Published November 1, 2023
| Version v1
Publication
Open
Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence
Creators
- 1. Sungkyunkwan University
- 2. Benha University
- 3. Suez University
- 4. Center for Research in Molecular Medicine and Chronic Diseases
- 5. Universidade de Santiago de Compostela
- 6. University of Padua
- 7. University of Pisa
- 8. Association of Electronic and Information Technologies
- 9. University of the Basque Country
- 10. Instituto Andaluz de Ciencias de la Tierra
- 11. Universidad de Granada
Description
Artificial intelligence (AI) is currently being utilized in a wide range of sophisticated applications, but the outcomes of many AI models are challenging to comprehend and trust due to their black-box nature. Usually, it is essential to understand the reasoning behind an AI model's decision-making. Thus, the need for eXplainable AI (XAI) methods for improving trust in AI models has arisen. XAI has become a popular research subject within the AI field in recent years. Existing survey papers have tackled the concepts of XAI, its general terms, and post-hoc explainability methods but there have not been any reviews that have looked at the assessment methods, available tools, XAI datasets, and other related aspects. Therefore, in this comprehensive study, we provide readers with an overview of the current research and trends in this rapidly emerging area with a case study example. The study starts by explaining the background of XAI, common definitions, and summarizing recently proposed techniques in XAI for supervised machine learning. The review divides XAI techniques into four axes using a hierarchical categorization system: (i) data explainability, (ii) model explainability, (iii) post-hoc explainability, and (iv) assessment of explanations. We also introduce available evaluation metrics as well as open-source packages and datasets with future research directions. Then, the significance of explainability in terms of legal demands, user viewpoints, and application orientation is outlined, termed as XAI concerns. This paper advocates for tailoring explanation content to specific user types. An examination of XAI techniques and evaluation was conducted by looking at 410 critical articles, published between January 2016 and October 2022, in reputed journals and using a wide range of research databases as a source of information. The article is aimed at XAI researchers who are interested in making their AI models more trustworthy, as well as towards researchers from other disciplines who are looking for effective XAI methods to complete tasks with confidence while communicating meaning from data.
Translated Descriptions
⚠️
This is an automatic machine translation with an accuracy of 90-95%
Translated Description (Arabic)
يتم حاليًا استخدام الذكاء الاصطناعي (AI) في مجموعة واسعة من التطبيقات المتطورة، لكن نتائج العديد من نماذج الذكاء الاصطناعي يصعب فهمها والثقة بها نظرًا لطبيعتها السوداء. عادة، من الضروري فهم الأسباب الكامنة وراء صنع القرار في نموذج الذكاء الاصطناعي. وبالتالي، نشأت الحاجة إلى أساليب الذكاء الاصطناعي القابلة للتفسير (XAI) لتحسين الثقة في نماذج الذكاء الاصطناعي. أصبح XAI موضوعًا بحثيًا شائعًا في مجال الذكاء الاصطناعي في السنوات الأخيرة. تناولت أوراق المسح الحالية مفاهيم XAI ومصطلحاتها العامة وطرق التفسير اللاحقة، ولكن لم تكن هناك أي مراجعات نظرت في طرق التقييم والأدوات المتاحة ومجموعات بيانات XAI والجوانب الأخرى ذات الصلة. لذلك، في هذه الدراسة الشاملة، نقدم للقراء نظرة عامة على البحث الحالي والاتجاهات في هذا المجال الناشئ بسرعة مع مثال على دراسة الحالة. تبدأ الدراسة بشرح خلفية XAI، والتعاريف الشائعة، وتلخيص التقنيات المقترحة مؤخرًا في XAI للتعلم الآلي الخاضع للإشراف. تقسم المراجعة تقنيات XAI إلى أربعة محاور باستخدام نظام تصنيف هرمي: (1) قابلية شرح البيانات، (2) قابلية شرح النموذج، (3) قابلية