل إنتاج العمل، ومراقبة العملية بأكملها، يجب أن تُنسب الملكية للبشر.” [P11].
ومع ذلك، أوصى العديد من المشاركين الآخرين بإعطاء ملكية الجزء من العمل المدعوم بالذكاء الاصطناعي أو تصنيفه كمساهمات من الذكاء الاصطناعي وضمان الشفافية بشأن الأدوات المستخدمة. قدم P18 تشبيهاً مناسباً، “يجب أن تكون نسبة الملكية لاستخدام الذكاء الاصطناعي في الأعمال المستقبلية مشابهة لكيفية تقديم الباحثين لنتائج أبحاثهم، مع ذكر المصادر والأنظمة المستخدمة كمراجع.” [P18] وقد تم تأكيد ذلك من قبل P19، “يجب أن تُعتبر الملكية جهدًا جماعيًا بمشاركة من مختلف أعضاء الفريق، بما في ذلك GenAI.” [P19].
5.6.2 تدهور المهارات والبطالة. أشار العديد من المشاركين إلى قلقهم من أن العمل البشري في تصميم تجربة المستخدم قد لا يُقدَّر كما ينبغي، مما قد يؤدي إلى شعور المصممين بالرضا عن النفس وتراجع مهاراتهم. وهم يخشون أن يؤدي استحواذ الذكاء الاصطناعي على المهام التي كانت تُعتبر أساسية للمصممين إلى تدهور المهارات. هذه المخاوف، التي ذُكرت بشكل أساسي من قبل المصممين الكبار (انظر الجدول 1)، ذات صلة خاصة بالمصممين المبتدئين، الذين قد يفوتون فرص تطوير مهاراتهم بشكل أكبر. وبالتالي، قد ينتقل المصممون البشريون إلى أدوار إدارية أو عامة أكثر. تصور P16 (مصمم كبير) هذا، حيث قال: “قد يقع المصممون في فخ الرضا عن النفس. قد يكون لدى المصممين المبتدئين فرص أقل لتعلم عملية التصميم، ولكن بدلاً من ذلك، قد يركزون على إتقان أنظمة الذكاء الاصطناعي. ونتيجة لذلك، قد يتطور المصممون إلى مصممين عامين.” [P16]
أبرز المشاركون بشكل خاص أن البطالة وتدهور المهارات قد تتطور إلى قضايا اجتماعية كبيرة إذا لم يتم التعامل معها بجدية من البداية. أعرب P4 (مصمم مبتدئ) وP5 (مصمم أول) عن قلقهما من أن مهارات المصممين البشريين قد تصبح ضيقة مع تزايد تخصصنا في مهام معينة إذا استحوذ الذكاء الاصطناعي على العديد من أجزاء العمل في سير العمل التصميمي، مما قد يحد من إبداعنا كمصممين: “يمكن أن يضعنا الذكاء الاصطناعي في فقاعة. نشعر أننا نقوم بعمل على مستوى أعلى، وأن حاجز المهنة للدخول في مجال التصميم أصبح أعلى. ولكن بطريقة ما، نحن مجبرون على أن نكون أكثر تخصصًا في نطاق محدود من المهام. قد يؤدي فقدان الإشراف على سير العمل في التصميم والتطوير إلى تعريض إبداعنا للخطر.” [P5].
عكس P12، الذي هو أيضًا مصمم كبير، بتفكير عميق تأثير GenAI على التعليم والمهن للمصممين المبتدئين، “تنضج خبرتنا من خلال الممارسة المستمرة على مر السنين، والتي تشمل ليس فقط مهارات التصميم ولكن أيضًا التفكير التصميمي والتعاطف الحقيقي مع المستخدمين. إذا قررت الشركات توظيف مشغلين أو مشغلين للذكاء الاصطناعي في المستقبل لاستبدال أدوار المصممين المبتدئين، كيف يمكن لهؤلاء المشغلين أو مشغلين الذكاء الاصطناعي الحصول على التدريب المناسب الذي يعادل ما نتلقاه في مدارس التصميم أو من خلال ممارسة التصميم؟ والأهم من ذلك، كيف يمكن للمصممين المبتدئين اكتساب مهارات جديدة أو تحسين خبراتهم إذا استحوذ الذكاء الاصطناعي على المهام التي يحتاجون إلى ممارستها للحفاظ على مهاراتهم في التصميم؟”[P12].
5.6.3 الخصوصية والشفافية والوصول المتساوي إلى GenAI. أشار المشاركون إلى أن الخصوصية تمثل مصدر قلق كبير في مجال GenAI. حيث إن هذه النماذج الذكية تستوعب كميات كبيرة من البيانات لتتعلم الأنماط وتولد المحتوى، هناك احتمال للتعرض غير المقصود لمعلومات حساسة، مثل التفاصيل الشخصية أو السرية التي لم يقصد المستخدمون مشاركتها: “إذا كنت أرغب في استخدام ChatGPT لاستخراج المعلومات من الوثائق التقنية، لا أعرف كيف يمكنني منع تسرب المعلومات الحساسة. بالنسبة لحياتي الشخصية أيضًا، ليس من الواضح كم من بياناتي الشخصية ستظل محفوظة لتدريب نماذج الذكاء الاصطناعي بشكل أكبر.”[P19]
أشار العديد من المشاركين إلى أهمية تدريب الذكاء الاصطناعي على بيانات غير متحيزة وضمان الوصول المتساوي إلى أدوات الذكاء الاصطناعي التوليدي. وذلك بسبب المخاوف المتعلقة بالفجوة الاجتماعية وإمكانية أن تحد سياسات الدول من قدرة الناس على التعلم والوصول إلى أدوات الذكاء الاصطناعي، فضلاً عن احتمال أن تؤدي البيانات المتحيزة إلى نتائج مشكوك فيها. “الشركات الكبرى و
“قد تكتسب الدول المتقدمة السيطرة على البيانات المدخلة في الذكاء الاصطناعي. كما أنه من الأسهل لها الوصول إلى أدوات الذكاء الاصطناعي التوليدي مقارنة ببعض الدول النامية.” [P6]. أضاف P2، “من المهم ضمان أن الأفراد الذين لديهم مستوى تعليمي أقل ولديهم معرفة محدودة بالذكاء الاصطناعي التوليدي يمكنهم استخدام هذه الأدوات لتعزيز عملهم وحياتهم اليومية.” [P2].

5.7 الموضوع 6: مستقبل الذكاء الاصطناعي التوليدي في تصميم تجربة المستخدم

الموضوع 6 قدم رؤى حول رؤى ورغبات المشاركين المتعلقة بدمج الذكاء الاصطناعي التوليدي في مجال التصميم.
5.7.1 تصور التصميم التعاوني مع GenAI. تخيل بعض المشاركين نظامًا مثلثيًا يتضمن التعاون ومدخلات من المستخدمين والمصممين وGenAI. تهدف هذه المقاربة إلى تقليل التحيزات وضمان عدم تجاهل الفروق الدقيقة الحاسمة، الضرورية للابتكارات التصميمية. كما ذكر P11، “قد يغفل المصممون البشر عن بعض الفروق الدقيقة التي قد تكون مهمة للمستخدمين. أتمنى أن يعزز GenAI هذا الجانب من خلال تقديم مدخلاته المستندة إلى بيانات ومعرفة واسعة من الماضي، مما قد يرفع تصميم تجربة المستخدم إلى المستوى التالي.” [P11].
5.7.2 تعزيز معرفة الذكاء الاصطناعي والوصول إليه للمصممين. أعرب عدد من المشاركين عن أملهم في زيادة معرفة الذكاء الاصطناعي من خلال التعليم، مما يعود بالفائدة على المصممين ذوي الخبرة والمبتدئين. أشار P13 (مصمم مبتدئ) إلى أن “الطلاب قد يشعرون بالقلق حيال هذه الثورة الكبيرة في الذكاء الاصطناعي. التعليم ضروري لإعداد المصممين المبتدئين والبارعين لتأثير الذكاء الاصطناعي وكيفية الاستفادة منه لتعزيز مهارات التصميم البشرية لدينا.” [P13]. بالإضافة إلى ذلك، يتصور عدد قليل من المشاركين أن أدوات الذكاء الاصطناعي المستقبلية ستكون أكثر سهولة للمصممين. بعبارة أخرى، يرغبون في أدوات ذكاء اصطناعي تقدم قابلية استخدام بصرية وبديهية أكبر، متجاوزة الاعتماد الوحيد على المدخلات النصية. “سماعة Apple Vision Pro الأخيرة ملهمة. تصميمها البصري والبديهي يبدو واعدًا. وبالتالي، يجب أن يتطور الذكاء الاصطناعي أيضًا ليصبح أكثر بصرية وبديهية ومتاحة عالميًا لجميع المستخدمين.” [P12].
5.7.3 الدور المرغوب فيه لـ Al في تعزيز عمليات التصميم. أعرب العديد من المشاركين عن رغبتهم في أن تتطور GenAI أكثر، لدعمهم في مهام التصميم المختلفة. كانوا يأملون أن تتمكن GenAI من المساعدة في تصميم الرسوميات، وكتابة التعليمات، وتوليد المكونات الذكية، وتقديم أمثلة تصميم ناجحة، وأتمتة نقل التصاميم إلى فرق التطوير. أشار P4 إلى: “أتوقع طرقًا أسهل لكتابة التعليمات أو توليد المكونات تلقائيًا التي تتكيف مع الشاشات المختلفة، وتوصي بانتقالات الرسوم المتحركة الذكية، وحتى توليد محتوى ثلاثي الأبعاد لبيئات غامرة في المستقبل.” [P4]. كما أعرب العديد من المشاركين الآخرين عن رغباتهم في دعم GenAI المحدد في مهام مختلفة في المستقبل. على سبيل المثال، كانوا يأملون أن تتمكن الذكاء الاصطناعي من تلخيص المعرفة المتطورة للبحث والاستراتيجية (P10 وP18 وP19)، وأن تعمل كمستخدم اصطناعي لتسريع السبرينت (P1 وP18)، وتقديم إرشادات مهنية في تصميم الوصول (P12)، ومساعدة المصممين على إتقان والتبديل بسرعة بين أدوات التصميم المختلفة (P18)، والمساعدة في تصنيف وتجميع البيانات الخام النوعية من دراسات المستخدمين (P5 وP9 وP10 وP12 وP18)، وترجمة التصاميم بسلاسة إلى التطوير (P5 وP18 وP19).
P18 سلط الضوء بشكل خاص على المشهد المتغير باستمرار لأدوات التصميم، معبراً عن رغبة في أن “تساعد GenAI في تقليل الوقت الذي يقضيه المصممون في مواكبة التكنولوجيا المتطورة بسرعة. تتغير أدوات التصميم كل بضع سنوات. إذا كان بإمكان الذكاء الاصطناعي مساعدة المصممين في الانتقال بسلاسة بين الأدوات القديمة والجديدة أو دمج كل شيء في أداة واحدة، فسيكون لدى المصممين المزيد من الوقت لفهم احتياجات المستخدمين والتصميم.” [P18]. عبر P5 عن أمنيته في أن “تساعدني أدوات GenAI في تقييم جدوى التصاميم من جانب التطوير خلال مرحلة التصميم المبكرة وتساعد في توثيق التصاميم في Figma، مما يسهل الانتقال السلس إلى فريق التطوير بصيغة يمكن الوصول إليها وفهمها من قبل المطورين. حالياً، لا يزال يتعين علي توثيق التصاميم يدوياً، وأخذ العديد من لقطات الشاشة، وشرحها شفهياً لفريق التطوير.” [P5].
5.7.4 تحدي تعريف مقاييس نتائج GenAI. وفقًا لعدة مشاركين، فإن رؤية مستقبلية أخرى لـ GenAI تتضمن تحدي تعريف مقاييس لقياس جودة المخرجات التي تنتجها الذكاء الاصطناعي. كما علق P17، “إذا استخدمت العديد من الشركات نفس أدوات GenAI، فقد يؤدي ذلك إلى تصميمات مشابهة ويفتقر إلى التنوع والابتكار. أيضًا، يمكن أن يؤدي الذكاء الاصطناعي إلى توليد العديد من المخرجات ويترك التحدي لنا، البشر، لاختيار من بينها.” [P17]. “إنه بالفعل تحدٍ لاختيار وتحسين نتائج الذكاء الاصطناعي. نحن بحاجة إلى إرشادات حول كيفية قياس أو التحقق من مخرجات الذكاء الاصطناعي.” [P18]

6 المناقشة

تشير نتائجنا الرئيسية إلى أن مصممي تجربة المستخدم ذوي الخبرة يعترفون بإمكانات GenAI كأداة مساعدة للمهام المتكررة وتعزيز الإنتاجية، ومع ذلك يظلون حذرين بشأن اعتماد أدوات GenAI بأنفسهم. ويرجع ذلك إلى قيودهم في فهم التصميم الموجه نحو المستخدم وتأثيرها المحتمل على مهاراتهم الإبداعية، بالإضافة إلى الحدود الأساسية للمتعة والقدرة على اتخاذ القرار في التعاون بين البشر والذكاء الاصطناعي. بينما يُنظر إلى GenAI على أنه مغير لقواعد اللعبة في معالجة التحديات داخل ممارسة تصميم تجربة المستخدم الحالية، مثل الاحتكاك في تسليم التصميمات والقيود في موارد تصميم تجربة المستخدم، أعرب المصممون عن مخاوف جدية بشأن تدهور المهارات، خاصة بالنسبة للمصممين المبتدئين. إنهم يدعون إلى تطور أدوات GenAI إلى أنظمة بديهية ومرئية وذكية تعاونية. يهدف هذا التطور إلى تحقيق توازن بين الإبداع البشري وكفاءة الذكاء الاصطناعي، مع التأكيد على الاستخدام المسؤول للذكاء الاصطناعي والتعاون بين البشر والذكاء الاصطناعي للحفاظ على السيطرة البشرية في الإبداع واتخاذ القرار. نتناول هذه الجوانب أدناه، حيث نناقش الآثار المترتبة على التعاون بين البشر وGenAI، وبشكل خاص حقوق الطبع والنشر والملكية، والإبداع البشري والقدرة على اتخاذ القرار، ومعرفة الذكاء الاصطناعي والوصول إليه.

6.1 المتعة والقدرة على اتخاذ القرار في التعاون بين البشر وGenAI

تعتبر المتعة والمعنى كلمتين رئيسيتين متكررتين في محادثاتنا مع مصممي تجربة المستخدم. حيث يسعى مصممو تجربة المستخدم ليس فقط للتفوق في عملهم ولكن أيضًا لاستخراج المتعة منه. وقد وجدت الأعمال الأخيرة أن تجربة المستخدم للمستخدمين الأوائل لـ ChatGPT تأثرت بالسمات الهيدونية، وليس فقط العملية [88]. يمكن النظر إلى المتعة في العمل من خلال عدسة نظرية الدافع، بما في ذلك الدوافع الخارجية والداخلية [48]. تدور الدوافع الخارجية حول ما إذا كانت التكنولوجيا تخدم بكفاءة كوسيلة لإنجاز المهام، مع اعتبار التكنولوجيا مثل GenAI كأداة. تركز الدوافع الداخلية على ما إذا كان مصممو تجربة المستخدم يجدون عملهم مجزيًا بطبيعته وأن التفاعل مع التكنولوجيا ممتع، مع اعتبار التكنولوجيا مثل GenAI كلعبة [8]. نحن نؤكد أنه في تصميم تجربة المستخدم، حيث تلعب الإبداع وحل المشكلات أدوارًا مركزية، تصبح هذه الدوافع الداخلية ذات أهمية قصوى. لا يُدفع المصممون فقط من خلال النتائج النهائية؛ بل يسعون أيضًا إلى الإشباع والمتعة طوال عملية التصميم. بعبارة أخرى، يجب أن يمكّن مستقبل التعاون بين الذكاء الاصطناعي والبشر الأفراد من التركيز على المهام التي تهمهم حقًا، مما يسمح للذكاء الاصطناعي بالتعامل مع الجوانب الأكثر رتابة. لا تعزز هذه الطريقة الإنتاجية فحسب، بل تعزز أيضًا رضا العمل [59].
مرتبطًا ارتباطًا وثيقًا بالمتعة، وجدنا أن القدرة على اتخاذ القرار في التعاون بين البشر والذكاء الاصطناعي هي أيضًا موضوع متكرر. هل سيتولى الذكاء الاصطناعي مزيدًا من القدرة على اتخاذ القرار بينما يلعب البشر دورًا إشرافيًا؟ أم سيحتفظ البشر بالسيطرة الأساسية ولديهم القدرة على تجاوز قرارات الذكاء الاصطناعي عند الضرورة؟ بدلاً من ذلك، هل سيتعاون الذكاء الاصطناعي والبشر في نهج مختلط للقدرة على اتخاذ القرار، حيث يساهم كلاهما في اتخاذ القرار والتنفيذ [91]؟ يعتقد مصممو تجربة المستخدم أنه يجب علينا التمييز بين المهام الروتينية المحددة جيدًا والمهام المعقدة التي تتطلب الإبداع وحل المشكلات. يجلب البشر الفهم السياقي، والحكم الأخلاقي، والمساءلة، ويجب أن يقودوا العمل المعقد الذي يتطلب الإبداع، والتعاطف، والأخلاق، واتخاذ القرارات المعقدة. يجب أن تكون المهام التي يقودها الذكاء الاصطناعي والقرارات التي يتخذها الذكاء الاصطناعي شفافة (انظر، أنظمة الذكاء الاصطناعي المضللة [105])، وقابلة للتفسير [21]، وسهلة التعديل من قبل المتعاونين البشريين. السيناريو المثالي يتضمن تعلم البشر من
رؤى الذكاء الاصطناعي، بينما يتحسن الذكاء الاصطناعي بناءً على ملاحظات البشر (انظر على سبيل المثال، [30] لمراجعة التحديات المتعلقة بالتعلم المعزز من ملاحظات البشر). بشكل عام، وجدنا أن معتقدات مصممي تجربة المستخدم تتماشى مع تلك الخاصة بـ Fanni وآخرين [26]، الذين يدعون إلى “القدرة النشطة للبشر” ويؤكدون أن سياسات الذكاء الاصطناعي يجب أن تتضمن أحكامًا تمكّن المستخدمين من تحدي أو تصحيح القرارات المدفوعة بالذكاء الاصطناعي بسرعة (انظر، الذكاء الاصطناعي القابل للتحدي [66]). تهدف هذه الطريقة إلى تمكين الاستقلالية الفردية وتعزيز حقوق الإنسان الأساسية في العصر الرقمي. نجد أن الأبعاد السابقة للمتعة والقدرة على اتخاذ القرار ليست فقط عوامل بشرية فريدة، ولكننا نشك في أنها ستلعب دورًا متزايدًا في مدى احتفاظ مصممي تجربة المستخدم باستخدام مثل هذه الأدوات.

6.2 GenAI يتداخل، البشر يستنتجون، وإرهاق الإبداع

في ثقافة العمل السريعة المدفوعة بالكفاءة اليوم، يعد تصميم تجربة المستخدم أمرًا أساسيًا للحصول على التصميم الصحيح والتصميم المناسب [12]. يجمع فريقًا للمرور عبر كل دورة، متقدمًا من فكرة إلى نموذج أولي إلى بحث المستخدم ضمن أطر زمنية قصيرة [18]. إن السرعة الملحوظة لـ GenAI في توليد المحتوى والحلول لديها القدرة على تسهيل الدورات السريعة المتحركة. ومع ذلك، يجب أن نكون واعين أنه بينما يمكن لـ GenAI إنتاج كميات هائلة من المخرجات في لحظة، فإن البشر هم المبتكرون وراء هذه الإبداعات. في الأساس، يقوم الذكاء الاصطناعي بالتداخل، وتقدير القيم أو توليد النتائج ضمن نطاق البيانات الموجودة. يتناول Thoring وآخرون [93] هذا من منظور نظرية الإبداع التطوري [87]، حيث تعتبر التباين والاختيار مكونات رئيسية. بينما يقوم GenAI بالتداخل (التباين)، فإنه يفتقر حاليًا إلى القدرة على الاستنتاج من الفضاء الكامن والاستمرار في أراضٍ غير معروفة أو الاندماج في فضاء كامن آخر دون السيطرة البشرية الإشرافية (الاختيار) [93]. في الواقع، قد يتفوق الذكاء الاصطناعي في “العمل المحيطي” [83] أو المهام المتكررة ضمن أنماط البيانات المعروفة. ومع ذلك، نظرًا لأن الذكاء الاصطناعي يتم تدريبه باستخدام بيانات معروفة، فإن الاعتماد بشكل كبير على الذكاء الاصطناعي يمكن أن يؤدي إلى حلول موحدة (على سبيل المثال، جميع واجهات المستخدم التي تم إنشاؤها بواسطة الذكاء الاصطناعي تبدو متشابهة) [6] أو تضخيم التحيزات في البيانات (على سبيل المثال، التصميم الناتج يفضل مجموعات ديموغرافية معينة [4، 19، 25]. بالمقابل، اعتبر الأطفال، الذين يتلقون حوالي أربعة أو خمسة أوامر من حيث البيانات اللغوية أقل من نماذج اللغة الكبيرة، ويتفوقون بشكل كبير على قدرات الذكاء الاصطناعي [31]. في الواقع، نحن البشر نتفوق في الاستنتاج، ونقوم بعمل توقعات تتجاوز نطاق البيانات المعروفة باستخدام ليس فقط إبداعنا، وحدسنا، ومعرفتنا بالمجال، ولكن أيضًا نستفيد من الأسس متعددة الوسائط والطبيعة الاجتماعية والتفاعلية للمدخلات الحسية المستلمة [31]. في جوهره، يتفوق البشر في “العمل الأساسي” الذي يتعرفون عليه، مما يساهم ليس فقط في نجاحهم، ولكن أيضًا في سعادتهم [83].
تدور عقبة محتملة أخرى حول خطر استنفاد الإبداع البشري في مواكبة وتيرة إنتاج الذكاء الاصطناعي. يمكن أن تمارس سرعة الذكاء الاصطناعي المتواصلة ضغطًا هائلًا على المبدعين البشريين. بينما لا يتعرض الذكاء الاصطناعي لعوامل جسدية مثل التعب، يُتوقع من البشر الابتكار باستمرار وبسرعة على نطاق واسع. من ناحية، تضع سرعة الذكاء الاصطناعي معيارًا صعبًا للمبدعين البشريين؛ ومن ناحية أخرى، فإنها تخفض أيضًا متطلبات التدريب لتصبح فنانًا أو مبدعًا، حيث يمكن للعديد من الأشخاص إنشاء لوحات فنية دون الحاجة إلى الإمساك بفرشاة. علاوة على ذلك، مع تعلم الذكاء الاصطناعي من الإبداعات البشرية وصقل قدراته، يرفع المعايير لما يُعتبر مبتكرًا. قد يشعر المبدعون البشريون بشكل عام، ومصممو تجربة المستخدم بشكل خاص، بأنهم مضطرون لدفع حدودهم للتنافس مع الذكاء الاصطناعي، مما يمكن أن يكون مرهقًا عقليًا وعاطفيًا. للحفاظ على الإبداع البشري، يبدو أن المبدعين البشريين سيكونون أفضل حالًا إذا ركزوا على المهام التي تتطلب التفكير النقدي والعمق العاطفي، حيث يكون اللمسة البشرية لا يمكن استبدالها. لتحقيق ذلك، يجب تقديم أنظمة دعم وتعليم لتخفيف الضغط على البشر والمساعدة في التخفيف من خطر استنفاد الإبداع، وهو ما سنناقشه لاحقًا.