التفسير اللاحق، و (4) تقييم التفسيرات. نقدم أيضًا مقاييس التقييم المتاحة بالإضافة إلى حزم ومجموعات بيانات مفتوحة المصدر مع اتجاهات البحث المستقبلية. بعد ذلك، يتم تحديد أهمية قابلية التفسير من حيث المطالب القانونية ووجهات نظر المستخدم وتوجه التطبيق، والتي يطلق عليها مخاوف XAI. تدعو هذه الورقة إلى تخصيص محتوى الشرح لأنواع محددة من المستخدمين. تم إجراء فحص لتقنيات XAI وتقييمها من خلال النظر في 410 مقالات نقدية، نُشرت بين يناير 2016 وأكتوبر 2022، في مجلات مرموقة واستخدام مجموعة واسعة من قواعد البيانات البحثية كمصدر للمعلومات. تستهدف المقالة باحثي XAI المهتمين بجعل نماذج الذكاء الاصطناعي الخاصة بهم أكثر جدارة بالثقة، وكذلك الباحثين من التخصصات الأخرى الذين يبحثون عن طرق XAI فعالة لإكمال المهام بثقة أثناء توصيل المعنى من البيانات.Translated Description (French)
L'intelligence artificielle (IA) est actuellement utilisée dans un large éventail d'applications sophistiquées, mais les résultats de nombreux modèles d'IA sont difficiles à comprendre et à faire confiance en raison de leur nature de boîte noire. Habituellement, il est essentiel de comprendre le raisonnement derrière la prise de décision d'un modèle d'IA. Ainsi, le besoin de méthodes d'IA eXplainable (XAI) pour améliorer la confiance dans les modèles d'IA s'est fait sentir. XAI est devenu un sujet de recherche populaire dans le domaine de l'IA ces dernières années. Les documents d'enquête existants ont abordé les concepts de XAI, ses termes généraux et ses méthodes d'explicabilité post-hoc, mais aucun examen n'a porté sur les méthodes d'évaluation, les outils disponibles, les ensembles de données XAI et d'autres aspects connexes. Par conséquent, dans cette étude approfondie, nous fournissons aux lecteurs un aperçu de la recherche actuelle et des tendances dans ce domaine en émergence rapide avec un exemple d'étude de cas. L'étude commence par expliquer le contexte de XAI, les définitions communes et résumer les techniques récemment proposées dans XAI pour l'apprentissage automatique supervisé. La revue divise les techniques XAI en quatre axes à l'aide d'un système de catégorisation hiérarchique : (i) explicabilité des données, (ii) explicabilité du modèle, (iii) explicabilité post-hoc et (iv) évaluation des explications. Nous introduisons également des métriques d'évaluation disponibles ainsi que des packages et des ensembles de données open source avec des orientations de recherche futures. Ensuite, l'importance de l'explicabilité en termes d'exigences juridiques, de points de vue des utilisateurs et d'orientation des applications est décrite, appelée préoccupations XAI. Cet article préconise d'adapter le contenu de l'explication à des types d'utilisateurs spécifiques. Un examen des techniques et de l'évaluation de la XAI a été mené en examinant 410 articles critiques, publiés entre janvier 2016 et octobre 2022, dans des revues réputées et en utilisant un large éventail de bases de données de recherche comme source d'information. L'article s'adresse aux chercheurs de XAI qui souhaitent rendre leurs modèles d'IA plus fiables, ainsi qu'aux chercheurs d'autres disciplines qui recherchent des méthodes XAI efficaces pour effectuer des tâches en toute confiance tout en communiquant le sens des données.Translated Description (Spanish)