6.3 محو الأمية في الذكاء الاصطناعي والذكاء الاصطناعي التشاركي في تصميم تجربة المستخدم

في تصميم تجربة المستخدم، هناك تزايد في وجود أدوات تحتوي على ميزات الذكاء الاصطناعي أو تدعم إضافات الذكاء الاصطناعي. يتم دمج الذكاء الاصطناعي تدريجياً في مراحل مختلفة من سير عمل تجربة المستخدم. تبدأ معرفة الذكاء الاصطناعي بفهم أساسي للذكاء الاصطناعي.
المفاهيم، بما في ذلك كيفية عمل التعلم الآلي، وأنواع أنظمة الذكاء الاصطناعي، وقدراتها، وقيودها، وآثارها. هذا الفهم ضروري لمصممي تجربة المستخدم للاستفادة من قوة الذكاء الاصطناعي، وضمان الوصول المتساوي إلى التعليم والأدوات المتعلقة بالذكاء الاصطناعي، والمشاركة بنشاط في تشكيل وتطوير تقنيات الذكاء الاصطناعي، خاصة تلك المستخدمة في عمليات التصميم الخاصة بهم. من خلال اعتماد نهج الذكاء الاصطناعي التشاركي، يصبح مصممو تجربة المستخدم جزءًا لا يتجزأ من الحوار حول مستقبل تطوير الذكاء الاصطناعي التوليدي – مما يمكّن مصممي تجربة المستخدم من تعزيز ملاحظاتهم وخبراتهم، وضمان توافق أدوات الذكاء الاصطناعي بشكل فعال مع النوايا البشرية وممارسات تصميم تجربة المستخدم.
عنصر أساسي آخر من محو الأمية في الذكاء الاصطناعي هو أنه يجب على مصممي تجربة المستخدم أن يكونوا قادرين على التفكير النقدي والتشكيك في المخرجات التي ينتجها الذكاء الاصطناعي. هناك حاجة إلى إطار شامل أو إرشادات لتقييم نتائج الذكاء الاصطناعي. يمكن أن يشمل هذا الإطار تقييم مصداقية الذكاء الاصطناعي، وأدائه، ودقته، وتأثيره على تجارب المستخدمين [89]، بالإضافة إلى توفير وسيلة هيكلية للطعن [66]. ما سبق يتردد صدى الجهود الأخيرة [50] لضمان تشكيل لجان تتكون من خبراء في الذكاء الاصطناعي، وخبراء في الأخلاقيات، والأهم من ذلك، محترفين مبدعين، للعمل معًا للإشراف على تقدم الذكاء الاصطناعي التوليدي. ستكون هذه المجموعات الأكثر تجهيزًا لتقديم معايير وأفضل الممارسات لتقييم مخرجات الذكاء الاصطناعي التوليدي والتخفيف من الأضرار على المحترفين المبدعين (انظر، [50]).
في دراستنا، أكد مصممو تجربة المستخدم على أهمية منح الفضل للمبدعين الأصليين للأعمال الفنية أو الأنماط المستخدمة في الأعمال التي تعتمد على الذكاء الاصطناعي. وقد ذكروا استخدام تقنية البلوكشين (انظر، [24])، حيث تبرز آليات الأصل الحسابية [94] التي تتعقب وتوثق أصول ومراحل وتحولات المحتوى أو القطع الفنية التي تم إنشاؤها بواسطة الذكاء الاصطناعي. وبالتالي، يتضمن ذلك تسجيل معلومات حول كيفية إنشاء قطعة من المحتوى، مثل النصوص أو الصور أو وسائل الإعلام الأخرى، والنماذج والخوارزميات التي تم استخدامها، وأي مصادر بيانات ساهمت في إنشاء المحتوى. من خلال ذلك، سيسهل هذا نسب الملكية إذا استندوا إلى مخرجات النموذج طوال ممارستهم لتجربة المستخدم. وجدنا أن العديد من مصممي تجربة المستخدم في دراستنا يعتقدون أن البشر يجب أن يكونوا مالكي نتائج التصميم. ومع ذلك، كان هذا صحيحًا فقط إذا كانت أدوات الذكاء الاصطناعي تعمل كأدوات مساعدة، وظل المصممون البشر في المقعد الأمامي من خلال قيادة العملية بأكملها، وإضافة لمسات تفصيلية ودمج مدخلات المستخدمين. ومع ذلك، أوصوا بنسب الملكية إلى الجزء من العمل المدعوم بالذكاء الاصطناعي أو وسمه كمساهم من الذكاء الاصطناعي مع ضمان الشفافية حول الأدوات المستخدمة. ومن المثير للاهتمام، في دراسة حديثة، وجد دراكسلر وآخرون [20] أن الناس غالبًا لا يكشفون عن الذكاء الاصطناعي كمؤلف عند استخدام نصوص مخصصة تم إنشاؤها بواسطة الذكاء الاصطناعي في الكتابة، على الرغم من أنهم ينسبون الملكية إلى الذكاء الاصطناعي. عند مقارنة الكتابة البشرية وكتّاب الظل من الذكاء الاصطناعي، وجدوا أن نسب التأليف إلى النفس أكثر شيوعًا عند الكتابة بدعم من الذكاء الاصطناعي. نظرًا للاختلاف في نسب الملكية، من الضروري تطوير إطار عمل لنسب التأليف يأخذ في الاعتبار المستخدم وعلاقته بالنموذج التوليدي.
إن الخطاب حول حقوق الطبع والنشر، والخصوصية، والملكية في ممارسة تصميم تجربة المستخدم مرتبط بشكل جوهري بالخطاب الأوسع حول الذكاء الاصطناعي التوليدي ككل. سواء كان الأمر يتعلق بالنصوص أو الصور أو الفيديو، فإن ما سبق يثير تساؤلًا حول ما إذا كانت المخرجات الناتجة عن الذكاء الاصطناعي التوليدي أخلاقية ومتوافقة قانونيًا في البداية. في الواقع، العمل الأخير الذي قام به جيانغ وآخرون [50] في تقييم تأثير الذكاء الاصطناعي البصري على الفنانين يرسم صورة قاتمة بشأن الأضرار المعلنة على الفنانين، والتي يمكن أن تمتد إلى أعمال إبداعية أخرى (بما في ذلك أعمال مصممي تجربة المستخدم). على سبيل المثال، قد لا تنتهي “الاستخدام العادل” ضمن قانون حقوق الطبع والنشر الأمريكي دائمًا بحماية الفنان، سواء بسبب التحديد من حالة إلى أخرى، أو التكلفة العالية في متابعة المعارك القانونية. سواء كانت الطريقة المستقبلية تتضمن تحسين أساليب العلامات المائية لضمان حقوق الطبع والنشر والمساءلة القابلة للتتبع [94، 107]، أو التطوير الإضافي للتنظيم والسياسة على مستوى (دولي) أو (وطني) (مثل قانون الذكاء الاصطناعي الأوروبي. من الواضح أن تطوير الذكاء الاصطناعي العام يحتاج إلى أن يتم التصرف بشأنه بشكل مسؤول – من أجل وما يتجاوز ذلك.
ممارسة تصميم تجربة المستخدم. وبالتالي، نعتقد أن الاستناد إلى المبادئ الأساسية المتمحورة حول الإنسان (مثل، التوافق مع رفاهية الإنسان، واحترام الخصوصية، ووجود حوكمة ورقابة مناسبة) في تطوير الذكاء الاصطناعي المسؤول [34] هو أمر ضروري على الفور لحماية المصممين في المستقبل.

6.5 مستقبل تصميم تجربة المستخدم المدعوم بالذكاء الاصطناعي: المخاوف والفرص

تؤكد التقدم السريع في الذكاء الاصطناعي التوليدي وتصميم تجربة المستخدم على ضرورة المراقبة المستمرة والبحث في كلا المجالين. مع استمرار تطور أدوات الذكاء الاصطناعي التوليدي، من الضروري التأكد من أنها تتماشى مع القيم الإنسانية والأخلاقيات والعمليات الإبداعية، على الرغم من أي عدم يقين قد ينشأ من القضايا التجريبية والمفاهيمية غير المحلولة حاليًا في علم النفس البشري. من أجل ذلك، من خلال الإجابة على أسئلة بحثنا حول التصورات الحالية (RQ1) والمستقبلية (RQ2) للذكاء الاصطناعي التوليدي، تعتبر هذه الدراسة أساسًا لاستكشاف الفرص والتحديات التي يقدمها الذكاء الاصطناعي التوليدي في تصميم تجربة المستخدم، حيث نؤكد على أهمية التعاون المسؤول والمستنير بين البشر والذكاء الاصطناعي في هذا المجال، مما يؤدي إلى السؤال: هل هذه المخاوف من الذكاء الاصطناعي التوليدي في تصميم تجربة المستخدم مبررة؟
لقد لاحظنا (القسم 5.5) أن مصممي تجربة المستخدم ذوي الخبرة كانوا دفاعيين بشأن فكرة أن “الذكاء الاصطناعي يمكن أن يحل محل المصممين البشريين”. من ناحية، يتمتع المصممون ذوو الخبرة بثقة في قدراتهم الفائقة من حيث مهارات التصميم، وتصادم الحكمة والتنوير في التواصل بين البشر، وامتلاك تعاطف عالٍ تجاه المستخدمين. ومع ذلك، أكدوا (القسم 5.6) على المخاطر الحقيقية التي قد يشكلها الذكاء الاصطناعي التوليدي على قوة العمل من مصممي تجربة المستخدم المبتدئين، مما يطرح معضلة: إذا لم يقم المصممون البشريون المبتدئون بالعمل الروتيني الذي يمكن أن يحل محله الذكاء الاصطناعي التوليدي، فكيف سيصلون إلى مستويات الخبرة التي لا يمكن للذكاء الاصطناعي استبدالها؟ من ناحية أخرى، حتى المصممين ذوي الخبرة أعربوا عن مخاوفهم بشأن الذكاء الاصطناعي بشكل عام، والذكاء الاصطناعي التوليدي بشكل خاص، وكيف يمكن أن يجعل خبرتهم جزئيًا غير ذات قيمة في المستقبل. على وجه التحديد، كانت بعض المخاوف تدور حول القدرات الناشئة للذكاء الاصطناعي، حيث مع تطور الذكاء الاصطناعي ليصبح أكثر تعقيدًا (مثل: المزيد من المعلمات مع بيانات تدريب أكثر تنوعًا وحجمًا)، هناك احتمال أن يتقدم خارج النطاق أو السيطرة المقصودة [6،80]. من ناحية أخرى، قد لا يكون بعض المشاركين لدينا على دراية بالوتيرة السريعة للتطورات – على سبيل المثال، يمكن ضبط نماذج الذكاء الاصطناعي التوليدي على مجموعات بيانات فريدة من أصحاب المصلحة [104]، مما سيرتقي بعملية التفكير الإبداعي، والعصف الذهني، والتوثيق ضمن عملية تصميم تجربة المستخدم. ومع ذلك، تستمر المخاطر حتى لو أدت النماذج المضبوطة إلى حالات حيث تولد أنظمة الذكاء الاصطناعي التوليدي تصاميم أو تتخذ قرارات لا تتماشى مع مبادئ تجربة المستخدم المنقحة، أو رضا أصحاب المصلحة، أو الأسوأ، تنتهك القيم الإنسانية، أو الأخلاقيات، أو الأعراف الثقافية [100، 108]، حتى عندما قد يتولى المصممون مقعد القيادة في اختيار وتنقيح المخرجات مع مطالبات إضافية (راجع، [15]) على النماذج المدربة بشكل مخصص. تتماشى هذه المخاوف مع تصورات المستخدمين حول وعي الآلة (لـ GPT-3) [84]، مما يؤدي إلى عدة توترات، من بينها الأكثر صلة بحالتنا هي التفاعل والتجربة، السيطرة والاستقلالية، والجمود والعفوية. علاوة على ذلك، كما هو الحال في الفن الرقمي، قد يكون الأشخاص الذين يظهرون معتقدات قوية حول الإبداع البشري متحيزين لتقليل قيمة مخرجات تصميم الذكاء الاصطناعي التوليدي باعتبارها أقل إبداعًا وإلهامًا بشكل عام إذا علموا أنها صنعت باستخدام الذكاء الاصطناعي التوليدي [68].
ومع ذلك، فإن عدم القدرة على التنبؤ، والاستقطاب، ومثل هذه التوترات غالبًا ما تشير إلى فرص جديدة حيث يمكن أن تؤدي القدرات الناشئة للذكاء الاصطناعي إلى أساليب مبتكرة لحل المشكلات ومفاهيم تصميم جديدة تمامًا تلهم المبدعين البشريين، ويمكن أن تعطل بشكل جذري الممارسة (تذكر التحولات النموذجية في العلوم الطبيعية [54]). ومع ذلك، بينما نرى بالفعل اضطرابًا جذريًا في عدة عناصر من ممارسة تصميم تجربة المستخدم (انظر، الشكل 3)، لا يزال هناك الكثير مما يجب تحسينه (والحذر بشأنه) مع زيادة قدرات الذكاء الاصطناعي التوليدي. في الأساس، نرى أنه من الضروري ضمان توافق متناغم بين التعاون بين الإنسان والذكاء الاصطناعي في ممارسة تصميم تجربة المستخدم، حيث تتطلب الأساليب والاستراتيجيات التأكد من أن أنظمة الذكاء الاصطناعي التوليدي تفهم وتحترم قيم المصممين البشريين وتفسر وتستجيب لمخاوفهم بفعالية. من أجل ذلك، ندعو إلى مزيد من البحث في مجال الذكاء الاصطناعي المسؤول وما بعده لمواصلة وضع الإنسان في مركز تطوير التكنولوجيا، والأهم من ذلك، في جهود المصالحة في حال اختلاف القيم الأخلاقية بين المجموعات [49]. في الواقع، كما
يجادل راكوفا وآخرون [78] أنه من أجل تمكين العمل المسؤول في مجال الذكاء الاصطناعي بشكل أفضل، تحتاج المنظمات إلى تحديث ممارساتها السائدة، مما يتطلب معالجة الممارسات العملية السائدة، والممارسات العملية الناشئة، ورسم مستقبلها الطموح. لتحقيق هذه الغاية، فإن الطريق إلى الأمام هو الاستمرار في الاستفادة من الذكاء الاصطناعي التشاركي وتصميم حساس للقيم [5، 21، 33، 47] لإنشاء ذكاء اصطناعي من أجل الخير الاجتماعي [95]، وفي النهاية تعزيز آليات حماية إنسانية وإبداعية، مما يسمح لممارسات تصميم تجربة المستخدم وممارسيها بالتكيف والازدهار بشكل سلس.

6.6 قيود الدراسة والعمل المستقبلي

في تحقيقنا، اخترنا نهج البحث الشامل لفهم التأثير العام الذي قد تحدثه GenAI على تصميم تجربة المستخدم، والذي يأتي بطبيعة الحال على حساب العمق – لم نتعمق في أي منتجات أو أدوات أو سير عمل معينة في تجربة المستخدم. وبالتالي، كان من الاستراتيجي إعطاء الأولوية للعرض على العمق، مما أتاح لنا التقاط مجموعة واسعة من وجهات النظر والاعتبارات ضمن مجال تجربة المستخدم. وهذا يمثل خطوة مهمة للأعمال المستقبلية للتعمق في أدوات وحالات استخدام محددة لتقديم تحليل أكثر عمقًا. ثانيًا، كانت مقابلاتنا محدودة بحجم العينة الصغيرة. ) من المشاركين في الشركات الناشئة. غالبًا ما تتمتع الشركات الناشئة بهياكل فرق فريدة وقيود على الموارد، مما يؤدي إلى أساليب مختلفة في معالجة تحديات تصميم تجربة المستخدم مقارنة بالشركات الراسخة. وهذا يدعو إلى دراسة مستقبلية لممارسات التنظيم في نظم الشركات الناشئة، وكيف يتم تضمين الذكاء الاصطناعي التوليدي ضمن هذه الهياكل. علاوة على ذلك، على الرغم من أن الغالبية العظمى من المشاركين لدينا كانت لديهم خبرة في استخدام أدوات الذكاء الاصطناعي التوليدي مثل chatGPT (GPT-3.5) وMidjourney، فقد دعا البحث المشاركين لمشاهدة عروض توضيحية لأدوات الذكاء الاصطناعي التوليدي دون استخدام فعلي. قد تكون هذه التفاعلات المحدودة قد أغفلت رؤى عملية قد تنشأ فقط من التجربة المباشرة مع هذه الأدوات.

7 الخاتمة

عملنا يهدف إلى التحقيق في كيفية إدراك مصممي تجربة المستخدم لأدوات الذكاء الاصطناعي التوليدي، وكيف ينعكس ذلك في سير العمل والممارسات الحالية لديهم. علاوة على ذلك، حددنا الفرص والمخاطر التي يتصورها مصممو تجربة المستخدم لمستقبل التعاون بين الإنسان والذكاء الاصطناعي في ممارسة تصميم تجربة المستخدم. من خلال مقابلات معمقة مع 20 مصمم تجربة مستخدم لديهم سنوات خبرة متنوعة، وجدنا أن المصممين ذوي الخبرة واثقون من أصالتهم وإبداعهم ومهاراتهم التعاطفية، ويعتبرون دور الذكاء الاصطناعي التوليدي بشكل عام مساعدًا. وأكدوا على عوامل “الاستمتاع” و”الوكالة” كعوامل فريدة للبشر، حيث سيظل البشر دائمًا في مقعد القيادة على مخرجات الذكاء الاصطناعي. ومع ذلك، كانت هناك مخاوف جدية بشأن التحديات التي قد تواجه المصممين المبتدئين، الذين قد يتأثرون بتدهور المهارات، واستبدال الوظائف، وإرهاق الإبداع. نستخلص عدة تداعيات للتعاون المسؤول بين الإنسان والذكاء الاصطناعي في تصميم تجربة المستخدم، تتعلق بشكل خاص بحقوق الطبع والنشر والملكية، والإبداع البشري، ومعرفة الذكاء الاصطناعي والوصول إليه. ندعو إلى مواصلة الجهود لتطوير ذكاء اصطناعي مسؤول لضمان آليات حماية الإنسان والإبداع، مما يسمح لمهنيي تصميم تجربة المستخدم بالتكيف بسرعة وبشكل أنيق مع فجر الذكاء الاصطناعي التوليدي وتأثيره المزعزع على ممارسة تصميم تجربة المستخدم.

شكر وتقدير

نشكر جميع المشاركين لدينا على مشاركتهم في دراستنا، وتقديم وجهات نظرهم القيمة حول الذكاء الاصطناعي التوليدي.