La inteligencia artificial (IA) se está utilizando actualmente en una amplia gama de aplicaciones sofisticadas, pero los resultados de muchos modelos de IA son difíciles de comprender y confiar debido a su naturaleza de caja negra. Por lo general, es esencial comprender el razonamiento detrás de la toma de decisiones de un modelo de IA. Por lo tanto, ha surgido la necesidad de métodos de IA explicables (XAI) para mejorar la confianza en los modelos de IA. XAI se ha convertido en un tema de investigación popular dentro del campo de la IA en los últimos años. Los documentos de la encuesta existentes han abordado los conceptos de XAI, sus términos generales y los métodos de explicación post-hoc, pero no ha habido ninguna revisión que haya analizado los métodos de evaluación, las herramientas disponibles, los conjuntos de datos de XAI y otros aspectos relacionados. Por lo tanto, en este estudio exhaustivo, proporcionamos a los lectores una descripción general de la investigación actual y las tendencias en esta área que emerge rápidamente con un ejemplo de estudio de caso. El estudio comienza explicando los antecedentes de XAI, las definiciones comunes y resumiendo las técnicas propuestas recientemente en XAI para el aprendizaje automático supervisado. La revisión divide las técnicas de XAI en cuatro ejes utilizando un sistema de categorización jerárquico: (i) explicabilidad de los datos, (ii) explicabilidad del modelo, (iii) explicabilidad post-hoc y (iv) evaluación de las explicaciones. También presentamos las métricas de evaluación disponibles, así como los paquetes y conjuntos de datos de código abierto con futuras direcciones de investigación. Luego, se describe la importancia de la explicabilidad en términos de demandas legales, puntos de vista del usuario y orientación de la aplicación, denominadas preocupaciones XAI. Este documento aboga por adaptar el contenido de la explicación a tipos de usuarios específicos. Se realizó un examen de las técnicas y la evaluación de XAI observando 410 artículos críticos, publicados entre enero de 2016 y octubre de 2022, en revistas de renombre y utilizando una amplia gama de bases de datos de investigación como fuente de información. El artículo está dirigido a investigadores de XAI que estén interesados en hacer que sus modelos de IA sean más confiables, así como a investigadores de otras disciplinas que buscan métodos XAI efectivos para completar tareas con confianza mientras comunican el significado de los datos.Files
1-s2.0-S1566253523001148-main.pdf.pdf
Files
(8.1 MB)
| Name | Size | Download all |
|---|---|---|
|
md5:30332a78f8a90ca7b258709eda4429e7
|
8.1 MB | Preview Download |
Additional details
Additional titles
- Translated title (Arabic)
- الذكاء الاصطناعي القابل للتفسير (XAI): ما نعرفه وما تبقى لتحقيق الذكاء الاصطناعي الجدير بالثقة
- Translated title (French)
- Intelligence artificielle explicable (XAI) : ce que nous savons et ce qui reste pour atteindre une intelligence artificielle digne de confiance
- Translated title (Spanish)
- Inteligencia Artificial Explicable (XAI): Lo que sabemos y lo que queda para lograr una Inteligencia Artificial Confiable
Identifiers
- Other
- https://openalex.org/W4366262984
- DOI
- 10.1016/j.inffus.2023.101805
References
- https://openalex.org/W1034019924
- https://openalex.org/W104847522
- https://openalex.org/W141083031
- https://openalex.org/W1497605902
- https://openalex.org/W1501005121
- https://openalex.org/W1546644035
- https://openalex.org/W1660264423
- https://openalex.org/W1727290854
- https://openalex.org/W1751198022
- https://openalex.org/W1772653913
- https://openalex.org/W1787224781
- https://openalex.org/W1789187189
- https://openalex.org/W1821462560
- https://openalex.org/W1851970720
- https://openalex.org/W1901365844
- https://openalex.org/W1927620212
- https://openalex.org/W1966087765
- https://openalex.org/W1967847904
- https://openalex.org/W1968843430
- https://openalex.org/W1971078475
- https://openalex.org/W1971882425
- https://openalex.org/W1975675278
- https://openalex.org/W1981856831
- https://openalex.org/W1988160630
- https://openalex.org/W1989968174
- https://openalex.org/W1990088911
- https://openalex.org/W1993572674
- https://openalex.org/W2000257498
- https://openalex.org/W2001468398
- https://openalex.org/W2013787102
- https://openalex.org/W2021191744
- https://openalex.org/W2026161499
- https://openalex.org/W2026805626
- https://openalex.org/W2029440072
- https://openalex.org/W2034637247
- https://openalex.org/W2036363240
- https://openalex.org/W2036779186
- https://openalex.org/W2041280856
- https://openalex.org/W2044140042
- https://openalex.org/W2046945713
- https://openalex.org/W2053075547
- https://openalex.org/W2054640944
- https://openalex.org/W2058051343
- https://openalex.org/W2058486605
- https://openalex.org/W2060277733
- https://openalex.org/W2063046703
- https://openalex.org/W2063689928
- https://openalex.org/W2069466022
- https://openalex.org/W2070427365
- https://openalex.org/W2083620785
- https://openalex.org/W2083844448
- https://openalex.org/W2084341220
- https://openalex.org/W2084701050
- https://openalex.org/W2088203270
- https://openalex.org/W2090026354
- https://openalex.org/W2092493343
- https://openalex.org/W2096434649
- https://openalex.org/W2096674496
- https://openalex.org/W2100171245
- https://openalex.org/W2108331388
- https://openalex.org/W2110842818
- https://openalex.org/W2111406701
- https://openalex.org/W2112649152
- https://openalex.org/W2118341388
- https://openalex.org/W2119858020
- https://openalex.org/W2125847307
- https://openalex.org/W2129639577
- https://openalex.org/W2131251708
- https://openalex.org/W2133606111
- https://openalex.org/W2134584261
- https://openalex.org/W2134682899
- https://openalex.org/W2135090752
- https://openalex.org/W2140553307
- https://openalex.org/W2142148616
- https://openalex.org/W2142910214
- https://openalex.org/W2145470394
- https://openalex.org/W2150165932
- https://openalex.org/W2150662287
- https://openalex.org/W2154157725
- https://openalex.org/W2158585626
- https://openalex.org/W2158599326
- https://openalex.org/W2159750251
- https://openalex.org/W2163532598
- https://openalex.org/W2164878629
- https://openalex.org/W2166305015
- https://openalex.org/W2170018669
- https://openalex.org/W2186091528
- https://openalex.org/W2187089797
- https://openalex.org/W2191167248
- https://openalex.org/W2195388612
- https://openalex.org/W2240067561
- https://openalex.org/W2251148681
- https://openalex.org/W2255700062
- https://openalex.org/W2259088801
- https://openalex.org/W2264742718
- https://openalex.org/W2282821441
- https://openalex.org/W2295107390
- https://openalex.org/W2295680652
- https://openalex.org/W2309231107
- https://openalex.org/W2318937711
- https://openalex.org/W2327262288
- https://openalex.org/W2343061342
- https://openalex.org/W2367397349
- https://openalex.org/W2394669110
- https://openalex.org/W2398344594
- https://openalex.org/W2420245003
- https://openalex.org/W2475287302
- https://openalex.org/W2479935243
- https://openalex.org/W2481995907
- https://openalex.org/W2493343568
- https://openalex.org/W2504241250
- https://openalex.org/W2536379393
- https://openalex.org/W2536885573
- https://openalex.org/W2542889218
- https://openalex.org/W2551974706
- https://openalex.org/W2553981914
- https://openalex.org/W2559655401
- https://openalex.org/W2576489192
- https://openalex.org/W2587299461
- https://openalex.org/W2588936615
- https://openalex.org/W2592141406
- https://openalex.org/W2594475271
- https://openalex.org/W2596567068
- https://openalex.org/W2601530120
- https://openalex.org/W2606882704
- https://openalex.org/W2607162077
- https://openalex.org/W2607223307
- https://openalex.org/W2610600445
- https://openalex.org/W2657631929
- https://openalex.org/W2725284386
- https://openalex.org/W2740486019
- https://openalex.org/W2742340224
- https://openalex.org/W2751298778
- https://openalex.org/W2752194699
- https://openalex.org/W2752332392
- https://openalex.org/W2753415590
- https://openalex.org/W2754395205
- https://openalex.org/W2762968537
- https://openalex.org/W2764024122
- https://openalex.org/W2768346313