REFERENCES

[1] Abdallah MH Abbas, Khairil Imran Ghauth, and Choo-Yee Ting. 2022. User experience design using machine learning: a systematic review. IEEE Access 10 (2022), 51501-51514.
[2] Saleema Amershi, Dan Weld, Mihaela Vorvoreanu, Adam Fourney, Besmira Nushi, Penny Collisson, Jina Suh, Shamsi Iqbal, Paul N. Bennett, Kori Inkpen, Jaime Teevan, Ruth Kikin-Gil, and Eric Horvitz. 2019. Guidelines for Human-AI Interaction. In Proceedings of the 2019 CHI Conference
on Human Factors in Computing Systems (Glasgow, Scotland Uk) (CHI ’19). Association for Computing Machinery, New York, NY, USA, 1-13. https://doi.org/10.1145/3290605.3300233
[3] Kareem Ayoub and Kenneth Payne. 2016. Strategy in the age of artificial intelligence. 7ournal of strategic studies 39, 5-6 (2016), 793-819.
[4] Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (Virtual Event, Canada) (FAccT ’21). Association for Computing Machinery, New York, NY, USA, 610-623. https://doi.org/10.1145/3442188.3445922
[5] Abeba Birhane, William Isaac, Vinodkumar Prabhakaran, Mark Diaz, Madeleine Clare Elish, Iason Gabriel, and Shakir Mohamed. 2022. Power to the People? Opportunities and Challenges for Participatory AI. In Equity and Access in Algorithms, Mechanisms, and Optimization (Arlington, VA, USA) (EAAMO ’22). Association for Computing Machinery, New York, NY, USA, Article 6, 8 pages. https://doi.org/10.1145/3551624.3555290
[6] Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S. Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, Erik Brynjolfsson, Shyamal Buch, Dallas Card, Rodrigo Castellon, Niladri Chatterji, Annie Chen, Kathleen Creel, Jared Quincy Davis, Dora Demszky, Chris Donahue, Moussa Doumbouya, Esin Durmus, Stefano Ermon, John Etchemendy, Kawin Ethayarajh, Li Fei-Fei, Chelsea Finn, Trevor Gale, Lauren Gillespie, Karan Goel, Noah Goodman, Shelby Grossman, Neel Guha, Tatsunori Hashimoto, Peter Henderson, John Hewitt, Daniel E. Ho, Jenny Hong, Kyle Hsu, Jing Huang, Thomas Icard, Saahil Jain, Dan Jurafsky, Pratyusha Kalluri, Siddharth Karamcheti, Geoff Keeling, Fereshte Khani, Omar Khattab, Pang Wei Koh, Mark Krass, Ranjay Krishna, Rohith Kuditipudi, Ananya Kumar, Faisal Ladhak, Mina Lee, Tony Lee, Jure Leskovec, Isabelle Levent, Xiang Lisa Li, Xuechen Li, Tengyu Ma, Ali Malik, Christopher D. Manning, Suvir Mirchandani, Eric Mitchell, Zanele Munyikwa, Suraj Nair, Avanika Narayan, Deepak Narayanan, Ben Newman, Allen Nie, Juan Carlos Niebles, Hamed Nilforoshan, Julian Nyarko, Giray Ogut, Laurel Orr, Isabel Papadimitriou, Joon Sung Park, Chris Piech, Eva Portelance, Christopher Potts, Aditi Raghunathan, Rob Reich, Hongyu Ren, Frieda Rong, Yusuf Roohani, Camilo Ruiz, Jack Ryan, Christopher Ré, Dorsa Sadigh, Shiori Sagawa, Keshav Santhanam, Andy Shih, Krishnan Srinivasan, Alex Tamkin, Rohan Taori, Armin W. Thomas, Florian Tramèr, Rose E. Wang, William Wang, Bohan Wu, Jiajun Wu, Yuhuai Wu, Sang Michael Xie, Michihiro Yasunaga, Jiaxuan You, Matei Zaharia, Michael Zhang, Tianyi Zhang, Xikun Zhang, Yuhui Zhang, Lucia Zheng, Kaitlyn Zhou, and Percy Liang. 2022. On the Opportunities and Risks of Foundation Models. arXiv:2108.07258 [cs.LG]
[7] Zalán Borsos, Raphaël Marinier, Damien Vincent, Eugene Kharitonov, Olivier Pietquin, Matt Sharifi, Olivier Teboul, David Grangier, Marco Tagliasacchi, and Neil Zeghidour. 2022. AudioLM: a Language Modeling Approach to Audio Generation. arXiv:2209.03143 [cs.SD]
[8] Petter Bae Brandtzæg, Asbjørn Følstad, and Jan Heim. 2018. Enjoyment: Lessons from Karasek. Springer International Publishing, Cham, 331-341. https://doi.org/10.1007/978-3-319-68213-6_21
[9] Virginia Braun and Victoria Clarke. 2012. Thematic analysis. In APA Handbook of Research Methods in Psychology: Vol. 2. Research Designs. American Psychological Association, 750 First St. NE, Washington, DC 20002-4242, 57-71.
[10] Lizzy Burnam. 2023. We Surveyed 1093 Researchers About How They Use AI-Here’s What We Learned. https://www.userinterviews.com/blog/ai-in-ux-research-report
[11] Patrick Butlin. 2021. AI Alignment and Human Reward. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society (Virtual Event, USA) (AIES ’21). Association for Computing Machinery, New York, NY, USA, 437-445. https://doi.org/10.1145/3461702.3462570
[12] Bill Buxton. 2010. Sketching user experiences: getting the design right and the right design. Morgan kaufmann, Burlington, Massachusetts, USA.
[13] Kelly Caine. 2016. Local Standards for Sample Size at CHI. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems (San Jose, California, USA) (CHI ’16). Association for Computing Machinery, New York, NY, USA, 981-992. https://doi.org/10.1145/2858036.2858498
[14] Tara Capel and Margot Brereton. 2023. What is Human-Centered about Human-Centered AI? A Map of the Research Landscape. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 359, 23 pages. https://doi.org/10.1145/3544548.3580959
[15] Minsuk Chang, Stefania Druga, Alexander J. Fiannaca, Pedro Vergani, Chinmay Kulkarni, Carrie J Cai, and Michael Terry. 2023. The Prompt Artists. In Proceedings of the 15th Conference on Creativity and Cognition (, Virtual Event, USA,) (C&C’23). Association for Computing Machinery, New York, NY, USA, 75-87. https://doi.org/10.1145/3591196.3593515
[16] Li-Yuan Chiou, Peng-Kai Hung, Rung-Huei Liang, and Chun-Teng Wang. 2023. Designing with AI: An Exploration of Co-Ideation with Image Generators. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 1941-1954. https://doi.org/10.1145/3563657.3596001
[17] Victoria Clarke, Virginia Braun, and Nikki Hayfield. 2015. Thematic analysis. Qualitative psychology: A practical guide to research methods 222, 2015 (2015), 248.
[18] Tiago Silva Da Silva, Milene Selbach Silveira, Frank Maurer, and Theodore Hellmann. 2012. User experience design and agile development: From theory to practice. Fournal of Software Engineering and Applications 5, 10 (2012), 743-751. https://doi.org/10.4236/jsea.2012.510087
[19] Smit Desai, Tanusree Sharma, and Pratyasha Saha. 2023. Using ChatGPT in HCI Research-A Trioethnography. In Proceedings of the 5th International Conference on Conversational User Interfaces (Eindhoven, Netherlands) (CUI ’23). Association for Computing Machinery, New York, NY, USA, Article 8, 6 pages. https://doi.org/10.1145/3571884.3603755
[20] Fiona Draxler, Anna Werner, Florian Lehmann, Matthias Hoppe, Albrecht Schmidt, Daniel Buschek, and Robin Welsch. 2024. The AI Ghostwriter Effect: When Users do not Perceive Ownership of AI-Generated Text but Self-Declare as Authors. ACM Transactions on Computer-Human Interaction 31, 2 (Feb. 2024), 1-40. https://doi.org/10.1145/3637875
[21] Upol Ehsan and Mark O. Riedl. 2020. Human-Centered Explainable AI: Towards a Reflective Sociotechnical Approach. In HCI International 2020 Late Breaking Papers: Multimodality and Intelligence, Constantine Stephanidis, Masaaki Kurosu, Helmut Degen, and Lauren Reinerman-Jones (Eds.).
Springer International Publishing, Cham, 449-466.
[22] Damian Okaibedi Eke. 2023. ChatGPT and the rise of generative AI: Threat to academic integrity? Journal of Responsible Technology 13 (2023), 100060.
[23] Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock. 2023. GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. arXiv:2303.10130 [econ.GN]
[24] Chris Elsden, Arthi Manohar, Jo Briggs, Mike Harding, Chris Speed, and John Vines. 2018. Making Sense of Blockchain Applications: A Typology for HCI. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems (Montreal QC, Canada) (CHI ’18). Association for Computing Machinery, New York, NY, USA, 1-14. https://doi.org/10.1145/3173574.3174032
[25] Ziv Epstein, Aaron Hertzmann, Investigators of Human Creativity, Memo Akten, Hany Farid, Jessica Fjeld, Morgan R Frank, Matthew Groh, Laura Herman, Neil Leach, et al. 2023. Art and the science of generative AI. Science 380, 6650 (2023), 1110-1111.
[26] Rosanna Fanni, Valerie Steinkogler, Giulia Zampedri, and Jo Pierson. 2022. Enhancing human agency through redress in Artificial Intelligence Systems. AI & SOCIETY 38 (06 2022). https://doi.org/10.1007/s00146-022-01454-7
[27] Ed Felten, Manav Raj, and Robert Seamans. 2023. How will Language Modelers like ChatGPT Affect Occupations and Industries? arXiv:2303.01157 [econ.GN]
[28] K. J. Kevin Feng, Maxwell James Coppock, and David W. McDonald. 2023. How Do UX Practitioners Communicate AI as a Design Material? Artifacts, Conceptions, and Propositions. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (, Pittsburgh, PA, USA,) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 2263-2280. https://doi.org/10.1145/3563657.3596101
[29] Jennifer Fereday and Eimear Muir-Cochrane. 2006. Demonstrating Rigor Using Thematic Analysis: A Hybrid Approach of Inductive and Deductive Coding and Theme Development. International Journal of Qualitative Methods 5, 1 (2006), 80-92. https://doi.org/10.1177/160940690600500107 arXiv:https://doi.org/10.1177/160940690600500107
[30] Giorgio Franceschelli and Mirco Musolesi. 2023. Reinforcement Learning for Generative AI: State of the Art, Opportunities and Open Research Challenges. arXiv:2308.00031 [cs.LG]
[31] Michael C. Frank. 2023. Bridging the data gap between children and large language models. Trends in Cognitive Sciences 27, 11 (2023), 990-992. https://doi.org/10.1016/j.tics.2023.08.007
[32] Jonas Frich, Lindsay MacDonald Vermeulen, Christian Remy, Michael Mose Biskjaer, and Peter Dalsgaard. 2019. Mapping the Landscape of Creativity Support Tools in HCI. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (Glasgow, Scotland Uk) (CHI ’19). Association for Computing Machinery, New York, NY, USA, 1-18. https://doi.org/10.1145/3290605.3300619
[33] Batya Friedman. 1996. Value-Sensitive Design. Interactions 3, 6 (dec 1996), 16-23. https://doi.org/10.1145/242485.242493
[34] Ozlem Ozmen Garibay, Brent Winslow, Salvatore Andolina, Margherita Antona, Anja Bodenschatz, Constantinos Coursaris, Gregory Falco, Stephen M. Fiore, Ivan Garibay, Keri Grieman, John C. Havens, Marina Jirotka, Hernisa Kacorri, Waldemar Karwowski, Joe Kider, Joseph Konstan, Sean Koon, Monica Lopez-Gonzalez, Iliana Maifeld-Carucci, Sean McGregor, Gavriel Salvendy, Ben Shneiderman, Constantine Stephanidis, Christina Strobel, Carolyn Ten Holter, and Wei Xu. 2023. Six Human-Centered Artificial Intelligence Grand Challenges. International fournal of Human Computer Interaction 39, 3 (2023), 391-437. https://doi.org/10.1080/10447318.2022.2153320 arXiv:https://doi.org/10.1080/10447318.2022.2153320
[35] Frederic Gmeiner, Humphrey Yang, Lining Yao, Kenneth Holstein, and Nikolas Martelaro. 2023. Exploring Challenges and Opportunities to Support Designers in Learning to Co-create with AI-based Manufacturing Design Tools. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 226, 20 pages. https://doi.org/10.1145/3544548.3580999
[36] Abid Haleem, Mohd Javaid, and Ravi Pratap Singh. 2022. An era of ChatGPT as a significant futuristic support tool: A study on features, abilities, and challenges. BenchCouncil Transactions on Benchmarks, Standards and Evaluations 2, 4 (2022), 100089. https://doi.org/10.1016/j.tbench.2023.100089
[37] Perttu Hämäläinen, Mikke Tavast, and Anton Kunnari. 2023. Evaluating Large Language Models in Generating Synthetic HCI Research Data: a Case Study. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 433, 19 pages. https://doi.org/10.1145/3544548.3580688
[38] Ariel Han and Zhenyao Cai. 2023. Design Implications of Generative AI Systems for Visual Storytelling for Young Learners. In Proceedings of the 22nd Annual ACM Interaction Design and Children Conference (Chicago, IL, USA) (IDC ’23). Association for Computing Machinery, New York, NY, USA, 470-474. https://doi.org/10.1145/3585088.3593867
[39] Gunnar Harboe and Elaine M. Huang. 2015. Real-World Affinity Diagramming Practices: Bridging the Paper-Digital Gap. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (Seoul, Republic of Korea) (CHI ’15). Association for Computing Machinery, New York, NY, USA, 95-104. https://doi.org/10.1145/2702123.2702561
[40] Rex Hartson and Pardha Pyla. 2018. The UX Book: Process and Guidelines for Ensuring a Quality User Experience (2st ed.). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA.
[41] Monique Hennink and Bonnie N. Kaiser. 2022. Sample sizes for saturation in qualitative research: A systematic review of empirical tests. Social Science & Medicine 292 (2022), 114523. https://doi.org/10.1016/j.socscimed.2021.114523
[42] Jonathan Ho, William Chan, Chitwan Saharia, Jay Whang, Ruiqi Gao, Alexey Gritsenko, Diederik P. Kingma, Ben Poole, Mohammad Norouzi, David J. Fleet, and Tim Salimans. 2022. Imagen Video: High Definition Video Generation with Diffusion Models. arXiv:2210.02303 [cs.CV]
[43] Matthew K. Hong, Shabnam Hakimi, Yan-Ying Chen, Heishiro Toyoda, Charlene Wu, and Matt Klenk. 2023. Generative AI for Product Design: Getting the Right Design and the Design Right. arXiv:2306.01217 [cs.HC]
[44] Eric Horvitz. 1999. Principles of mixed-initiative user interfaces. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (Pittsburgh, Pennsylvania, USA) (CHI ’99). Association for Computing Machinery, New York, NY, USA, 159-166. https://doi.org/10.1145/302979. 303030
[45] Stephanie Houde, Steven I Ross, Michael Muller, Mayank Agarwal, Fernando Martinez, John Richards, Kartik Talamadupula, and Justin D Weisz. 2022. Opportunities for Generative AI in UX Modernization. In 7oint Proceedings of the ACM IUI Workshops 2022, March 2022, Helsinki, Finland. Association for Computing Machinery, New York, NY, USA.
[46] Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2023. A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions. arXiv:2311.05232 [cs.CL]
[47] Nanna Inie, Jeanette Falk, and Steve Tanimoto. 2023. Designing Participatory AI: Creative Professionals’ Worries and Expectations about Generative AI. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI EA ’23). Association for Computing Machinery, New York, NY, USA, Article 82, 8 pages. https://doi.org/10.1145/3544549.3585657
[48] Alice M Isen and Johnmarshall Reeve. 2005. The influence of positive affect on intrinsic and extrinsic motivation: Facilitating enjoyment of play, responsible work behavior, and self-control. Motivation and emotion 29 (2005), 295-323.
[49] Maurice Jakesch, Zana Buçinca, Saleema Amershi, and Alexandra Olteanu. 2022. How Different Groups Prioritize Ethical Values for Responsible AI. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (Seoul, Republic of Korea) (FAccT ’22). Association for Computing Machinery, New York, NY, USA, 310-323. https://doi.org/10.1145/3531146.3533097
[50] Harry H. Jiang, Lauren Brown, Jessica Cheng, Mehtab Khan, Abhishek Gupta, Deja Workman, Alex Hanna, Johnathan Flowers, and Timnit Gebru. 2023. AI Art and Its Impact on Artists. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society (Montréal, QC, Canada) (AIES ’23). Association for Computing Machinery, New York, NY, USA, 363-374. https://doi.org/10.1145/3600211.3604681
[51] Zachary Kaiser. 2019. Creativity as computation: Teaching design in the age of automation. Design and Culture 11, 2 (2019), 173-192.
[52] Taenyun Kim, Maria D Molina, Minjin Rheu, Emily S Zhan, and Wei Peng. 2023. One AI Does Not Fit All: A Cluster Analysis of the Laypeople’s Perception of AI Roles. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 29, 1-20 pages. https://doi.org/10.1145/3544548.3581340
[53] Joke Kort, APOS Vermeeren, and Jenneke E Fokker. 2007. Conceptualizing and measuring user experience. , 57-64 pages.
[54] Thomas S. Kuhn. 1962. The Structure of Scientific Revolutions. University of Chicago Press, Chicago.
[55] Chinmay Kulkarni, Stefania Druga, Minsuk Chang, Alex Fiannaca, Carrie Cai, and Michael Terry. 2023. A Word is Worth a Thousand Pictures: Prompts as AI Design Material. arXiv:2303.12647 [cs.HC]
[56] Tomas Lawton, Kazjon Grace, and Francisco J Ibarrola. 2023. When is a Tool a Tool? User Perceptions of System Agency in Human-AI Co-Creative Drawing. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 1978-1996. https://doi.org/10.1145/3563657.3595977
[57] Mina Lee, Megha Srivastava, Amelia Hardy, John Thickstun, Esin Durmus, Ashwin Paranjape, Ines Gerard-Ursin, Xiang Lisa Li, Faisal Ladhak, Frieda Rong, Rose E. Wang, Minae Kwon, Joon Sung Park, Hancheng Cao, Tony Lee, Rishi Bommasani, Michael Bernstein, and Percy Liang. 2024. Evaluating Human-Language Model Interaction. arXiv:2212.09746 [cs.CL]
[58] Florian Lehmann. 2023. Mixed-Initiative Interaction with Computational Generative Systems. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHIEA ’23). Association for Computing Machinery, New York, NY, USA, Article 501, 6 pages. https://doi.org/10.1145/3544549.3577061
[59] Robert W Lent and Steven D Brown. 2006. Integrating person and situation perspectives on work satisfaction: A social-cognitive view. 7ournal of vocational behavior 69, 2 (2006), 236-247.
[60] Joseph CR Licklider. 1960. Man-Computer Symbiosis. IRE Transactions on Human Factors in Electronics HFE-1, 1(1960), 4-11.
[61] Lassi A. Liikkanen. 2019. It AIn’t Nuttin’ New – Interaction Design Practice After the AI Hype. In Human-Computer Interaction – INTERACT 2019, David Lamas, Fernando Loizides, Lennart Nacke, Helen Petrie, Marco Winckler, and Panayiotis Zaphiris (Eds.). Springer International Publishing, Cham, 600-604.
[62] Zhiyu Lin, Upol Ehsan, Rohan Agarwal, Samihan Dani, Vidushi Vashishth, and Mark Riedl. 2023. Beyond Prompts: Exploring the Design Space of Mixed-Initiative Co-Creativity Systems. arXiv:2305.07465 [cs.AI]
[63] Duri Long and Brian Magerko. 2020. What is AI Literacy? Competencies and Design Considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA) (CHI ’20). Association for Computing Machinery, New York, NY, USA, 1-16. https://doi.org/10.1145/3313831.3376727
[64] Robyn Longhurst. 2003. Semi-structured interviews and focus groups. Key methods in geography 3, 2 (2003), 143-156.
[65] Yuwen Lu, Chengzhi Zhang, Iris Zhang, and Toby Jia-Jun Li. 2022. Bridging the Gap Between UX Practitioners’ Work Practices and AI-Enabled Design Support Tools. In Extended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems (New Orleans, LA, USA) (CHI EA ’22). Association for Computing Machinery, New York, NY, USA, Article 268, 7 pages. https://doi.org/10.1145/3491101.3519809
[66] Henrietta Lyons, Eduardo Velloso, and Tim Miller. 2021. Conceptualising Contestability: Perspectives on Contesting Algorithmic Decisions. Proc. ACM Hum.-Comput. Interact. 5, CSCW1, Article 106 (apr 2021), 25 pages. https://doi.org/10.1145/3449180
[67] Nora McDonald, Sarita Schoenebeck, and Andrea Forte. 2019. Reliability and Inter-rater Reliability in Qualitative Research: Norms and Guidelines for CSCW and HCI Practice. Proceedings of the ACM on Human-Computer Interaction 3 (11 2019), 1-23. https://doi.org/10.1145/3359174
[68] Kobe Millet, Florian Buehler, Guanzhong Du, and Michail D. Kokkoris. 2023. Defending humankind: Anthropocentric bias in the appreciation of AI art. Computers in Human Behavior 143 (2023), 107707. https://doi.org/10.1016/j.chb.2023.107707
[69] Bin Ning, Fang Liu, and Zhixiong Liu. 2023. Creativity Support in AI Co-creative Tools: Current Research, Challenges and Opportunities. In 2023 26th International Conference on Computer Supported Cooperative Work in Design (CSCWD). IEEE, New York, NY, USA, 5-10. https: //doi.org/10.1109/CSCWD57460.2023.10152832
[70] Lorelli S. Nowell, Jill M. Norris, Deborah E. White, and Nancy J. Moules. 2017. Thematic Analysis: Striving to Meet the Trustworthiness Criteria. International fournal of Qualitative Methods 16, 1 (2017), 1609406917733847. https://doi.org/10.1177/1609406917733847 arXiv:https://doi.org/10.1177/1609406917733847
[71] Shakked Noy and Whitney Zhang. 2023. Experimental evidence on the productivity effects of generative artificial intelligence. Available at SSRN 4375283 381, 6654 (2023), 187-192. https://doi.org/10.1126/science.adh2586
[72] OpenAI, Josh Achiam, Steven Adler, and Sandhini Agarwal et al. 2023. GPT-4 Technical Report. arXiv:2303.08774 [cs.CL]
[73] Antti Oulasvirta, Niraj Ramesh Dayama, Morteza Shiripour, Maximilian John, and Andreas Karrenbauer. 2020. Combinatorial optimization of graphical user interface designs. Proc. IEEE 108, 3 (2020), 434-464.
[74] Joon Sung Park, Lindsay Popowski, Carrie Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2022. Social Simulacra: Creating Populated Prototypes for Social Computing Systems. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology (Bend, OR, USA) (UIST ’22). Association for Computing Machinery, New York, NY, USA, Article 74, 18 pages. https://doi.org/10.1145/3526113.3545616
[75] John V Pavlik. 2023. Collaborating with ChatGPT: Considering the implications of generative artificial intelligence for journalism and media education. Journalism & Mass Communication Educator 78, 1 (2023), 84-93.
[76] Andrew Pearson. 2023. The rise of CreAltives: Using AI to enable and speed up the creative process. Journal of AI, Robotics & Workplace Automation 2, 2 (2023), 101-114.
[77] Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer. 2023. The Impact of AI on Developer Productivity: Evidence from GitHub Copilot. arXiv:2302.06590 [cs.SE]
[78] Bogdana Rakova, Jingying Yang, Henriette Cramer, and Rumman Chowdhury. 2021. Where Responsible AI Meets Reality: Practitioner Perspectives on Enablers for Shifting Organizational Practices. Proc. ACM Hum.-Comput. Interact. 5, CSCW1, Article 7 (apr 2021), 23 pages. https://doi.org/10. 1145/3449081
[79] Kevin Roose. 2022. The brilliance and weirdness of ChatGPT. https://www.nytimes.com/2022/12/05/technology/chatgpt-ai-twitter.html
[80] S. Russell. 2019. Human Compatible: Artificial Intelligence and the Problem of Control. Penguin Publishing Group, New York, NY, USA. https: //books.google.nl/books?id=M1eFDwAAQBAJ
[81] Jana I Saadi and Maria C Yang. 2023. Generative Design: Reframing the Role of the Designer in Early-Stage Design Process. 7ournal of Mechanical Design 145, 4 (2023), 041411.
[82] Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. 2022. Photorealistic text-to-image diffusion models with deep language understanding. Advances in Neural Information Processing Systems 35 (2022), 36479-36494.
[83] NANNA SANDBERG, TOM HOY, and MARTIN ORTLIEB. 2019. My AI versus the company AI: how knowledge workers conceptualize forms of AI assistance in the workplace. In Ethnographic Praxis in Industry Conference Proceedings, Vol. 2019. Wiley Online Library, American Anthropological Association, Arlington, Virginia, USA, 125-143.
[84] Ava Elizabeth Scott, Daniel Neumann, Jasmin Niess, and Paweł W. Woźniak. 2023. Do You Mind? User Perceptions of Machine Consciousness. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 374, 19 pages. https://doi.org/10.1145/3544548.3581296
[85] Ben Shneiderman. 2020. Human-centered artificial intelligence: Reliable, safe & trustworthy. International fournal of Human-Computer Interaction 36, 6 (2020), 495-504.
[86] Ben Shneiderman. 2020. Human-centered artificial intelligence: Three fresh ideas. AIS Transactions on Human-Computer Interaction 12, 3 (2020), 109-124.
[87] D.K. Simonton. 1999. Origins of Genius: Darwinian Perspectives on Creativity. Oxford University Press, Oxford, UK. https://books.google.li/books? id=a2LnBwAAQBAJ
[88] Marita Skjuve, Asbjørn Følstad, and Petter Bae Brandtzaeg. 2023. The User Experience of ChatGPT: Findings from a Questionnaire Study of Early Users. In Proceedings of the 5th International Conference on Conversational User Interfaces (Eindhoven, Netherlands) (CUI ’23). Association for Computing Machinery, New York, NY, USA, Article 2, 10 pages. https://doi.org/10.1145/3571884.3597144
[89] Irene Solaiman, Zeerak Talat, William Agnew, Lama Ahmad, Dylan Baker, Su Lin Blodgett, Hal Daumé III au2, Jesse Dodge, Ellie Evans, Sara Hooker, Yacine Jernite, Alexandra Sasha Luccioni, Alberto Lusoli, Margaret Mitchell, Jessica Newman, Marie-Therese Png, Andrew Strait, and Apostol Vassilev. 2023. Evaluating the Social Impact of Generative AI Systems in Systems and Society. arXiv:2306.05949 [cs.CY]
[90] Åsne Stige, Efpraxia D Zamani, Patrick Mikalef, and Yuzhen Zhu. 2023. Artificial intelligence (AI) for user experience (UX) design: a systematic literature review and future research agenda. https://doi.org/10.1108/ITP-07-2022-0519
[91] S Shyam Sundar. 2020. Rise of machine agency: A framework for studying the psychology of human-AI interaction (HAII). Fournal of ComputerMediated Communication 25, 1 (2020), 74-88.
[92] Jakob Tholander and Martin Jonsson. 2023. Design Ideation with AI – Sketching, Thinking and Talking with Generative Machine Learning Models. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 1930-1940. https://doi.org/10.1145/3563657.3596014
[93] Katja Thoring, Sebastian Huettemann, and Roland M Mueller. 2023. THE AUGMENTED DESIGNER: A RESEARCH AGENDA FOR GENERATIVE AI-ENABLED DESIGN. Proceedings of the Design Society 3 (2023), 3345-3354.
[94] Joel E Tohline et al. 2008. Computational provenance. Computing in Science & Engineering 10, 03 (2008), 9-10.
[95] Steven Umbrello and Ibo Poel. 2021. Mapping value sensitive design onto AI for social good principles. AI and Ethics 1 (08 2021), 3. https: //doi.org/10.1007/s43681-021-00038-3
[96] Russ Unger and Carolyn Chandler. 2012. A Project Guide to UX Design: For user experience designers in the field or in the making. New Riders, Nora, Indianapolis, USA.
[97] Michael Veale and Frederik Zuiderveen Borgesius. 2021. Demystifying the Draft EU Artificial Intelligence Act – Analysing the good, the bad, and the unclear elements of the proposed approach. Computer Law Review International 22, 4 (Aug. 2021), 97-112. https://doi.org/10.9785/cri-2021-220402
[98] Florent Vinchon, Todd Lubart, Sabrina Bartolotta, Valentin Gironnay, Marion Botella, Samira Bourgeois-Bougrine, Jean-Marie Burkhardt, Nathalie Bonnardel, Giovanni Emanuele Corazza, Vlad Glăveanu, et al. 2023. Artificial Intelligence & Creativity: A manifesto for collaboration. The Journal of Creative Behavior 57, 4 (2023), 472-484.
[99] Qian Wan and Zhicong Lu. 2023. GANCollage: A GAN-Driven Digital Mood Board to Facilitate Ideation in Creativity Support. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 136-146. https://doi.org/10.1145/3563657.3596072
[100] Norbert Wiener. 1960. Some Moral and Technical Consequences of Automation. Science 131, 3410 (1960), 1355-1358. https://doi.org/10.1126/ science.131.3410.1355 arXiv:https://www.science.org/doi/pdf/10.1126/science.131.3410.1355
[101] Wen Xu and Katina Zammit. 2020. Applying Thematic Analysis to Education: A Hybrid Approach to Interpreting Data in Practitioner Research. International Fournal of Qualitative Methods 19 (2020), 1609406920918810. https://doi.org/10.1177/1609406920918810 arXiv:https://doi.org/10.1177/1609406920918810
[102] Bin Yang, Long Wei, and Zihan Pu. 2020. Measuring and improving user experience through artificial intelligence-aided design. Frontiers in Psychology 11 (2020), 595374.
[103] Qian Yang, Aaron Steinfeld, Carolyn Rosé, and John Zimmerman. 2020. Re-examining Whether, Why, and How Human-AI Interaction Is Uniquely Difficult to Design. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (, Honolulu, HI, USA, ) (CHI ’20). Association for Computing Machinery, New York, NY, USA, 1-13. https: //doi.org/10.1145/3313831.3376301
[104] Yang Yuan. 2023. On the Power of Foundation Models. In Proceedings of the 40th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 202), Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett (Eds.). PMLR, USA, 40519-40530. https://proceedings.mlr.press/v202/yuan23b.html
[105] Xiao Zhan, Yifan Xu, and Stefan Sarkadi. 2023. Deceptive AI Ecosystems: The Case of ChatGPT. In Proceedings of the 5th International Conference on Conversational User Interfaces (Eindhoven, Netherlands) (CUI ’23). Association for Computing Machinery, New York, NY, USA, Article 3, 6 pages. https://doi.org/10.1145/3571884.3603754
[106] Chengzhi Zhang, Weijie Wang, Paul Pangaro, Nikolas Martelaro, and Daragh Byrne. 2023. Generative Image AI Using Design Sketches as Input: Opportunities and Challenges. In Proceedings of the 15th Conference on Creativity and Cognition (Virtual Event, USA) (C&C ’23). Association for Computing Machinery, New York, NY, USA, 254-261. https://doi.org/10.1145/3591196.3596820
[107] Haonan Zhong, Jiamin Chang, Ziyue Yang, Tingmin Wu, Pathum Chamikara Mahawaga Arachchige, Chehara Pathmabandu, and Minhui Xue. 2023. Copyright Protection and Accountability of Generative AI: Attack, Watermarking and Attribution. In Companion Proceedings of the ACM Web Conference 2023 (Austin, TX, USA) (WWW’23 Companion). Association for Computing Machinery, New York, NY, USA, 94-98. https://doi.org/10.1145/3543873.3587321
[108] Simon Zhuang and Dylan Hadfield-Menell. 2020. Consequences of Misaligned AI. In Advances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (Eds.), Vol. 33. Curran Associates, Inc., Red Hook, New York, USA, 15763-15773.