- https://openalex.org/W2769506033
- https://openalex.org/W2783643842
- https://openalex.org/W2786672974
- https://openalex.org/W2787546749
- https://openalex.org/W2787666871
- https://openalex.org/W2788403449
- https://openalex.org/W2789070602
- https://openalex.org/W2792126732
- https://openalex.org/W2792641098
- https://openalex.org/W2793165286
- https://openalex.org/W2793566095
- https://openalex.org/W2795530988
- https://openalex.org/W2795559220
- https://openalex.org/W2802093198
- https://openalex.org/W2807828695
- https://openalex.org/W2810511505
- https://openalex.org/W2810611310
- https://openalex.org/W2869036160
- https://openalex.org/W2891503716
- https://openalex.org/W2892104346
- https://openalex.org/W2899682268
- https://openalex.org/W2902476877
- https://openalex.org/W2910028250
- https://openalex.org/W2910705748
- https://openalex.org/W2911227954
- https://openalex.org/W2913323966
- https://openalex.org/W2913693384
- https://openalex.org/W2914854991
- https://openalex.org/W2916797267
- https://openalex.org/W2916904544
- https://openalex.org/W2917767525
- https://openalex.org/W2918497321
- https://openalex.org/W2939861854
- https://openalex.org/W2942157335
- https://openalex.org/W2942444880
- https://openalex.org/W2943207870
- https://openalex.org/W2945976633
- https://openalex.org/W2951025380
- https://openalex.org/W2952602496
- https://openalex.org/W2954312269
- https://openalex.org/W2956281901
- https://openalex.org/W2958514452
- https://openalex.org/W2959325723
- https://openalex.org/W2960259221
- https://openalex.org/W2962772482
- https://openalex.org/W2962790223
- https://openalex.org/W2962793481
- https://openalex.org/W2962843949
- https://openalex.org/W2962858109
- https://openalex.org/W2962884579
- https://openalex.org/W2963095307
- https://openalex.org/W2963125461
- https://openalex.org/W2963374347
- https://openalex.org/W2963382180
- https://openalex.org/W2963609017
- https://openalex.org/W2963798744
- https://openalex.org/W2963852406
- https://openalex.org/W2963917042
- https://openalex.org/W2964010828
- https://openalex.org/W2964060211
- https://openalex.org/W2964231383
- https://openalex.org/W2964277809
- https://openalex.org/W2964303497
- https://openalex.org/W2966362896
- https://openalex.org/W2968511602
- https://openalex.org/W2969224142
- https://openalex.org/W2969670093
- https://openalex.org/W2970245929
- https://openalex.org/W2971201880
- https://openalex.org/W2972854536
- https://openalex.org/W2973631113
- https://openalex.org/W2979273157
- https://openalex.org/W2981646383
- https://openalex.org/W2981731882
- https://openalex.org/W2984353433
- https://openalex.org/W2984593886
- https://openalex.org/W2991138092
- https://openalex.org/W2994881943
- https://openalex.org/W2995188596
- https://openalex.org/W2995523160
- https://openalex.org/W2995739402
- https://openalex.org/W2996514457
- https://openalex.org/W2997560917
- https://openalex.org/W2999615587
- https://openalex.org/W3000716014
- https://openalex.org/W3000965188
- https://openalex.org/W3005340895
- https://openalex.org/W3006064320
- https://openalex.org/W3013308250
- https://openalex.org/W3014140399
- https://openalex.org/W3015977534
- https://openalex.org/W3023001449
- https://openalex.org/W3026235644
- https://openalex.org/W3027472889
- https://openalex.org/W3035253074
- https://openalex.org/W3035517615
- https://openalex.org/W3036453007
- https://openalex.org/W3045238219
- https://openalex.org/W3048102985
- https://openalex.org/W3048549109
- https://openalex.org/W3083483195
- https://openalex.org/W3087507349
- https://openalex.org/W3088014761
- https://openalex.org/W3088836971
- https://openalex.org/W3088909400
- https://openalex.org/W3093718041
- https://openalex.org/W3093861859
- https://openalex.org/W3096301653