  1. ChatGPT: https://chat.openai.com
    DALL-E 2: https://openai.com/dall-e-2
    Stable Diffusion: https://stablediffusionweb.com
    Midjourney: https://www.midjourney.com
    Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s).
    © 2024 Copyright held by the owner/author(s).
    Manuscript submitted to ACM


DOI: https://doi.org/10.1145/3613904.3642114
Publication Date: 2024-05-11

User Experience Design Professionals’ Perceptions of Generative Artificial Intelligence

JIE LI, EPAM, NetherlandsHANCHENG CAO, Stanford University, United StatesLAURA LIN, Google, United StatesYOUYANG HOU, Notion Labs, United StatesRUIHAO ZHU, Cornell University, United StatesABDALLAH EL ALI, Centrum Wiskunde & Informatica, Netherlands

Abstract

Among creative professionals, Generative Artificial Intelligence (GenAI) has sparked excitement over its capabilities and fear over unanticipated consequences. How does GenAI impact User Experience Design (UXD) practice, and are fears warranted? We interviewed 20 UX Designers, with diverse experience and across companies (startups to large enterprises). We probed them to characterize their practices, and sample their attitudes, concerns, and expectations. We found that experienced designers are confident in their originality, creativity, and empathic skills, and find GenAI’s role as assistive. They emphasized the unique human factors of “enjoyment” and “agency”, where humans remain the arbiters of “AI alignment”. However, skill degradation, job replacement, and creativity exhaustion can adversely impact junior designers. We discuss implications for human-GenAI collaboration, specifically copyright and ownership, human creativity and agency, and AI literacy and access. Through the lens of responsible and participatory AI, we contribute a deeper understanding of GenAI fears and opportunities for UXD.

CCS Concepts: • Human-centered computing Empirical studies in HCI.
Additional Key Words and Phrases: Generative AI, UX Designers, Responsible AI, User Experience, Human-AI Collaboration

ACM Reference Format:

Jie Li, Hancheng Cao, Laura Lin, Youyang Hou, Ruihao Zhu, and Abdallah El Ali. 2024. User Experience Design Professionals’ Perceptions of Generative Artificial Intelligence. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’24), May 11-16, 2024, Honolulu, HI, USA. ACM, New York, NY, USA, 27 pages. https://doi.org/10.1145/3613904.3642114

1 INTRODUCTION

The landscape of Artificial Intelligence (AI) is rapidly evolving as adaptable foundation models, built on deep neural networks and self-supervised learning, have gained widespread adoption [6]. These span transformer-based large language models (e.g., GPT-3, BERT), visual (e.g., DALL-E, Florence), or multimodal (e.g., UniLM) models. The power of these models has recently been harnessed, exemplified by the launch of Generative AI (GenAI) tools, such as ChatGPT , DALL-E , Stable Diffusion , and Midjourney . These models have dramatically pushed forward the edge of AI capabilities, where GenAI tools are capable of processing extensive data and learning statistical patterns, enabling them
to generate novel multimodal output based on natural-language inputs (i.e., prompts). They have demonstrated the ability to produce diverse types of content, including text [72], images [82], video [42], and audio [7]. For example, ChatGPT is viewed as the finest chatbot ever [36] with over a million people signed up to use it in only five days [79]. Enthusiastic fans posted examples of ChatGPT producing computer code, essays, and poems; while others type in surreal prompts to Midjourney or DALL-E 2, ranging from requests like”paint a portrait of your boss patting a tiger in the style of Rembrandt”, to “depict a gathering of middle aged dinosaurs sipping coffee and contemplating the meaning of life”, and these tools will return a startlingly accurate depiction in an instant. On the one hand, the widespread adoption of GenAI tools has sparked excitement within society due to their remarkable advancements; on the other hand, it results in anxiety about unanticipated consequences that may be implicitly induced and beyond human control [4]. Moreover, concerns arise regarding homogenization, as any weaknesses in these models have can propagate to downstream applications [6].
The expanding capabilities of GenAI have the potential to revolutionize various aspects of modern life, and significantly impacting individuals who earn their living by creating content, including designers, copywriters, journalists, and tenured professors. Among those professionals, there is a growing recognition of potential harms, such as reputation damage, economic and job loss, plagiarism, and copyright infringement [22,25,50,75]. In the context of participatory AI, it is of utmost importance to recognize that creative professionals and content creators, not just technical experts, possess valuable knowledge, expertise, and interests crucial for the responsible development of AI [5]. Several efforts are underway to explore the impact of AI on UX (e.g., AI in UX Research report [10]). As a step to better understand and address such growing concerns, in this paper, we specifically explore the potential impact of GenAI tools on User Experience Design (UXD) practice, by involving UX Designers, important creative professionals, in our conversations. In contemporary UX Design, the goal is to align with both user needs and business values. Unlike other creative professions, UX Design is not purely artistic work, but revolves around meeting end users’ demands. This entails not only designing the overall look and feel of a digital product [96], but also involves crafting effective navigation systems, labels, and content categorization to ensure that users can effortlessly locate and access the information they require [40]. The UX Design process is an iterative, user-centered design process that requires designers to be creative and collaborative, possess a problem-solving and sense-making mindset, and exhibit empathy towards users [73]. Successful UX Design must go through several iterations and is often linked to how well designers manage to comprehend and translate users’ requirements into corresponding functionality and pleasing aesthetics [90]. With the ability to automate repetitive design tasks, such as creating wireframes, generating layout variations, or producing design prototypes [45], GenAI tools can significantly speed up the UX Design process, allowing UX Designers to explore a wider range of design options and iterate quickly to create more effective and engaging user experiences [81]. While GenAI tools offer many opportunities for UX Designers, it also presents potential threats related to job displacement, ethical concerns, lack of human touch, data quality and bias, and intellectual property issues [47].

2 RESEARCH QUESTIONS AND CONTRIBUTIONS

To harness the power of GenAI in supporting UX Designers and to ensure the responsible disclosure of further AI advancements, our objective is threefold. First, we aim to gain an overview of current UX Design practices, including the tools used, their limitations, as well as the existing workflows and challenges. Identifying existing practices and challenges provides a baseline against which the impact and utility of GenAI tools can be assessed. This also helps in contextualizing GenAI within current workflows. For instance, larger enterprises and smaller companies often operate differently in terms of UX design workflows. Understanding these contexts is crucial for evaluating how GenAI tools
might fit into or potentially alter these various workflows. Second, we seek to comprehend UX Designers’ perceptions toward GenAI – we intend to explore how UX Designers can leverage the potential of GenAI to overcome the existing limitations in tools and challenges in workflows, allowing them to ultimately craft impactful and meaningful user experiences now and in the future. Lastly, we seek to understand how and in what capacity GenAI may impact User Experience Design (UXD) practice, and whether fears that may arise are warranted. To this end, we pose the following research questions (RQs):
  • RQ1: How do UX Designers perceive GenAI tools, and what is their view on the potential for incorporating these tools into their current workflows?
  • RQ2: What opportunities and risks do UX Designers envision for the future of human-AI collaboration in UX Design practice?
To address these questions, we conducted one-on-one in-depth interviews with 20 UX Designers who have diverse years of experience and currently work in companies ranging from startups to large enterprises with over 10,000 employees. These companies are mainly located in four countries across Europe and the United States. We interviewed participants to characterize their practices, and inquire about their attitudes, concerns, and expectations.
Our work contributes to a deeper understanding of current UX Design practices, challenges as well GenAI fears and opportunities for UXD. We find that overall, experienced UX designers are confident in their skills in terms of originality, creativity, and user empathy skills. They believe that GenAI can serve as an assistive tool to help with repetitive and general tasks, and for enhancing productivity – however, only for those who are already skilled designers. They emphasized the unique human factors of “enjoyment” and “agency”, where humans will remain as the final arbiters of “AI alignment” despite anticipations of future emergent AI abilities. However, they expressed serious concerns about skill degradation and unemployment, especially for junior designers, who may lose opportunities to systematically study design, and may instead end up being trained as prompters. Furthermore, they emphasized the importance of society paying sufficient attention to potential problems that may arise with such GenAI advancements, such as copyright and ownership issues, adverse impacts on human creativity (“creativity exhaustion”) through GenAI output speed and homogenization, and concerns over promoting AI literacy and ensuring equal access to stay relevant. Ultimately, while UX Design practice may constantly be in flux, we believe GenAI’s impact on this profession requires immediate adaptation. We discuss the implications of these findings, within the lens of ensuring responsible AI development and human-AI collaboration within UX Design practice.

3.1 Al’s Impact on (Creative) Work

A key line of related literature studies the impact of AI on work. Felton et al. [27] suggest that the top occupations exposed to language modeling include telemarketers and a variety of post-secondary teachers such as English language and literature, foreign language and literature, and history teachers. Eloundou et al. suggests [23] around of the U.S. workforce could have at least of their work tasks affected by the introduction of large language models, while approximately of workers may see at least of their tasks impacted. Solaiman et al. [89] describe specific social impact GenAI systems may have over society – the framework defines seven categories of social impact: bias, stereotypes, and representational harms; cultural values and sensitive content; disparate performance; privacy and data protection; financial costs; environmental costs; and data and content moderation labor costs. Other works have empirically shown the considerable productivity gains through the use of state of AI tools such as ChatGPT [71, 77].
A number of works have discussed the impacts of AI on creative works, which are closely relevant to UX design and research [32]. Pearson [76] delves into the role of modern creative AI technologies as potential muses for artists and creators, highlighting their capacity to enable novel forms of image creation, music composition, animation, and video production. Kulkarni et al. [55] investigates the use of text-to-image models (TTIs) in collaborative design through involving 14 non-professional designers. They reveal that TTIs facilitate rapid exploration of design spaces and support fluid collaboration, with text prompts acting as reflective design aids that facilitates exploration, iteration, and reflection in pair design. Ning et al.[69] investigate the design space of Artificial Intelligence technology-driven Creativity Support Tools (AI-CSTs), highlighting AI-CSTs’ impact on workflows, their potential as co-creators, and strategies for handling AI errors, providing insights into AI-CSTs’ design and technology requirements. Epstein et al. [25] examine the impact of generative AI tools on traditional artistic practices, addressing questions of authorship, ownership, and the potential transformation of creative work and employment. Chang et al. [15] studied the perception of artists who leveraged text-to-image AI models for artwork, finding that 1) artists hold the text prompt and the resulting image can be considered collectively as a form of artistic expression (prompts as art), and 2) prompt templates (prompts with “slots” for others to fill in with their own words) are developed to create generative art styles. Vinchon et al. [98] proposed four potential futures of AI human relationship for creative work, namely “Co-Cre-AI-tion”, “Organic”, “Plagiarism 3.0”, and “Shut down”. More closely related to the present work, Inie et al. [47] discussed creative professionals’ worries and expectations about generative AI, concluding that creative professionals should better understand, cope with, adapt to as well as exploit AI. Long and Magerko [63] provided a concrete definition of AI literacy by identifying 16 core competencies (e.g., recognizing AI, human role in AI, ethics) that humans need to effectively interact with and critically evaluate AI, along with 15 recommended design considerations that foster increased user understanding of AI (e.g., explainability, promote transparency, low barrier to entry). Complementary to the foregoing, we focus and conduct in-depth analysis dedicated to understand how AI impacts one specific occupation closely related to the HCI community – UX Design practice.

3.2 Human-AI Collaboration for Human Empowerment

The relationship between human and machine intelligence has been discussed since the early days of human computer interaction [44, 60]. Given the increasing performance and prevalence of AI in society, much attention have been drawn to study and design for effective and responsible human-AI collaboration in recent years. Amershi et al. [2] proposed design guidelines for human-AI interaction. Through multiple rounds of evaluation with practitioners, they condense over 150 AI-related design recommendations into 18 aspects concerning four different time phases, and test their guideline against 20 AI-fused products with various categories (e.g., e-commerce, web search) and features (e.g., recommendations, search). Yang et al. [103] investigated whether, why, and how human-AI interaction exerts unique difficulties in design. Through synthesizing prior research with their own design, research, and teaching experiences, they identified AI capability uncertainty and AI output complexity as two sources of unique design challenges of AI. Over two decades ago, Horvitz [44] presented a set of principles for building mixed-initiative user interfaces (UIs) that enable users and intelligent agents to collaborate efficiently. The proposed principles are still widely applicable to today’s human-AI collaboration, such as “providing mechanisms for efficient agent-user collaboration to refine results” and “employing socially appropriate behaviors for agent-user interaction”. Lehman [58] also investigates mixed-initiative human-AI interactions and collaborative work with generative systems. The author focuses on designing and evaluating functional prototypes through web-based experiments, exploring concepts like initiative, intent, and control, showing that the levels of initiative and control afforded by the UIs influence perceived authorship when writing text.
Kim et al. [52] used computational methods to categorize ten different AI roles prevalent in our everyday life and compared laypeople’s perceptions of them using online survey data, e.g., AI are considered with roles such as tools (low in both human involvement and AI autonomy), servants (high human involvement and low AI autonomy), assistants (low human involvement and high AI autonomy), and mediators (high in both dimensions). They found that people assessed AI mediators the most favorably, and AI tools the least. Relatedly, Scott et al. [84] studied whether lay people perceive AI as “conscious”, finding dynamic tensions between denial and speculation, thinking and feeling, interaction and experience, control and independence, and rigidity and spontaneity. Capel and Brereton [14] explored the emerging field of Human-Centered Artificial Intelligence (HCAI) by reviewing the existing literature on the subject. Their work identifies established HCAI clusters, and highlights emerging areas, including Interaction with AI and Ethical AI. Additionally, they proposes a new definition of HCAI and encourage greater collaboration between AI and HCI research while suggesting new HCAI constructs. Similarly, Shneiderman [86] explored the concept of HCAI as a transformative approach to AI system design, emphasizing the importance of placing humans at the core of design thinking, offering three key ideas: (1) a two-dimensional HCAI framework that balances human control and automation, (2) a shift from emulating humans to empowering people, and (3) a three-level governance structure for creating trustworthy HCAI systems. These ideas propose a reframe in design discussions with the potential to yield greater benefits for individuals, communities, and society, although they will require validation and refinement through further research and development. Indeed, such HCAI methods [85] aim to ultimately produce Reliable, Safe & Trustworthy (RST) designs, fostering human self-efficacy, creativity, and responsibility while significantly improving performance. Birhane et al. [5] reviewed participatory methods and practices within the AI and Machine Learning pipeline and acknowledged that wider publics beyond technical experts have knowledge, expertise, and interests that are essential to the design, development, and deployment of AI. These prior works set the stage for ensuring, at both a research and policy level, for ensuring responsible human-AI collaboration that ultimately empower people and practices. We draw on these works to help interpret our findings on how UX Designers envision current and future human-AI collaboration.

3.3 Use of AI in UX Design and Research

Studying User Interface (UI) and corresponding User Experience (UX), and how to build tools to better support UX design and research has long been a focal subject in the HCI community [12, 32, 53]. As AI increasingly shows its potential in transforming work practices, researchers have started studying and discussing the implications of AI for UX design and research [51, 90]. So far research has demonstrated the immense potential of AI for UX Design, with prototypes and applications available at key processes, including understanding the context of use, uncovering user requirements, aiding problem solving, evaluating design, and for assisting development of solutions [1,90]. Generally, most believe that AI will likely result in AI-augmented creativity support tools rather than a full transformative shift [61, 98]. Tholander et al. [92] investigated how generative machine learning and large language models may play a part in creative design processes of ideation, early prototyping and sketching, demonstrating practical usefulness and limitations of the system in design ideation processes, as well as how user interaction, and broader discourse around AI, shapes user expectation of AI’s capabilities and potentials. Feng et al. [28] found that UX practitioners can be more hands off when designing prototypes since AI can provide support. They introduced in detail how UX practitioners communicate with AI and use AI as design materials for their daily work. Other works have also proposed specific prototypes to support better user experiences [38, 99]. Evaluations of such systems point to the potential of AI for augmenting human creativity, as AI often brings distinct perspective that opens up new avenues for artistic expression[16], and fosters a two-way exchange of ideas between users and AI [56]. Lin et al. [62] explored the design
space of mixed-initiative co-creativity systems where humans and AI systems could communicate creative intent to each other.
Meanwhile, research has also revealed challenges of incorporating AI into the UX design process. Gmeiner et al. [35] showed designers face many challenges in learning to effectively co-create with current generative AI systems, including challenges in understanding and adjusting AI outputs and in communicating their design goals. For example, when expecting AI tools to provide them with project-relevant work examples, or exhibit more context awareness for the specific problem at hand. Zhang et al. [106] further demonstrated limitations of AI for architectural design, in that the AI tool “fails to understand the Architectural domain-specific terms” and “generates surreal images unsuitable for construction purposes”. Other works demonstrate similar challenges of prompt engineering in using AI for product design [43]. On the other hand, Abbas et al. [1] found that while the majority of UX designers had no expertise with ML as a design tool, it holds great potential for improving productivity. Thoring et al. [93] define a research agenda for GenAI-enabled design, highlighting key areas include creating better guided prompts, better interfaces for output interpolation, and getting AI to think outside of the latent space box (i.e., extrapolation). With regard to the potential of AI in UX research, Yang et al. [102] proposed a methodology to simulate user experience by using AI-aided design technology in mobile application design. Hamalainen et al. [37] leveraged OpenAI’s GPT-3 model to generate open-ended questionnaire responses about user experiences over video games, where they find that GPT-3 can yield believable accounts of HCI experiences that is hard to be distinguished from real human responses. Park et al. [74] leveraged GPT-3 to simulate synthetic users and conversations over social computing platforms prototypes, where they find generated responses are hard to be differentiated from actual community behavior. By contrast to the majority of literature that focus on evaluating the use of AI for UX design and research in specific settings or technologies, our work contributes a deeper understanding of current practices, AI adoption, and perceptions of UX practitioners regarding emerging generative AI technologies, which sets the stage for envisioning how UX as a profession may evolve in the future with the ever increasing capabilities of AI.

4 METHOD

We employed a semi-structured interview methodology for this study [64]. We initially distributed a screening questionnaire with the goal of recruiting UX Designers who have diverse educational backgrounds and industry experiences, including years of experience, working at various company sizes, and residing in at least two different continents (namely Europe and North America). The screening questionnaire was distributed across social media networks (e.g., LinkedIn) and mailing lists (including design schools), targeting UXD communities. We received 48 responses, all of which were reviewed. While all respondents met our recruitment criteria, only 20 participants chose to participate. This mix of participants was well balanced in terms of our recruiting requirements. These 20 participants engaged with us in online one-on-one 60-minute interviews between April and July, 2023. 20 participants were deemed sufficient given saturation points exceeding 17 interviews in sociological research [41] and local sample size standards in HCI (remote interviews usually have ) [13]. Each interview was conducted by an interviewer (author), while a second researcher (also author) was present to take notes. All interview sessions were additionally audio recorded. The interview process, data collection, and storage strictly adhered to the GDPR and ethical policy and data protection guidelines set forth by the last author’s research institute.
Characteristics UX Designers
Work experience >5 years P1, P5, P6, P8, P10, P11, P12, P14, P16, P17, P18
3-5 years P2, P7
<3 years P3, P4, P9, P13, P15, P19, P20
Company size >10,000 P5, P6, P7, P8, P9, P11, P14, P15, P16, P17
1,000-10,000 P1, P2, P10, P13, P18
500-1,000 P20
<50 P3, P4, P12, P19
Table 1. Years of work experience and company size (employee count) for our 20 UX Designer participants (P1-P20)
GenAI video demos GenAI tools included Date posted on YouTube
Crafting a mobile interface Midjourney (V4), Stability Photoshop Plugin (0.7.0) & ChatGPT (Jan. 30, 2023 version) Feb. 20, 2023
Drafting a design proposal (only used the part of Notion AI: Notion AI (2.21) Mar. 15, 2023
Creating UI components Prompt2Design (V3) May. 6, 2023
Organizing ideation notes MiroAI (Beta) Mar. 8, 2023
Table 2. GenAI video demonstrations used in the interview

4.1 Participants

To ensure traceability in presenting the results, we labeled the 20 participants as P1-P20. Among the 20 participants, there were 12 females and 8 males. Three participants fell within the 18-25 age group, while 12 were in the 26-35 group. An additional five participants belonged to the age group. 11 of them had over five years experience as UX Designers, two had three to five years, and seven had fewer than three years of experience. For their experience using GenAI tools, all 20 participants used ChatGPT regularly, and eight of them regularly used GPT-4 (known as ChatGPT Plus). Thirteen participants tried Midjourney and occasionally used it for inspiration. Seven participants also mentioned their occasional use of Notion AI. Four tried Miro AI, and two used Bard.
The 20 participants were from 15 different companies, with ten participants located in Europe, including The Netherlands, Sweden, Romania, Hungary, and the other 10 participants located across the United States. Moreover, ten participants were employed by enterprises with over 10,000 employees (across well known Big Tech companies), five worked in internationally known enterprises with 1,000 to 10,000 employees, one was employed at a company with 500 to 1,000 employees, and four worked in small companies or startups with fewer than 50 employees. To ensure data privacy, we ensure that where participants work cannot be traced back to them (Table 1).

4.2 Interview questions and procedure

The interview questions are scripted to gather answers for the research questions presented in Sec. 1, which are divided into four parts. Part 1 begins with general questions that aim to gather information about participants’ backgrounds, work experiences, common challenges in UX Design, and perceived limitations in their current tools. Part 2 comprises a set of questions focused on understanding participants’ current experiences and knowledge of GenAI tools, as well as their attitudes toward using GenAI in UX Design work. In Part 3, participants are shown a collection of videos
Fig. 1. The analysis resulted in six major themes, where each theme contained two to four sub-themes.
consisting of four demonstrations from YouTube using various GenAI tools for UX Design, ideation, research data synthesis, and front-end development work (Table 2). We used these video demonstrations primarily as probes to gauge our participants to reflect on these tools in general and how they relate to their practices, and not to assess whether or not they have used these specific tools themselves. Subsequent follow-up questions delve into their impressions of the GenAI tools shown in the videos, their envisioned roles for AI in human-AI collaboration, and their concerns regarding ethics and ownership. Part 4 concludes the interview with questions soliciting further reflections, advice or recommendations for UX Designers interested in integrating GenAI into their work and identifying their most desired, innovative functionalities of current and future GenAI. The complete list of interview questions is available in Supplementary Material A.
The interview process lasted approximately 60 minutes and involved the following steps:
  • Step 1 ( 5 mins ): Introduction given by the interview facilitator to ensure participants understand the purpose of the interview, have read, comprehended, and signed the informed consent form;
  • Step 2 ( : General questions about background, workflows, and UX practice;
  • Step 3 ( 20 mins): Attitudes, concerns, and expectations of GenAI Systems;
  • Step 4 ( 20 mins): Envisioning GenAI’s role in human-AI collaboration (including watching a video demonstration of a sample of current GenAI tools);
  • Step 5 ( 5 mins): Concluding questions.
Fig. 2. Eight categories of tools commonly used in UX Design practices
Fig. 3. Activities that can and cannot be completed by GenAI in the UX Design workflow

4.3 Data Analysis

We used Miro for remote collaboration on data analysis. The audio recordings of the interviews were transcribed. We analyzed the data following a deductive/inductive hybrid thematic analysis approach [29, 101]: First, the transcripts were coded by five researchers using a deductive coding approach, where each researcher identified topics in the transcripts independently according to four categories in our code manual, created based on our research questions and interview guide. After the coding phase, the five researchers held a half-day workshop to compare and discuss the
identified topics and converted the coded data into 415 digital statement cards on Miro. Each statement card consists of the original quote from the participant, participant ID (P1-P20), and a one-sentence statement or summary of the quote on the card. Instead of calculating statistical inter-rater reliability (IRR) for the analysis, the consensus among the five researchers was reached through daily meetings, workshops and discussions [67]. Given that the categories in the code manual were still too broad, we then adopted an inductive coding approach [9,17] to surface sub-themes from the statement cards using the affinity diagramming technique [39], and to ensure the broader categories are consolidated. To help ensure our analysis meets the trustworthiness criteria (cf., [70]), we describe our hybrid approach in detail in Supplementary Material B. The analysis resulted in six major themes (Fig. 1): (1) Limitations in current UX Design tools: (a) frictional design handoff; (b) fragmented tools; and (c) lack of advanced design support. (2) Challenges in current UX Design: (a) managing stakeholders expectations; (b) understanding users; (c) satisfying both business and user needs; and (d) lacking domain knowledge. (3) UX Designers’ attitudes towards GenAI: (a) overall impression of the videos; and (b) shortcomings of GenAI tools in the videos. (4) Human-AI collaboration: (a) threats and collaboration with GenAI; (b) work replaceable by GenAI; and (c) work irreplaceable by GenAI. (5) Ethical concerns and ownership: (a) originality, copyright, and ownership; (b) skill degradation, and unemployment; (c) privacy, transparency, and equal access to GenAI. (6) The future of GenAI in UX Design: (a) envisioning collaboration with GenAI; (b) promoting AI literacy and accessibility for designers; (c) AI’s desired role in enhancing design processes; and (d) challenges of defining AI-generated outcome metrics.

5 FINDINGS

This section illustrates the current UX Design tools and workflows, and provides a detailed presentation of the findings under the six themes identified in the data analysis (Sec. 4.3 and Fig. 1). Fig. 3 displays the activities in the UX Design workflow that UX Designers in our study stated can and cannot be completed by GenAI.

5.1 Current UX Design Tools and Workflow

Understanding the limitations and challenges of current UX design workflows is crucial for creating a baseline to assess the impact and utility of GenAI tools. This knowledge not only informs how GenAI can address existing issues and enhance workflows, but also aids in contextualizing these tools within different operational environments, from larger enterprises to smaller companies. This is essential for evaluating the integration and potential modifications GenAI tools might bring to these diverse workflows, where one size may not fit all. Moreover, this understanding plays a key role in setting realistic expectations for the adoption of GenAI tools, providing insights into the aspects likely to be embraced or resisted by UX designers.
We synthesized eight categories of frequently used tools that support UX Designers in their daily work (Fig. 2), as well as eight typical steps within UX Design practice – from gathering requirements from business stakeholders and users, to ideation, prototyping, user testing, iteration, and enabling the realization of solutions with the development team. Table 3 illustrates the main activities and tools used in each step. We observe that Figma was mentioned by all participants as a design tool to support high-fidelity prototyping and handoff to development teams. Miro was also frequently mentioned for brainstorming and synthesizing user research data.
In terms of workflow, we observed that larger enterprises (with over 10,000 employees) often have clear divisions of roles between UX Designers, Product Managers, and UX Researchers. In contrast, smaller companies, such as startups with fewer than 50 employees, often lack such clear divisions. UX Designers in smaller companies are responsible for conducting user research and may spend more time convincing stakeholders to prioritize UX. Furthermore, workflows
Steps Main Activities Tools
1 Business Requirements
– Analyze stakeholders’ requirements
– Conduct a stakeholder kickoff meeting to align on expectations, business objectives, and technical challenges.
Brainstorm
2 User Requirements
– Define research goals and identify the necessary features to achieve those goals.
– Synthesize insights from user research to create personas and empathy maps.
– Dive into user’s motivation, pain points, and challenges to inform design decisions.
Brainstorm & User Research
3
– Convert research insights into design ideas through sketches.
– Conduct internal reviews to refine ideas with the design team.
– Organize workshops with stakeholders to consolidate design solutions.
Brainstorm, Documentation & Presentation
4 Prototyping
– Make interactive prototypes that visually and functionally represent the proposed design solutions.
– Ensure prototypes capture user interactions, workflows, and navigation paths.
Design
5
– Conduct user testing to evaluate the prototypes.
– Apply online (or offline) moderated (or unmoderated) user testing to guide participants through predefined tasks.
Communication & User Research
6 Iteration
– Analyze user testing results, identifying areas for improvement.
– Iterate on the design based on the gained insights
Design & Non Digital
7 Solution – Collaborate with stakeholders to ensure the design addresses user needs and aligns with the business goals. Documentation & Presentation
8 Design Handoff
– Provide the development team with design assets and documentation and guide them to ensure the design’s consistency and accuracy across different screen sizes.
– Address technical constraints and make UI adjustments to optimize the user interface for the final product.
Documentation, Presentation & Design
Table 3. The typical UX Design workflow in 2023
in smaller companies tend to be shorter than those in larger enterprises due to fewer people involved and lower communication costs. However, smaller companies may also omit or simplify some steps due to resource limitations, such as a limited budget for extensive user evaluations. P6, who has experience working in both small companies and large enterprises, explained, “In large companies, our [UX Designers’] roles are well-defined. Although we also face a complex business landscape and stakeholders with different opinions, we have collaborators like UX Researchers and Product
Managers to handle those aspects, and we can focus more on design. In small companies, we [UX Designers] must take on the roles of researchers, Product Managers, and designers.”

5.2 Theme 1: Limitations of Current UX Design Tools

5.2.1 Limitation 1: Frictional design handoff to development teams. Participants pointed out that handing over design specifications, assets, and documentation to the development team is a rather frictional process. The current design tools do not support a seamless transition from design to development in a format accessible and understandable by the developers: “[We need] more gateways between designers and the developers, improving prototyping workflows by enabling Figma prototypes to feed directly into the development coding workflow, to reduce the frictions in communication between design and development teams.”[P14]
5.2.2 Limitation 2: Fragmented Tools. Participants emphasized that the current UX Design tools are scattered. Each tool has specific strengths or functionalities, but transitioning seamlessly between them is challenging: “We use Miro for brainstorming and creating low-fidelity prototypes, and Figma for high-fidelity ones. However, transferring wireframes from Miro to Figma isn’t straightforward.”[P10] Some participants specifically noted that the current design tools lack support for user research: “Something valuable would be the ability to conduct usability testing and generate data within the design tool itself.” [P17].
5.2.3 Limitation 3: Lack of Advanced Design Support. Participants expressed dissatisfaction with the current design tools due to their lack of advanced support for complex animations, cross-platform designs, graphic design, and more intelligent design systems: “There are micro animation-type things that you don’t get from Figma, like the timelines that you would use in After Effects or Principle to fine-tune your animations.” [P17] “Design systems lack intelligence. I’d appreciate a design system that recommends suitable design components and automatically adjusts these components to fit various screens.” [P5].

5.3 Theme 2: Challenges in Current UX Design

5.3.1 Challenge 1: Managing stakeholder expectations and gaining buy-ins for UX.. The most frequently mentioned challenge is that UX Designers often spend over half of their time communicating and collaborating with a variety of stakeholders. This includes working with external vendors who take time to adapt to the work culture and style (P13), spending excessive time keeping stakeholders in sync, coordinating across time zones, and convincing them of the necessity of UX and gaining buy-in from management-level stakeholders (P1, P8, P12). Additionally, they must balance the expectations of stakeholders and clients with user needs to come up with an ideal solution (P18). P12 mentioned, “I have to convince others that UX is necessary and do a lot of work with developers to ensure the design is appropriately implemented. Some organizations take UX as lightly as a window dressing task and give more buy-in to the backend. So, we need to constantly educate them about the importance of scaling up UX.” Such management processes, stakeholder communication, and expectation alignment are aspects that heavily rely on crucial human-to-human communication, and an area which GenAI technologies do not currently support.
5.3.2 Challenge 2: Understanding end-users. Participants mentioned that conducting research to understand user needs is challenging in terms of finding out the right pain points of users and building the right products: “I think the challenge always lies in trying to understand who I’m designing for, what their needs, goals, and problems are, and whether the product or service that I’m working on fits into the broader landscape.”[P17]. Such a challenge lies at the heart of
UX Design and research [12]. While some approaches aim to simulate synthetic users and conversations (e.g., using GPT-3 [74]), it is unlikely this can currently substitute for a deep understanding of end-users by designers.
5.3.3 Challenge 3: Designing solutions that satisfy both business value and user needs. Participants pointed out that coming up with good design solutions that balance clients’ needs, users’ needs, business value, and visual novelty is challenging: “The UX field is quite mature, with existing design references and design systems aiding the work. The challenge is to come up with new ideas constantly and find a balance when considering both established design references, internal stakeholders’ expectations, user needs, and incorporating them into the new design.”[P18] Satisfying business needs can carry implications for originality, copyright, and ownership, all of which impact adoption (cf., Sec 5.6.1).
5.3.4 Challenge 4: Lacking domain knowledge and resources. Participants mentioned that lacking domain expertise and resources to access knowledge poses a significant challenge: “One of the challenges for UX Designers is acquiring extensive domain knowledge, such as creating designs customized for specialized fields like medical, automotive, fintech, and so on.”[P1]; “In a startup, we often face resource constraints that limit our access to experts, conducting user research or usability testing on a larger scale, which is crucial to ensure the feasibility and desirability of the products.”[P6]. We believe this may be a promising avenue by which GenAI tools, specifically text-based large language models such as ChatGPT, can empower designers. Despite current model hallucinations (cf., [46]), designers can have immediate, vast, and easy access to domain-specific information and resources.

5.4 Theme 3: UX Designers’ Attitudes toward GenAI

All participants agreed that GenAI tools can be powerful and convenient in supporting general tasks such as early-stage design ideation, basic UX writing, and basic coding. These tools provide a starting point for human designers to build upon, helping them overcome the fear of beginning from scratch. While many participants were optimistic about the potential of GenAI tools, envisioning these tools as a means to enhance design efficiency, some others expressed mixed feelings towards GenAI. On one hand, they acknowledged that AI could assist with repetitive tasks, but they underscored that UX Design is inherently user-centered. The reliability of AI-generated designs is questionable, as they may not always align with user needs and goals. Relying too heavily on AI to design products or interfaces without human input was seen as risky. All participants emphasized that the current quality of AI outputs, such as text accuracy, image resolution, and design details and functionalities, still requires significant human input and review, which may not necessarily save time.
All participants believed that human designers possess unique abilities, as they exhibit empathy towards users and derive enjoyment from the design process, even when starting ideation from scratch: “GenAI tools can help nowadays in generating basic ideas to help us populate the blank canvas, thereby aiding in overcoming the fear of the ’empty canvas’. It’s similar to when we need to write reports; you can start with a basic structure generated by ChatGPT and then build upon it. However, I fear that in the long run, I might become more of an editor, gradually losing my design superpower to fill in the blank canvas and be empathetic towards users.”[P8].
5.4.1 Perceived shortcomings of GenAI tools shown in the demo. There were two shortcomings perceived to be key amongst participants:
(1) Practicality and efficiency. Participants highlighted that the GenAI tools showcased in the demos are quite generic, and the design process still requires a substantial amount of manual work. This includes tasks such as redrawing icons and buttons and inputting accurate text-based prompts. “Designers are trained to think visually. Our design process
involves trial and error on sketchbooks. Designs emerge from numerous sketch trials, not from text-based prompts. [In the demo], you’re required to switch between various GenAI tools, Photoshop plugins, and perform a substantial amount of manual work, including redrawing and regenerating high-resolution images. I question whether these GenAI tools truly enhance the design workflow.”[P12]
(2) Accuracy and reliability. Participants questioned the overall accuracy and reliability of outcomes generated by AI tools. They hold the belief that human inputs are necessary to validate the AI-generated outcomes. P16 shared: “For visual design, GenAI tools are still quite limited in capturing nuances and intricate details. Regarding ideation, AI tools like Miro AI may fall short in adequately representing the insights of stakeholders who possess extensive experience, say, 20 years in a specialized field like healthcare.” [P16]. P19 described her negative experience using GenAI tools: “For Midjourney, I often found myself spending hours adjusting my prompts, yet still unable to achieve the envisioned results. As for Miro, the cards are only clustered based on keywords, requiring manual effort to create coherent clusters.” [P19]. Furthermore, P3 expressed doubts about the reliability of AI’s code generation capability: “I’ve actually attempted to have ChatGPT generate code for certain features or functionalities, but it didn’t quite produce the results I was looking for.”[P3].

5.5 Theme 4: Human-AI Collaboration

All participants described GenAI tools as helpers or assistants and recommended that “augmentation” and “enhancement” are good words to describe the assistance provided by GenAI in UX Design.
5.5.1 Perceptions of threats and collaboration with GenAI. Participants expressed varying levels of concern regarding the potential threats of GenAI. Some felt that if GenAI remained a generic tool under their control, it wouldn’t be a source of concern, while others worried about its ability to simulate designers’ reasoning. Recommendations included treating GenAI as a team member, with some participants emphasizing the importance of humans retaining ultimate decision-making power in human-AI collaboration. P18 commented: “If the technology does eventually reach a point where it can accurately simulate designers’ rationales, then I might start worrying about it.” [P18]. P19 recommended treating GenAI as a team member: “I would envision AI as part of the team, contributing to the overall process, but not solely responsible for the final outcome.”[P19]. P8 stressed that humans should have the ultimate decision-making power in human-AI collaboration: “It should depend on people to choose which part I want AI assistance, for the purpose of expediting tasks. I don’t like the word ‘replace’. I very much enjoy the UI and UX Design…I want AI to be my apprentices. Humans do the creative and complex parts, and AI apprentices fill in the colors or draw lines.”[P8].
5.5.2 General and repetitive work can be replaced by GenAI. All participants shared the belief that GenAI has the potential to replace repetitive tasks, such as generating “lorem ipsum” texts for design mockups, condensing design briefs and technical documentation, crafting repetitive design components, creating responsive designs for various screens, automatically documenting visual design elements into design systems, generating low-fidelity prototypes, and generating design variations for inspiration (see Fig. 3: “GenAI Can”). As P8 said, “I envision AI functioning as a productivity tool that quickly aids me in concrete tasks. For instance, AI could summarize PRDs [Product Requirement Documents] or explain technical backgrounds in plain language. Currently, these aspects require substantial effort from PMs [Product Managers] and engineers to convey to us designers.”[P8]. P12 expressed their wish for AI-supported responsive design, “I’m hopeful that GenAI can assist with tasks like responsive design for different screens and suggesting user flows across pages, not just on a single page.” [P12] Many other participants expressed their desire for GenAI support in various tasks, including providing design inspiration (P2, P4, and P8), as well as facilitating efficient UX writing (P18 and P19).
P19 shared: “In our startup, we don’t have a dedicated UX writing role. Our designers often use ChatGPT to assess the appropriateness of the UX content in our design.” [P19].
5.5.3 UX Design work cannot be replaced by GenAI. Many participants made it clear that much UX Design work simply cannot be replaced by AI, including the work that requires communicating between human stakeholders, being empathetic towards user needs, and designing user-centered functionalities. They stressed the importance of humans being the original creator and validator of the AI outcomes (see Fig. 3: “GenAI Cannot”).
(1) User inputs and human-human communication or collaboration. Participants confidently expressed that they do not currently perceive GenAI as a competitor, since human designers serve as ambassadors for user needs and continue to be the ultimate decision-makers and arbiters of AI alignment. They emphasized that GenAI cannot replace work that requires user inputs, and human-human collaborations: “GenAI might be capable of taking over basic, repetitive, and straightforward tasks, but it cannot replace service design and collaborative efforts that demand more systematic and abstract thinking.”[P6]. P14 additionally pointed out, “UX Design is not just about aesthetics, but also usability and the accessibility of elements. Real human users are essential for improving designs through actual user research and testing.” [P14]. P16 believed that human communication is not replaceable by AI, stating “I value insights from human colleagues, their experiences, ideas, and the knowledge they bring to the table. These do not have to be flawless, but they aid us in charting directions together. I do not believe AI can perform this aspect of work.”[P16].
(2) Human creativity and decision making. Some participants pointed out that human creativity and originality are strong human qualities that are difficult to be replaced by AI. AI outcomes heavily rely on past data, potentially leading to recurring results due to the limited data pool. Human designers remain the driving force capable of deeply comprehending design contexts, empathizing with users, and innovating based on the latest knowledge: “The creative work that humans enjoy doing should not be replaced by AI. AI is a trained model that is based on the past, while human creativity and originality are more future-oriented.” [P8]; “Creativity can be supported by AI, as it can remix and regenerate based on human-inputted creative data, but humans possess the originality.”[P4].
Several participants stressed that final decision-making cannot be delegated to GenAI. Human verification is deemed necessary for each stage of design and development projects: “AI must be intentionally used, monitored, and validated by humans. They are essentially Large Language Models, which are computers trained to converse like humans. The quality of the content and data provided by AI is not necessarily supported by actual research data verified by human researchers.” [P8].

5.6 Theme 5: Ethical Concerns and Ownership

Copyright, privacy, data biases and transparency, skill degradation and unemployment, social disparity, and equal access are frequently mentioned ethical concerns among all participants.
5.6.1 Originality, copyright, and ownership. Participants emphasized the significance of crediting the original creators of artworks or art styles utilized in UX work. P3 and P4 suggested that blockchain technology could be useful in ensuring direct ownership attribution to the original creators, regardless of the extent to which the original work has been modified. P4 stated, “Ownership must remain traceable regardless of how extensively the AI outputs have been modified, derived, or reproduced from the originals. Perhaps blockchain can provide a solution here.” P14 expressed her concerns over design plagiarism, “From the confidentiality point of view, what data can be placed into GenAI tools, how we can make sure that we are using legal sources and we are not violating GDPR regulations. There are already cases that designers copy others’ work to build their own portfolios. GenAI can make plagiarism in design much worse.”[P14].
Expanding on the subject of ownership, some participants believed that humans must be the owners, with GenAI tools serving as supportive aids to humans in achieving the final results. P11 commented, “AI is not human; ownership belongs to humans. AI is a tool, much like Figma. I don’t believe ownership should be assigned to technology or tools. If you are the one refining, crafting the prompt, contributing creativity and thoughts to AI for generating the work, and overseeing the entire process, ownership should be attributed to humans.” [P11].
However, several other participants recommended attributing ownership to the part of work supported by AI or labeling it as AI-contributed and ensuring transparency about the tools used. P18 provided a fitting analogy, “The ownership attribution for using AI in future work should be similar to how researchers present their research findings, mentioning the sources and systems used as references.” [P18] This was echoed by P19, “Ownership should be seen as a collective effort with input from various team members, including GenAI.” [P19].
5.6.2 Skill degradation and unemployment. Several participants highlighted their concern that human UX Design work might be less appreciated, potentially leading to complacency among human designers and a decline in their skills. They worry that AI taking over tasks previously considered essential for designers could contribute to skill degradation. These concerns, mentioned primarily by senior designers (refer to Table 1), are particularly relevant for junior designers, who could miss out on opportunities to further develop their skills. As such, human designers might transition into more managerial or generalist roles. P16 (a senior designer) envisioned this, stating, “Designers may fall into the trap of becoming complacent. Junior designers might have fewer chances to learn the design process but, instead, they may focus on mastering AI systems. Consequently, designers could evolve into generalists.”[P16]
Participants specifically highlighted that unemployment and skill degradation could evolve into significant societal issues if not addressed seriously from the outset. P4 (junior designer) and P5 (senior designer) expressed concerns that human designers’ skills might become narrow as we increasingly specialize in specific tasks if AI potentially takes over many parts of the work in the design workflow, potentially limiting our creativity as designers: “AI can put us into a bubble. We feel that we are doing higher-level work, and the career barrier of becoming a designer is getting higher. But somehow we are forced to be more specialized in a limited range of tasks. The loss of oversight over the design and development workflow may jeopardize our creativity.” [P5].
P12, also a senior designer, thoughtfully reflected on the impact of GenAI on education and careers of junior designers, “Our expertise matures through year-over-year practice, encompassing not only design skills but also design thinking and genuine user empathy. If companies decide to employ prompters or AI operators in the future to replace junior designers’ roles, how can these prompters or AI operators receive the proper training equivalent to what we receive in design schools or through design practice? More importantly, how can junior designers acquire new skills or senior designers improve their expertise if AI takes over the tasks they need to practice and uphold their design skills?”[P12].
5.6.3 Privacy, transparency, and equal access to GenAI. Participants pointed out that privacy is a significant concern within the realm of GenAI. As these AI models ingest large amounts of data to learn patterns and generate content, there is potential for the unintentional exposure of sensitive information, such as personal or confidential details that users did not intend to share: “If I want to use ChatGPT to extract information from technical documents, I don’t know how I should prevent sensitive information from being leaked. For my personal life as well, it is not clear how much of my personal data will be retained to further train the AI models.”[P19]
Several participants stressed the importance of AI being trained on unbiased data and ensuring equal access to GenAI tools. This is due to concerns about social disparity and the potential for country policies to limit people’s ability to learn and access AI tools, as well as the possibility of biased data leading to questionable outputs. “Big companies and
developed countries may gain control over the data fed into AI. It’s also easier for them to access GenAI tools compared to some developing countries.” [P6]. P2 added, “It is important to ensure that individuals who are relatively less educated and have limited knowledge of GenAI can still use these tools to enhance their work and daily lives.”[P2].

5.7 Theme 6: The Future of GenAI in UX Design

Theme 6 provided insights into participants’ visions and desires related to the integration of GenAI in the field of design.
5.7.1 Envisioning collaborative design with GenAI. Some participants envisioned a triangulated system that involves collaboration and input from users, designers, and GenAI. This approach aims to minimize biases and ensure that crucial nuances, essential for design innovations, are not overlooked. As P11 mentioned, “Human designers may overlook certain nuances that could hold importance for users. I wish for GenAI to enhance this aspect by providing its input grounded in vast past data and knowledge, potentially elevating user experience design to the next level.”[P11].
5.7.2 Promoting AI literacy and access for designers. Several participants expressed a hope for increased AI literacy through education, benefiting both experienced and junior designers. P13 (a junior designer) noted, “Students may feel apprehensive about this significant AI revolution. Education is essential to prepare both junior and senior designers for the impact of AI and how to leverage it to enhance our human design skills.”[P13]. Additionally, a few participants envision future AI tools to be more accessible for designers. In other words, they desire AI tools that offer greater visual and intuitive usability, moving beyond sole reliance on text-based inputs. “The recent Apple Vision Pro headset is inspiring. Its visual and intuitive design appears promising. Thus, AI should also evolve to become more visual, intuitive, and universally accessible to all users.” [P12].
5.7.3 Al’s desired role in enhancing design processes. Many participants expressed their desire for GenAI to mature further, supporting them in various design tasks. They hoped that GenAI could assist in designing graphics, writing prompts, generating smart components, providing successful design examples, and automating the handoff of designs to development teams. P4 noted, “I am anticipating easier ways to write prompts or automatically generate components that adapt to different screens, recommend smart animation transitions, and even generate 3D content for immersive environments in the future.” [P4]. Many other participants also voiced their desires for specific GenAI support in various tasks in the future. For instance, they hoped AI could summarize state-of-the-art knowledge for research and strategy (P10, P18, and P19), serve as a synthetic user to expedite sprints (P1 and P18), offer professional guidance in accessibility design (P12), help designers quickly master and switch between different design tools (P18), aid in labeling and clustering qualitative raw data from user studies (P5, P9, P10, P12, and P18), and seamlessly translate designs into development (P5, P18, and P19).
P18 specifically highlighted the constantly changing landscape of design tools, expressing a desire for “GenAI to reduce the time designers spend keeping up with rapidly evolving technology. Design tools change every few years. If AI could help designers transition seamlessly between old and new tools or consolidate everything into a single tool, designers would have more time to understand user needs and design.” [P18]. P5 expressed a wish for “GenAI tools to help me assess the feasibility of designs on the development side during the early design stage and assist in documenting designs in Figma, facilitating smooth handoffs to the development team in a format accessible and understandable to developers. Currently, I still have to manually document designs, take numerous screenshots, and verbally explain them to the development team.” [P5].
5.7.4 The challenge of defining GenAI outcome metrics. According to several participants, another envisioned future for GenAI involves the challenge of defining metrics to measure the quality of AI-generated output. As P17 commented, “If many businesses use the same GenAI tools, it may result in similar designs and lacking diversity and innovation. Also, AI can lead to generating numerous outputs and leave the challenge to us, humans, to choose among them.”[P17]. “It is already a challenge to select and fine-tune AI outcomes. We need guidelines on how to measure or validate AI’s output.” [P18]

6 DISCUSSION

Our key findings indicate that experienced UX Designers acknowledge GenAI’s potential as an assistive tool for repetitive tasks and productivity enhancement, yet they remain cautious about adopting GenAI tools themselves. This is due to their limitations in understanding user-centered design and their potential impact on their creative skills, as well as fundamental limits to enjoyment and agency in human-AI collaboration. While GenAI is seen as a game-changer in addressing challenges within current UXD practice, such as friction in design handoffs and limitations in UXD resources, designers expressed grave concerns about skill degradation, particularly for junior designers. They advocate for GenAI tools to evolve into intuitive, visually-oriented, and collaboratively intelligent systems. This evolution aims to balance human creativity and AI efficiency, emphasizing responsible AI use and human-AI collaboration to preserve human control in creativity and decision-making. We delve into these aspects below, where we discuss implications for human-GenAI collaboration, specifically copyright and ownership, human creativity and agency, and AI literacy and access.

6.1 Enjoyment and Agency in Human-GenAI Collaboration

Enjoyment and meaning are two recurring key words in our conversations with UX Designers. As UX Designers not only aim to excel in their work but also seek to derive pleasure from it. Recent work has found that the user experience of early users of ChatGPT was impacted by hedonic attributes, not just pragmatic ones [88]. Enjoyment in work can be viewed through the lens of motivation theory, including both extrinsic and intrinsic motivations [48]. Extrinsic motivations revolve around whether the technology efficiently serves as a means to accomplish tasks, treating technology such as GenAI as a tool. Intrinsic motivations center on whether UX Designers find their work inherently rewarding and the engagement with technology enjoyable, considering technology such as GenAI as a toy [8]. We argue that, in UX Design, where creativity and problem-solving play central roles, this intrinsic motivation becomes paramount. Designers are not just driven by the end results; they also seek fulfillment and enjoyment throughout the design process. In other words, the future of AI-human collaboration should empower individuals to focus on tasks that truly matter to them, allowing AI to handle more mundane aspects. This approach not only enhances productivity but also fosters job satisfaction [59].
Closely related to enjoyment, we found that agency in human-AI collaboration is also a recurring theme. Will AI take on more agency while humans play a supervisory role? Or will humans retain primary control and have the ability to override AI’s decisions when necessary? Alternatively, will AI and humans collaborate in a mixed-agency approach, where both contribute to decision-making and execution [91]? UX designers believe that we should differentiate between well-defined, routine tasks and complex, creative problem-solving tasks. Humans bring contextual understanding, ethical judgment, and accountability, and should lead complex work that requires creativity, empathy, ethics, and complex decision-making. Tasks led by AI and decisions made by AI should be transparent (cf., deceptive AI ecosystems [105]), explainable [21], and easily adjustable by human collaborators. The ideal scenario involves humans learning from
AI insights, while AI improves based on human feedback (see e.g., [30] for a review on challenges for Reinforcement Learning from Human Feedback). Overall, we found that UX designers’ beliefs align with those of Fanni et al. [26], who advocate for “active human agency” and argue that AI policies should include provisions that enable users to promptly challenge or rectify AI-driven decisions (cf., contestable AI [66]). This approach aims to empower individual autonomy and bolster fundamental human rights in the digital age. We find that the foregoing dimensions of enjoyment and agency to not only be unique human factors, but we suspect will increasingly play a role in the extent that UX Designers retain usage of such tools.

6.2 GenAI Interpolating, Humans Extrapolating, and Creativity Exhaustion

In today’s efficiency-driven agile working culture, UX Design is essential for getting the design right and the right design [12]. It brings together a team to go through each sprint, progressing from idea to prototype to user research within short time frames [18]. GenAI’s remarkable speed in generating content and solutions have the potential to facilitate the fast-moving sprints. However, we must be aware that while GenAI can produce vast quantities of outputs in an instant, humans are the innovators behind these creations. Essentially, AI is interpolating, estimating values or generating outcomes within the range of existing data. Thoring et al. [93] approach this from the lens of evolutionary creativity theory [87], where Variation and Selection are key components. While GenAI interpolates (Variation), it currently lacks the capability of extrapolating out of the latent space and continuing into unknown territory or merging into another latent space without supervisory human control (Selection) [93]. Indeed, AI might excel at “peripheral work” [83] or repetitive tasks within known data patterns. However, since AI is trained with known data, relying heavily on AI can lead to homogenized solutions (e.g., the AI-generated user interfaces all look similar) [6] or amplify biases in the data (e.g., the resulting design favoring certain demographic groups [4, 19, 25]. In contrast, consider children, who receive around four or five orders of magnitude less language data than LLMs, and vastly outperform capabilities of AI [31]. Indeed, we humans excel at extrapolating, making predictions beyond the known data range using not only our creativity, intuition, and domain knowledge, but also benefit from multimodal grounding and the social and interactive nature of received sensory input [31]. In essence, humans excel at the “core work” that they identify with, which contributes to not only their success, but also their happiness [83].
Another potential setback revolves around the danger of human creativity exhaustion in keeping up with the pace of GenAI output. GenAI’s relentless speed can exert immense pressure on human creators. While AI is not directly subject to physical factors like fatigue [3], humans are expected to innovate continuously and rapidly on a large scale. On one hand, AI’s speed sets a challenging standard for human creators; on the other hand, it also lowers the training requirements to become an artist or a creative, as many people can create artistic paintings without picking up a brush. Moreover, as AI learns from human creations and refines its abilities, it raises the bar for what is considered innovative. Human creators in general, and UX Designer in particular, may feel compelled to push their limits to compete with AI, which can be mentally and emotionally draining. To sustain human creativity, instead of competing with AI’s speed, human creators seem to be better off focusing on tasks that require critical thinking and emotional depth, where the human touch is irreplaceable. To that end, support systems and education should be introduced to alleviate the pressure on humans and help mitigate the risk of creativity exhaustion, a point to which we discuss next.

6.3 AI Literacy and Participatory AI in UX Design

In UX Design, there is a growing presence of tools with AI features or supporting AI plugins. AI is gradually being integrated into various stages of the UX workflow [65]. AI literacy begins with a fundamental understanding of AI
concepts, including how machine learning works, the types of AI systems, their capabilities, limitations, and implications. This understanding is crucial for UX designers to harness AI’s power, ensure equal access to AI education and tools, and actively participate in shaping and refining AI technologies, especially those used in their design processes. By adopting a participatory AI approach, UX Designers are part and parcel of the conversation around the future of GenAI development – this enables UX designers to strengthen their feedback and expertise, ensuring that AI tools align effectively with human intentions and UX design practices [57, 63].
Another core element of AI literacy is that UX Designers should be able to think critically and question AI-generated outputs. There is a need for a comprehensive framework or guidelines for evaluating AI outcomes. Such a framework could encompass the assessment of AI’s credibility, performance, accuracy, and impact on user experiences [89], as well as provide a structural means for contestability [66]. The foregoing echoes recent efforts [50] to ensure committees consisting of AI experts, ethics experts, and importantly creative professionals, are brought together to oversee GenAI’s progress. These groups would be best equipped to provide standards and best practices for evaluating GenAI output and mitigating harm on creative professionals (cf., [50]).
In our study, UX Designers emphasized the significance of crediting the original creators of artworks or styles utilized in AI work. They mention the use of blockchain technology (cf., [24]), where of special interest are computational provenance mechanisms [94] that track and document the origins, processes, and transformations of AI-generated content or artifacts. As such, this involves recording information about how a piece of content, such as text, images, or other media, was generated, which AI models and algorithms were used, and any data sources that contributed to the creation of the content. In so doing, this would facilitate the attribution of ownership should they draw on model output throughout their UX practice. We found that many UX Designers in our study believed that humans should be the owners of the design outcomes. However, this was so only if GenAI tools serve as supportive aids, and human designers remain in the driver seat by leading the entire process, adding detailed touches and synthesizing user inputs. However, they recommended attributing ownership to the part of the work supported by AI or labeling it as AI-contributed while ensuring transparency about the tools used. Interestingly, in a recent study, Draxler et al. [20] found that people often do not disclose AI as an author when using personalized AI-generated texts in writing, although they attribute ownership to the AI. Comparing human and AI ghostwriters, they found that attributing authorship to oneself is more prevalent when writing with AI support. Given disagreement in attributing ownership, it is necessary to develop an authorship attribution framework that takes the user and their relationship to the generative model into account.
The discourse around copyright, privacy, ownership for UX design practice is inherently connected to the wider discourse on GenAI as a whole. Whether dealing with text, images, or video, the foregoing raise the question of whether the GenAI output being generated is ethical and legally compliant to begin with. Indeed, recent work by Jiang et al. [50] in assessing the impact of visual GenAI on artists paints a bleak picture regarding the outspoken harms on artists, which can well extend to other creative works (including the work of UX designers). For example, “fair use” within US copyright law may not always end up protecting the artist, whether due to case-by-case determination, or the high cost in pursing legal battles. Whether the way forward constitutes better watermarking approaches to ensure copyright and traceable accountability [94, 107], or the further development of regulation and policy at an (inter-)national level (e.g., European AI Act [97]), it is clear GenAI development needs to be acted upon responsibly – for and beyond
UX Design practice. As such, we believe drawing on the key human-centered principles (e.g., aligned with human well-being, respecting privacy, and having appropriate governance and oversight) of responsible AI development [34] are of immediate necessity to protect designers moving forward.

6.5 The Future of GenAI-infused UX Design: Fears and Opportunities

The rapid advancements in GenAI and UX Design underscore the necessity of ongoing monitoring and research in both fields. As GenAI tools continue to evolve, it is essential to ensure that they align with human values, ethics, and creative processes, despite any uncertainty that may arise from currently unresolved empirical and conceptual issues in human psychology [11]. To that end, by answering our research questions of current (RQ1) and future (RQ2) perceptions of GenAI, this study serves as a foundation for exploring the opportunities and challenges presented by GenAI in UX Design, where we emphasize the importance of responsible and informed collaboration between humans and AI in this domain, which leads to the question: are such fears of GenAI for UX Design warranted?
We observed (Sec. 5.5) that experienced UX Designers’ were defensive about the very idea that “AI can potentially replace human designers”. Experienced designers, on the one hand, have confidence in their superpowers in terms of design skills, the collision of wisdom and enlightenment in human-human communication, and possessing high empathy towards users. However, they emphasized (Sec. 5.6) the real risks that GenAI could pose on the junior UX Design workforce, posing the dilemma: if junior human designers don’t do the mundane work that can essentially be replaced by GenAI, then how will they reach the seniority levels wherein GenAI cannot replace? On the other hand, even experienced designers raised fears about AI in general, and GenAI in particular, and how this may in the future render their expertise partially obsolete. Specifically, some concerns revolved around emergent AI capabilities, where as AI becomes more sophisticated (e.g., more parameters with more diverse and larger training data), there is the possibility of it advancing beyond the intended scope or control [6,80]. On the flip side, some of our participants may not have been aware of the rapid pace of developments – for example, GenAI models can be fine-tuned on unique stakeholder datasets [104], which would elevate ideation, brainstorming, and documentation within the UX design process. Nevertheless, the risks persist should even fine-tuned models lead to situations where GenAI systems generate designs or make decisions that are not aligned with refined user experience principles, stakeholder satisfaction, or worse, violate human values, ethics, or cultural norms [100, 108], even when designers may take the driver seat of selecting and refining the output with additional prompts (cf., [15]) on custom-trained models. These concerns align with user perceptions of machine consciousness (of GPT-3) [84], which lead to several tensions, of which most relevant to our use case are interaction and experience, control and independence, and rigidity and spontaneity. Furthermore, as in digital art, people who exhibit strong anthropocentric creativity beliefs may be biased to brush off GenAI design output as overall less creative and awe-inspiring should they know it was made with GenAI [68].
However, unpredictability, polarity, and such tensions often signify new opportunities as emerging AI capabilities can lead to innovative problem-solving approaches and entirely new design concepts that while inspire human creators, can radically disrupt the practice (reminiscent of paradigmatic shifts in the natural sciences [54]). However, while we already see a radical disruption to several elements of UX Design practice (cf., Fig. 3), much remains to be desired (and cautious about) as GenAI capabilities increase. Essentially, we contend that ensuring harmonious alignment between human-AI collaboration for UX Design practice, methods and strategies are required to ensure that GenAI systems comprehend and respect human designers’ values and interpret and respond to their concerns effectively. For this, we call upon further research in and beyond responsible AI to continue placing the human at the center of technology development, and importantly, on reconciliation efforts should ethical values differ amongst groups [49]. Indeed, as
Rakova et al. [78] argue, to better enable responsible AI work, organizations need to update their prevalent practices, which requires addressing prevalent work practices, emerging work practices, and mapping their aspirational future. To this end, a path forward is to continue drawing on participatory AI and value-sensitive design [5, 21, 33, 47] to create AI for social good [95], and ultimately foster human and creative safeguarding mechanisms, allowing UX Design practice and its practitioners to adapt and flourish gracefully.

6.6 Study Limitations and Future Work

In our investigation, we opted for a breadth-first approach to better understand the overall impact that GenAI may have on UX Design, which naturally comes at the cost of depth – we did not delve deeper into any particular UX products, tools, or workflows. As such, prioritizing breadth over depth was strategic, enabling the capture of a wide range of perspectives and considerations within the UX field. This serves as an important stepping stone for future works to delve into specific tools and use cases to provide a more in-depth analysis. Second, our interviews were limited by the small sample size ( ) of startup participants. Startups often have unique team structures and resource constraints, leading to different approaches to addressing UX Design challenges compared to established companies. This invites future work to study the organizational practices of startup ecosystems, and how GenAI is embedded within such structures. Furthermore, although the majority of our participants had experience using GenAI tools such as chatGPT (GPT-3.5) and Midjourney, the study invited participants to watch demonstrations of GenAI tools without actual hands-on usage. This limited interaction might have overlooked practical insights that could arise only from direct hands-on experience with such tools.

7 CONCLUSION

Our work set out to investigate how UX Designers perceive GenAI tools, and how this is reflected in their current workflows and practices. Further, we identified the opportunities and risks do UX Designers envision for the future of human-AI collaboration in UX Design practice. Through in-depth interviews with 20 UX Designers who have diverse years of experience, we found that experienced designers are confident in their originality, creativity, and empathic skills, and find GenAI’s role overall as assistive. They emphasized factors of “enjoyment” and “agency” as uniquely humans, where humans will always remain in the driver seat over AI output. However, there were serious concerns over setbacks for junior designers, who may be impacted by skill degradation, job replacement, and creativity exhaustion. We draw several implications for responsible human-AI collaboration in UX design, specifically related to copyright and ownership, human creativity, and AI literacy and access. We call on continued efforts to develop responsible AI for ensuring human and creative safeguarding mechanisms, allowing UX Design professionals to immediately yet gracefully adapt to the dawn of GenAI, and its disruptive impact on UX Design practice.

ACKNOWLEDGMENTS

We thank all our participants for taking part in our study, and offering us their valuable perspectives on GenAI.

REFERENCES

[1] Abdallah MH Abbas, Khairil Imran Ghauth, and Choo-Yee Ting. 2022. User experience design using machine learning: a systematic review. IEEE Access 10 (2022), 51501-51514.
[2] Saleema Amershi, Dan Weld, Mihaela Vorvoreanu, Adam Fourney, Besmira Nushi, Penny Collisson, Jina Suh, Shamsi Iqbal, Paul N. Bennett, Kori Inkpen, Jaime Teevan, Ruth Kikin-Gil, and Eric Horvitz. 2019. Guidelines for Human-AI Interaction. In Proceedings of the 2019 CHI Conference
on Human Factors in Computing Systems (Glasgow, Scotland Uk) (CHI ’19). Association for Computing Machinery, New York, NY, USA, 1-13. https://doi.org/10.1145/3290605.3300233
[3] Kareem Ayoub and Kenneth Payne. 2016. Strategy in the age of artificial intelligence. 7ournal of strategic studies 39, 5-6 (2016), 793-819.
[4] Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, and Shmargaret Shmitchell. 2021. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?. In Proceedings of the 2021 ACM Conference on Fairness, Accountability, and Transparency (Virtual Event, Canada) (FAccT ’21). Association for Computing Machinery, New York, NY, USA, 610-623. https://doi.org/10.1145/3442188.3445922
[5] Abeba Birhane, William Isaac, Vinodkumar Prabhakaran, Mark Diaz, Madeleine Clare Elish, Iason Gabriel, and Shakir Mohamed. 2022. Power to the People? Opportunities and Challenges for Participatory AI. In Equity and Access in Algorithms, Mechanisms, and Optimization (Arlington, VA, USA) (EAAMO ’22). Association for Computing Machinery, New York, NY, USA, Article 6, 8 pages. https://doi.org/10.1145/3551624.3555290
[6] Rishi Bommasani, Drew A. Hudson, Ehsan Adeli, Russ Altman, Simran Arora, Sydney von Arx, Michael S. Bernstein, Jeannette Bohg, Antoine Bosselut, Emma Brunskill, Erik Brynjolfsson, Shyamal Buch, Dallas Card, Rodrigo Castellon, Niladri Chatterji, Annie Chen, Kathleen Creel, Jared Quincy Davis, Dora Demszky, Chris Donahue, Moussa Doumbouya, Esin Durmus, Stefano Ermon, John Etchemendy, Kawin Ethayarajh, Li Fei-Fei, Chelsea Finn, Trevor Gale, Lauren Gillespie, Karan Goel, Noah Goodman, Shelby Grossman, Neel Guha, Tatsunori Hashimoto, Peter Henderson, John Hewitt, Daniel E. Ho, Jenny Hong, Kyle Hsu, Jing Huang, Thomas Icard, Saahil Jain, Dan Jurafsky, Pratyusha Kalluri, Siddharth Karamcheti, Geoff Keeling, Fereshte Khani, Omar Khattab, Pang Wei Koh, Mark Krass, Ranjay Krishna, Rohith Kuditipudi, Ananya Kumar, Faisal Ladhak, Mina Lee, Tony Lee, Jure Leskovec, Isabelle Levent, Xiang Lisa Li, Xuechen Li, Tengyu Ma, Ali Malik, Christopher D. Manning, Suvir Mirchandani, Eric Mitchell, Zanele Munyikwa, Suraj Nair, Avanika Narayan, Deepak Narayanan, Ben Newman, Allen Nie, Juan Carlos Niebles, Hamed Nilforoshan, Julian Nyarko, Giray Ogut, Laurel Orr, Isabel Papadimitriou, Joon Sung Park, Chris Piech, Eva Portelance, Christopher Potts, Aditi Raghunathan, Rob Reich, Hongyu Ren, Frieda Rong, Yusuf Roohani, Camilo Ruiz, Jack Ryan, Christopher Ré, Dorsa Sadigh, Shiori Sagawa, Keshav Santhanam, Andy Shih, Krishnan Srinivasan, Alex Tamkin, Rohan Taori, Armin W. Thomas, Florian Tramèr, Rose E. Wang, William Wang, Bohan Wu, Jiajun Wu, Yuhuai Wu, Sang Michael Xie, Michihiro Yasunaga, Jiaxuan You, Matei Zaharia, Michael Zhang, Tianyi Zhang, Xikun Zhang, Yuhui Zhang, Lucia Zheng, Kaitlyn Zhou, and Percy Liang. 2022. On the Opportunities and Risks of Foundation Models. arXiv:2108.07258 [cs.LG]
[7] Zalán Borsos, Raphaël Marinier, Damien Vincent, Eugene Kharitonov, Olivier Pietquin, Matt Sharifi, Olivier Teboul, David Grangier, Marco Tagliasacchi, and Neil Zeghidour. 2022. AudioLM: a Language Modeling Approach to Audio Generation. arXiv:2209.03143 [cs.SD]
[8] Petter Bae Brandtzæg, Asbjørn Følstad, and Jan Heim. 2018. Enjoyment: Lessons from Karasek. Springer International Publishing, Cham, 331-341. https://doi.org/10.1007/978-3-319-68213-6_21
[9] Virginia Braun and Victoria Clarke. 2012. Thematic analysis. In APA Handbook of Research Methods in Psychology: Vol. 2. Research Designs. American Psychological Association, 750 First St. NE, Washington, DC 20002-4242, 57-71.
[10] Lizzy Burnam. 2023. We Surveyed 1093 Researchers About How They Use AI-Here’s What We Learned. https://www.userinterviews.com/blog/ai-in-ux-research-report
[11] Patrick Butlin. 2021. AI Alignment and Human Reward. In Proceedings of the 2021 AAAI/ACM Conference on AI, Ethics, and Society (Virtual Event, USA) (AIES ’21). Association for Computing Machinery, New York, NY, USA, 437-445. https://doi.org/10.1145/3461702.3462570
[12] Bill Buxton. 2010. Sketching user experiences: getting the design right and the right design. Morgan kaufmann, Burlington, Massachusetts, USA.
[13] Kelly Caine. 2016. Local Standards for Sample Size at CHI. In Proceedings of the 2016 CHI Conference on Human Factors in Computing Systems (San Jose, California, USA) (CHI ’16). Association for Computing Machinery, New York, NY, USA, 981-992. https://doi.org/10.1145/2858036.2858498
[14] Tara Capel and Margot Brereton. 2023. What is Human-Centered about Human-Centered AI? A Map of the Research Landscape. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 359, 23 pages. https://doi.org/10.1145/3544548.3580959
[15] Minsuk Chang, Stefania Druga, Alexander J. Fiannaca, Pedro Vergani, Chinmay Kulkarni, Carrie J Cai, and Michael Terry. 2023. The Prompt Artists. In Proceedings of the 15th Conference on Creativity and Cognition (, Virtual Event, USA,) (C&C’23). Association for Computing Machinery, New York, NY, USA, 75-87. https://doi.org/10.1145/3591196.3593515
[16] Li-Yuan Chiou, Peng-Kai Hung, Rung-Huei Liang, and Chun-Teng Wang. 2023. Designing with AI: An Exploration of Co-Ideation with Image Generators. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 1941-1954. https://doi.org/10.1145/3563657.3596001
[17] Victoria Clarke, Virginia Braun, and Nikki Hayfield. 2015. Thematic analysis. Qualitative psychology: A practical guide to research methods 222, 2015 (2015), 248.
[18] Tiago Silva Da Silva, Milene Selbach Silveira, Frank Maurer, and Theodore Hellmann. 2012. User experience design and agile development: From theory to practice. Fournal of Software Engineering and Applications 5, 10 (2012), 743-751. https://doi.org/10.4236/jsea.2012.510087
[19] Smit Desai, Tanusree Sharma, and Pratyasha Saha. 2023. Using ChatGPT in HCI Research-A Trioethnography. In Proceedings of the 5th International Conference on Conversational User Interfaces (Eindhoven, Netherlands) (CUI ’23). Association for Computing Machinery, New York, NY, USA, Article 8, 6 pages. https://doi.org/10.1145/3571884.3603755
[20] Fiona Draxler, Anna Werner, Florian Lehmann, Matthias Hoppe, Albrecht Schmidt, Daniel Buschek, and Robin Welsch. 2024. The AI Ghostwriter Effect: When Users do not Perceive Ownership of AI-Generated Text but Self-Declare as Authors. ACM Transactions on Computer-Human Interaction 31, 2 (Feb. 2024), 1-40. https://doi.org/10.1145/3637875
[21] Upol Ehsan and Mark O. Riedl. 2020. Human-Centered Explainable AI: Towards a Reflective Sociotechnical Approach. In HCI International 2020 Late Breaking Papers: Multimodality and Intelligence, Constantine Stephanidis, Masaaki Kurosu, Helmut Degen, and Lauren Reinerman-Jones (Eds.).
Springer International Publishing, Cham, 449-466.
[22] Damian Okaibedi Eke. 2023. ChatGPT and the rise of generative AI: Threat to academic integrity? Journal of Responsible Technology 13 (2023), 100060.
[23] Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock. 2023. GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models. arXiv:2303.10130 [econ.GN]
[24] Chris Elsden, Arthi Manohar, Jo Briggs, Mike Harding, Chris Speed, and John Vines. 2018. Making Sense of Blockchain Applications: A Typology for HCI. In Proceedings of the 2018 CHI Conference on Human Factors in Computing Systems (Montreal QC, Canada) (CHI ’18). Association for Computing Machinery, New York, NY, USA, 1-14. https://doi.org/10.1145/3173574.3174032
[25] Ziv Epstein, Aaron Hertzmann, Investigators of Human Creativity, Memo Akten, Hany Farid, Jessica Fjeld, Morgan R Frank, Matthew Groh, Laura Herman, Neil Leach, et al. 2023. Art and the science of generative AI. Science 380, 6650 (2023), 1110-1111.
[26] Rosanna Fanni, Valerie Steinkogler, Giulia Zampedri, and Jo Pierson. 2022. Enhancing human agency through redress in Artificial Intelligence Systems. AI & SOCIETY 38 (06 2022). https://doi.org/10.1007/s00146-022-01454-7
[27] Ed Felten, Manav Raj, and Robert Seamans. 2023. How will Language Modelers like ChatGPT Affect Occupations and Industries? arXiv:2303.01157 [econ.GN]
[28] K. J. Kevin Feng, Maxwell James Coppock, and David W. McDonald. 2023. How Do UX Practitioners Communicate AI as a Design Material? Artifacts, Conceptions, and Propositions. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (, Pittsburgh, PA, USA,) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 2263-2280. https://doi.org/10.1145/3563657.3596101
[29] Jennifer Fereday and Eimear Muir-Cochrane. 2006. Demonstrating Rigor Using Thematic Analysis: A Hybrid Approach of Inductive and Deductive Coding and Theme Development. International Journal of Qualitative Methods 5, 1 (2006), 80-92. https://doi.org/10.1177/160940690600500107 arXiv:https://doi.org/10.1177/160940690600500107
[30] Giorgio Franceschelli and Mirco Musolesi. 2023. Reinforcement Learning for Generative AI: State of the Art, Opportunities and Open Research Challenges. arXiv:2308.00031 [cs.LG]
[31] Michael C. Frank. 2023. Bridging the data gap between children and large language models. Trends in Cognitive Sciences 27, 11 (2023), 990-992. https://doi.org/10.1016/j.tics.2023.08.007
[32] Jonas Frich, Lindsay MacDonald Vermeulen, Christian Remy, Michael Mose Biskjaer, and Peter Dalsgaard. 2019. Mapping the Landscape of Creativity Support Tools in HCI. In Proceedings of the 2019 CHI Conference on Human Factors in Computing Systems (Glasgow, Scotland Uk) (CHI ’19). Association for Computing Machinery, New York, NY, USA, 1-18. https://doi.org/10.1145/3290605.3300619
[33] Batya Friedman. 1996. Value-Sensitive Design. Interactions 3, 6 (dec 1996), 16-23. https://doi.org/10.1145/242485.242493
[34] Ozlem Ozmen Garibay, Brent Winslow, Salvatore Andolina, Margherita Antona, Anja Bodenschatz, Constantinos Coursaris, Gregory Falco, Stephen M. Fiore, Ivan Garibay, Keri Grieman, John C. Havens, Marina Jirotka, Hernisa Kacorri, Waldemar Karwowski, Joe Kider, Joseph Konstan, Sean Koon, Monica Lopez-Gonzalez, Iliana Maifeld-Carucci, Sean McGregor, Gavriel Salvendy, Ben Shneiderman, Constantine Stephanidis, Christina Strobel, Carolyn Ten Holter, and Wei Xu. 2023. Six Human-Centered Artificial Intelligence Grand Challenges. International fournal of Human Computer Interaction 39, 3 (2023), 391-437. https://doi.org/10.1080/10447318.2022.2153320 arXiv:https://doi.org/10.1080/10447318.2022.2153320
[35] Frederic Gmeiner, Humphrey Yang, Lining Yao, Kenneth Holstein, and Nikolas Martelaro. 2023. Exploring Challenges and Opportunities to Support Designers in Learning to Co-create with AI-based Manufacturing Design Tools. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 226, 20 pages. https://doi.org/10.1145/3544548.3580999
[36] Abid Haleem, Mohd Javaid, and Ravi Pratap Singh. 2022. An era of ChatGPT as a significant futuristic support tool: A study on features, abilities, and challenges. BenchCouncil Transactions on Benchmarks, Standards and Evaluations 2, 4 (2022), 100089. https://doi.org/10.1016/j.tbench.2023.100089
[37] Perttu Hämäläinen, Mikke Tavast, and Anton Kunnari. 2023. Evaluating Large Language Models in Generating Synthetic HCI Research Data: a Case Study. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 433, 19 pages. https://doi.org/10.1145/3544548.3580688
[38] Ariel Han and Zhenyao Cai. 2023. Design Implications of Generative AI Systems for Visual Storytelling for Young Learners. In Proceedings of the 22nd Annual ACM Interaction Design and Children Conference (Chicago, IL, USA) (IDC ’23). Association for Computing Machinery, New York, NY, USA, 470-474. https://doi.org/10.1145/3585088.3593867
[39] Gunnar Harboe and Elaine M. Huang. 2015. Real-World Affinity Diagramming Practices: Bridging the Paper-Digital Gap. In Proceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems (Seoul, Republic of Korea) (CHI ’15). Association for Computing Machinery, New York, NY, USA, 95-104. https://doi.org/10.1145/2702123.2702561
[40] Rex Hartson and Pardha Pyla. 2018. The UX Book: Process and Guidelines for Ensuring a Quality User Experience (2st ed.). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA.
[41] Monique Hennink and Bonnie N. Kaiser. 2022. Sample sizes for saturation in qualitative research: A systematic review of empirical tests. Social Science & Medicine 292 (2022), 114523. https://doi.org/10.1016/j.socscimed.2021.114523
[42] Jonathan Ho, William Chan, Chitwan Saharia, Jay Whang, Ruiqi Gao, Alexey Gritsenko, Diederik P. Kingma, Ben Poole, Mohammad Norouzi, David J. Fleet, and Tim Salimans. 2022. Imagen Video: High Definition Video Generation with Diffusion Models. arXiv:2210.02303 [cs.CV]
[43] Matthew K. Hong, Shabnam Hakimi, Yan-Ying Chen, Heishiro Toyoda, Charlene Wu, and Matt Klenk. 2023. Generative AI for Product Design: Getting the Right Design and the Design Right. arXiv:2306.01217 [cs.HC]
[44] Eric Horvitz. 1999. Principles of mixed-initiative user interfaces. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (Pittsburgh, Pennsylvania, USA) (CHI ’99). Association for Computing Machinery, New York, NY, USA, 159-166. https://doi.org/10.1145/302979. 303030
[45] Stephanie Houde, Steven I Ross, Michael Muller, Mayank Agarwal, Fernando Martinez, John Richards, Kartik Talamadupula, and Justin D Weisz. 2022. Opportunities for Generative AI in UX Modernization. In 7oint Proceedings of the ACM IUI Workshops 2022, March 2022, Helsinki, Finland. Association for Computing Machinery, New York, NY, USA.
[46] Lei Huang, Weijiang Yu, Weitao Ma, Weihong Zhong, Zhangyin Feng, Haotian Wang, Qianglong Chen, Weihua Peng, Xiaocheng Feng, Bing Qin, and Ting Liu. 2023. A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions. arXiv:2311.05232 [cs.CL]
[47] Nanna Inie, Jeanette Falk, and Steve Tanimoto. 2023. Designing Participatory AI: Creative Professionals’ Worries and Expectations about Generative AI. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI EA ’23). Association for Computing Machinery, New York, NY, USA, Article 82, 8 pages. https://doi.org/10.1145/3544549.3585657
[48] Alice M Isen and Johnmarshall Reeve. 2005. The influence of positive affect on intrinsic and extrinsic motivation: Facilitating enjoyment of play, responsible work behavior, and self-control. Motivation and emotion 29 (2005), 295-323.
[49] Maurice Jakesch, Zana Buçinca, Saleema Amershi, and Alexandra Olteanu. 2022. How Different Groups Prioritize Ethical Values for Responsible AI. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency (Seoul, Republic of Korea) (FAccT ’22). Association for Computing Machinery, New York, NY, USA, 310-323. https://doi.org/10.1145/3531146.3533097
[50] Harry H. Jiang, Lauren Brown, Jessica Cheng, Mehtab Khan, Abhishek Gupta, Deja Workman, Alex Hanna, Johnathan Flowers, and Timnit Gebru. 2023. AI Art and Its Impact on Artists. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society (Montréal, QC, Canada) (AIES ’23). Association for Computing Machinery, New York, NY, USA, 363-374. https://doi.org/10.1145/3600211.3604681
[51] Zachary Kaiser. 2019. Creativity as computation: Teaching design in the age of automation. Design and Culture 11, 2 (2019), 173-192.
[52] Taenyun Kim, Maria D Molina, Minjin Rheu, Emily S Zhan, and Wei Peng. 2023. One AI Does Not Fit All: A Cluster Analysis of the Laypeople’s Perception of AI Roles. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 29, 1-20 pages. https://doi.org/10.1145/3544548.3581340
[53] Joke Kort, APOS Vermeeren, and Jenneke E Fokker. 2007. Conceptualizing and measuring user experience. , 57-64 pages.
[54] Thomas S. Kuhn. 1962. The Structure of Scientific Revolutions. University of Chicago Press, Chicago.
[55] Chinmay Kulkarni, Stefania Druga, Minsuk Chang, Alex Fiannaca, Carrie Cai, and Michael Terry. 2023. A Word is Worth a Thousand Pictures: Prompts as AI Design Material. arXiv:2303.12647 [cs.HC]
[56] Tomas Lawton, Kazjon Grace, and Francisco J Ibarrola. 2023. When is a Tool a Tool? User Perceptions of System Agency in Human-AI Co-Creative Drawing. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 1978-1996. https://doi.org/10.1145/3563657.3595977
[57] Mina Lee, Megha Srivastava, Amelia Hardy, John Thickstun, Esin Durmus, Ashwin Paranjape, Ines Gerard-Ursin, Xiang Lisa Li, Faisal Ladhak, Frieda Rong, Rose E. Wang, Minae Kwon, Joon Sung Park, Hancheng Cao, Tony Lee, Rishi Bommasani, Michael Bernstein, and Percy Liang. 2024. Evaluating Human-Language Model Interaction. arXiv:2212.09746 [cs.CL]
[58] Florian Lehmann. 2023. Mixed-Initiative Interaction with Computational Generative Systems. In Extended Abstracts of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHIEA ’23). Association for Computing Machinery, New York, NY, USA, Article 501, 6 pages. https://doi.org/10.1145/3544549.3577061
[59] Robert W Lent and Steven D Brown. 2006. Integrating person and situation perspectives on work satisfaction: A social-cognitive view. 7ournal of vocational behavior 69, 2 (2006), 236-247.
[60] Joseph CR Licklider. 1960. Man-Computer Symbiosis. IRE Transactions on Human Factors in Electronics HFE-1, 1(1960), 4-11.
[61] Lassi A. Liikkanen. 2019. It AIn’t Nuttin’ New – Interaction Design Practice After the AI Hype. In Human-Computer Interaction – INTERACT 2019, David Lamas, Fernando Loizides, Lennart Nacke, Helen Petrie, Marco Winckler, and Panayiotis Zaphiris (Eds.). Springer International Publishing, Cham, 600-604.
[62] Zhiyu Lin, Upol Ehsan, Rohan Agarwal, Samihan Dani, Vidushi Vashishth, and Mark Riedl. 2023. Beyond Prompts: Exploring the Design Space of Mixed-Initiative Co-Creativity Systems. arXiv:2305.07465 [cs.AI]
[63] Duri Long and Brian Magerko. 2020. What is AI Literacy? Competencies and Design Considerations. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (Honolulu, HI, USA) (CHI ’20). Association for Computing Machinery, New York, NY, USA, 1-16. https://doi.org/10.1145/3313831.3376727
[64] Robyn Longhurst. 2003. Semi-structured interviews and focus groups. Key methods in geography 3, 2 (2003), 143-156.
[65] Yuwen Lu, Chengzhi Zhang, Iris Zhang, and Toby Jia-Jun Li. 2022. Bridging the Gap Between UX Practitioners’ Work Practices and AI-Enabled Design Support Tools. In Extended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems (New Orleans, LA, USA) (CHI EA ’22). Association for Computing Machinery, New York, NY, USA, Article 268, 7 pages. https://doi.org/10.1145/3491101.3519809
[66] Henrietta Lyons, Eduardo Velloso, and Tim Miller. 2021. Conceptualising Contestability: Perspectives on Contesting Algorithmic Decisions. Proc. ACM Hum.-Comput. Interact. 5, CSCW1, Article 106 (apr 2021), 25 pages. https://doi.org/10.1145/3449180
[67] Nora McDonald, Sarita Schoenebeck, and Andrea Forte. 2019. Reliability and Inter-rater Reliability in Qualitative Research: Norms and Guidelines for CSCW and HCI Practice. Proceedings of the ACM on Human-Computer Interaction 3 (11 2019), 1-23. https://doi.org/10.1145/3359174
[68] Kobe Millet, Florian Buehler, Guanzhong Du, and Michail D. Kokkoris. 2023. Defending humankind: Anthropocentric bias in the appreciation of AI art. Computers in Human Behavior 143 (2023), 107707. https://doi.org/10.1016/j.chb.2023.107707
[69] Bin Ning, Fang Liu, and Zhixiong Liu. 2023. Creativity Support in AI Co-creative Tools: Current Research, Challenges and Opportunities. In 2023 26th International Conference on Computer Supported Cooperative Work in Design (CSCWD). IEEE, New York, NY, USA, 5-10. https: //doi.org/10.1109/CSCWD57460.2023.10152832
[70] Lorelli S. Nowell, Jill M. Norris, Deborah E. White, and Nancy J. Moules. 2017. Thematic Analysis: Striving to Meet the Trustworthiness Criteria. International fournal of Qualitative Methods 16, 1 (2017), 1609406917733847. https://doi.org/10.1177/1609406917733847 arXiv:https://doi.org/10.1177/1609406917733847
[71] Shakked Noy and Whitney Zhang. 2023. Experimental evidence on the productivity effects of generative artificial intelligence. Available at SSRN 4375283 381, 6654 (2023), 187-192. https://doi.org/10.1126/science.adh2586
[72] OpenAI, Josh Achiam, Steven Adler, and Sandhini Agarwal et al. 2023. GPT-4 Technical Report. arXiv:2303.08774 [cs.CL]
[73] Antti Oulasvirta, Niraj Ramesh Dayama, Morteza Shiripour, Maximilian John, and Andreas Karrenbauer. 2020. Combinatorial optimization of graphical user interface designs. Proc. IEEE 108, 3 (2020), 434-464.
[74] Joon Sung Park, Lindsay Popowski, Carrie Cai, Meredith Ringel Morris, Percy Liang, and Michael S. Bernstein. 2022. Social Simulacra: Creating Populated Prototypes for Social Computing Systems. In Proceedings of the 35th Annual ACM Symposium on User Interface Software and Technology (Bend, OR, USA) (UIST ’22). Association for Computing Machinery, New York, NY, USA, Article 74, 18 pages. https://doi.org/10.1145/3526113.3545616
[75] John V Pavlik. 2023. Collaborating with ChatGPT: Considering the implications of generative artificial intelligence for journalism and media education. Journalism & Mass Communication Educator 78, 1 (2023), 84-93.
[76] Andrew Pearson. 2023. The rise of CreAltives: Using AI to enable and speed up the creative process. Journal of AI, Robotics & Workplace Automation 2, 2 (2023), 101-114.
[77] Sida Peng, Eirini Kalliamvakou, Peter Cihon, and Mert Demirer. 2023. The Impact of AI on Developer Productivity: Evidence from GitHub Copilot. arXiv:2302.06590 [cs.SE]
[78] Bogdana Rakova, Jingying Yang, Henriette Cramer, and Rumman Chowdhury. 2021. Where Responsible AI Meets Reality: Practitioner Perspectives on Enablers for Shifting Organizational Practices. Proc. ACM Hum.-Comput. Interact. 5, CSCW1, Article 7 (apr 2021), 23 pages. https://doi.org/10. 1145/3449081
[79] Kevin Roose. 2022. The brilliance and weirdness of ChatGPT. https://www.nytimes.com/2022/12/05/technology/chatgpt-ai-twitter.html
[80] S. Russell. 2019. Human Compatible: Artificial Intelligence and the Problem of Control. Penguin Publishing Group, New York, NY, USA. https: //books.google.nl/books?id=M1eFDwAAQBAJ
[81] Jana I Saadi and Maria C Yang. 2023. Generative Design: Reframing the Role of the Designer in Early-Stage Design Process. 7ournal of Mechanical Design 145, 4 (2023), 041411.
[82] Chitwan Saharia, William Chan, Saurabh Saxena, Lala Li, Jay Whang, Emily L Denton, Kamyar Ghasemipour, Raphael Gontijo Lopes, Burcu Karagol Ayan, Tim Salimans, et al. 2022. Photorealistic text-to-image diffusion models with deep language understanding. Advances in Neural Information Processing Systems 35 (2022), 36479-36494.
[83] NANNA SANDBERG, TOM HOY, and MARTIN ORTLIEB. 2019. My AI versus the company AI: how knowledge workers conceptualize forms of AI assistance in the workplace. In Ethnographic Praxis in Industry Conference Proceedings, Vol. 2019. Wiley Online Library, American Anthropological Association, Arlington, Virginia, USA, 125-143.
[84] Ava Elizabeth Scott, Daniel Neumann, Jasmin Niess, and Paweł W. Woźniak. 2023. Do You Mind? User Perceptions of Machine Consciousness. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (, Hamburg, Germany, ) (CHI ’23). Association for Computing Machinery, New York, NY, USA, Article 374, 19 pages. https://doi.org/10.1145/3544548.3581296
[85] Ben Shneiderman. 2020. Human-centered artificial intelligence: Reliable, safe & trustworthy. International fournal of Human-Computer Interaction 36, 6 (2020), 495-504.
[86] Ben Shneiderman. 2020. Human-centered artificial intelligence: Three fresh ideas. AIS Transactions on Human-Computer Interaction 12, 3 (2020), 109-124.
[87] D.K. Simonton. 1999. Origins of Genius: Darwinian Perspectives on Creativity. Oxford University Press, Oxford, UK. https://books.google.li/books? id=a2LnBwAAQBAJ
[88] Marita Skjuve, Asbjørn Følstad, and Petter Bae Brandtzaeg. 2023. The User Experience of ChatGPT: Findings from a Questionnaire Study of Early Users. In Proceedings of the 5th International Conference on Conversational User Interfaces (Eindhoven, Netherlands) (CUI ’23). Association for Computing Machinery, New York, NY, USA, Article 2, 10 pages. https://doi.org/10.1145/3571884.3597144
[89] Irene Solaiman, Zeerak Talat, William Agnew, Lama Ahmad, Dylan Baker, Su Lin Blodgett, Hal Daumé III au2, Jesse Dodge, Ellie Evans, Sara Hooker, Yacine Jernite, Alexandra Sasha Luccioni, Alberto Lusoli, Margaret Mitchell, Jessica Newman, Marie-Therese Png, Andrew Strait, and Apostol Vassilev. 2023. Evaluating the Social Impact of Generative AI Systems in Systems and Society. arXiv:2306.05949 [cs.CY]
[90] Åsne Stige, Efpraxia D Zamani, Patrick Mikalef, and Yuzhen Zhu. 2023. Artificial intelligence (AI) for user experience (UX) design: a systematic literature review and future research agenda. https://doi.org/10.1108/ITP-07-2022-0519
[91] S Shyam Sundar. 2020. Rise of machine agency: A framework for studying the psychology of human-AI interaction (HAII). Fournal of ComputerMediated Communication 25, 1 (2020), 74-88.
[92] Jakob Tholander and Martin Jonsson. 2023. Design Ideation with AI – Sketching, Thinking and Talking with Generative Machine Learning Models. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 1930-1940. https://doi.org/10.1145/3563657.3596014
[93] Katja Thoring, Sebastian Huettemann, and Roland M Mueller. 2023. THE AUGMENTED DESIGNER: A RESEARCH AGENDA FOR GENERATIVE AI-ENABLED DESIGN. Proceedings of the Design Society 3 (2023), 3345-3354.
[94] Joel E Tohline et al. 2008. Computational provenance. Computing in Science & Engineering 10, 03 (2008), 9-10.
[95] Steven Umbrello and Ibo Poel. 2021. Mapping value sensitive design onto AI for social good principles. AI and Ethics 1 (08 2021), 3. https: //doi.org/10.1007/s43681-021-00038-3
[96] Russ Unger and Carolyn Chandler. 2012. A Project Guide to UX Design: For user experience designers in the field or in the making. New Riders, Nora, Indianapolis, USA.
[97] Michael Veale and Frederik Zuiderveen Borgesius. 2021. Demystifying the Draft EU Artificial Intelligence Act – Analysing the good, the bad, and the unclear elements of the proposed approach. Computer Law Review International 22, 4 (Aug. 2021), 97-112. https://doi.org/10.9785/cri-2021-220402
[98] Florent Vinchon, Todd Lubart, Sabrina Bartolotta, Valentin Gironnay, Marion Botella, Samira Bourgeois-Bougrine, Jean-Marie Burkhardt, Nathalie Bonnardel, Giovanni Emanuele Corazza, Vlad Glăveanu, et al. 2023. Artificial Intelligence & Creativity: A manifesto for collaboration. The Journal of Creative Behavior 57, 4 (2023), 472-484.
[99] Qian Wan and Zhicong Lu. 2023. GANCollage: A GAN-Driven Digital Mood Board to Facilitate Ideation in Creativity Support. In Proceedings of the 2023 ACM Designing Interactive Systems Conference (Pittsburgh, PA, USA) (DIS ’23). Association for Computing Machinery, New York, NY, USA, 136-146. https://doi.org/10.1145/3563657.3596072
[100] Norbert Wiener. 1960. Some Moral and Technical Consequences of Automation. Science 131, 3410 (1960), 1355-1358. https://doi.org/10.1126/ science.131.3410.1355 arXiv:https://www.science.org/doi/pdf/10.1126/science.131.3410.1355
[101] Wen Xu and Katina Zammit. 2020. Applying Thematic Analysis to Education: A Hybrid Approach to Interpreting Data in Practitioner Research. International Fournal of Qualitative Methods 19 (2020), 1609406920918810. https://doi.org/10.1177/1609406920918810 arXiv:https://doi.org/10.1177/1609406920918810
[102] Bin Yang, Long Wei, and Zihan Pu. 2020. Measuring and improving user experience through artificial intelligence-aided design. Frontiers in Psychology 11 (2020), 595374.
[103] Qian Yang, Aaron Steinfeld, Carolyn Rosé, and John Zimmerman. 2020. Re-examining Whether, Why, and How Human-AI Interaction Is Uniquely Difficult to Design. In Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (, Honolulu, HI, USA, ) (CHI ’20). Association for Computing Machinery, New York, NY, USA, 1-13. https: //doi.org/10.1145/3313831.3376301
[104] Yang Yuan. 2023. On the Power of Foundation Models. In Proceedings of the 40th International Conference on Machine Learning (Proceedings of Machine Learning Research, Vol. 202), Andreas Krause, Emma Brunskill, Kyunghyun Cho, Barbara Engelhardt, Sivan Sabato, and Jonathan Scarlett (Eds.). PMLR, USA, 40519-40530. https://proceedings.mlr.press/v202/yuan23b.html
[105] Xiao Zhan, Yifan Xu, and Stefan Sarkadi. 2023. Deceptive AI Ecosystems: The Case of ChatGPT. In Proceedings of the 5th International Conference on Conversational User Interfaces (Eindhoven, Netherlands) (CUI ’23). Association for Computing Machinery, New York, NY, USA, Article 3, 6 pages. https://doi.org/10.1145/3571884.3603754
[106] Chengzhi Zhang, Weijie Wang, Paul Pangaro, Nikolas Martelaro, and Daragh Byrne. 2023. Generative Image AI Using Design Sketches as Input: Opportunities and Challenges. In Proceedings of the 15th Conference on Creativity and Cognition (Virtual Event, USA) (C&C ’23). Association for Computing Machinery, New York, NY, USA, 254-261. https://doi.org/10.1145/3591196.3596820
[107] Haonan Zhong, Jiamin Chang, Ziyue Yang, Tingmin Wu, Pathum Chamikara Mahawaga Arachchige, Chehara Pathmabandu, and Minhui Xue. 2023. Copyright Protection and Accountability of Generative AI: Attack, Watermarking and Attribution. In Companion Proceedings of the ACM Web Conference 2023 (Austin, TX, USA) (WWW’23 Companion). Association for Computing Machinery, New York, NY, USA, 94-98. https://doi.org/10.1145/3543873.3587321
[108] Simon Zhuang and Dylan Hadfield-Menell. 2020. Consequences of Misaligned AI. In Advances in Neural Information Processing Systems, H. Larochelle, M. Ranzato, R. Hadsell, M.F. Balcan, and H. Lin (Eds.), Vol. 33. Curran Associates, Inc., Red Hook, New York, USA, 15763-15773.

  1. ChatGPT: https://chat.openai.com
    DALL-E 2: https://openai.com/dall-e-2
    Stable Diffusion: https://stablediffusionweb.com
    Midjourney: https://www.midjourney.com
    Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the owner/author(s).
    © 2024 Copyright held by the owner/author(s).
    Manuscript submitted to ACM