- https://openalex.org/W3096580368
- https://openalex.org/W3100035092
- https://openalex.org/W3100711616
- https://openalex.org/W3101366597
- https://openalex.org/W3101981467
- https://openalex.org/W3111875263
- https://openalex.org/W3112230406
- https://openalex.org/W3116073702
- https://openalex.org/W3116131084
- https://openalex.org/W3116286104
- https://openalex.org/W3119566555
- https://openalex.org/W3123487819
- https://openalex.org/W3124474911
- https://openalex.org/W3125069671
- https://openalex.org/W3125145238
- https://openalex.org/W3125997628
- https://openalex.org/W3128172745
- https://openalex.org/W3128845074
- https://openalex.org/W3130784583
- https://openalex.org/W3131457744
- https://openalex.org/W3132091045
- https://openalex.org/W3134165704
- https://openalex.org/W3135028703
- https://openalex.org/W3136603773
- https://openalex.org/W3137125108
- https://openalex.org/W3138247573
- https://openalex.org/W3144635834
- https://openalex.org/W3150957536
- https://openalex.org/W3157438020
- https://openalex.org/W3159960173
- https://openalex.org/W3160005967
- https://openalex.org/W3164846705
- https://openalex.org/W3165174133
- https://openalex.org/W3165596704
- https://openalex.org/W3173849000
- https://openalex.org/W3176482836
- https://openalex.org/W3176714582
- https://openalex.org/W3177765094
- https://openalex.org/W3181414820
- https://openalex.org/W3182546273
- https://openalex.org/W3183198896
- https://openalex.org/W3186197713
- https://openalex.org/W3186608063
- https://openalex.org/W3188740654
- https://openalex.org/W3189411510
- https://openalex.org/W3190528386
- https://openalex.org/W3195567601
- https://openalex.org/W3195625625
- https://openalex.org/W3196674861
- https://openalex.org/W3197347140
- https://openalex.org/W3199384803
- https://openalex.org/W3199693875
- https://openalex.org/W3200028717
- https://openalex.org/W3206150780
- https://openalex.org/W3209901185
- https://openalex.org/W3210179814
- https://openalex.org/W3211108032
- https://openalex.org/W3212368439
- https://openalex.org/W3214172001
- https://openalex.org/W3214404698
- https://openalex.org/W3214886350
- https://openalex.org/W39725046
- https://openalex.org/W4200545033
- https://openalex.org/W4206060008
- https://openalex.org/W4206210655
- https://openalex.org/W4211007335
- https://openalex.org/W4211254520
- https://openalex.org/W4221159886
- https://openalex.org/W4224035735
- https://openalex.org/W4224984798
- https://openalex.org/W4225150645
- https://openalex.org/W4234836739
- https://openalex.org/W4237501282
- https://openalex.org/W4237711877
- https://openalex.org/W4237795037
- https://openalex.org/W4241307704
- https://openalex.org/W4241536761
- https://openalex.org/W4241832412
- https://openalex.org/W4243742982
- https://openalex.org/W4244296837
- https://openalex.org/W4245152641
- https://openalex.org/W4246698901
- https://openalex.org/W4246855866
- https://openalex.org/W4249103910
- https://openalex.org/W4250852845
- https://openalex.org/W4253050311
- https://openalex.org/W4282045566
- https://openalex.org/W4285138359
- https://openalex.org/W4285292424
- https://openalex.org/W4285605356
- https://openalex.org/W4285719527
- https://openalex.org/W4286432986
- https://openalex.org/W4287071543
- https://openalex.org/W4287101833
- https://openalex.org/W4287120154
- https://openalex.org/W4287243489
- https://openalex.org/W4287552728
- https://openalex.org/W4287690455
- https://openalex.org/W4288044972
- https://openalex.org/W4288109975
- https://openalex.org/W4288281523
- https://openalex.org/W4289123308
- https://openalex.org/W4289258965
- https://openalex.org/W4289699961
- https://openalex.org/W4290091102
- https://openalex.org/W4290716979
- https://openalex.org/W4293187584
- https://openalex.org/W4293583991
- https://openalex.org/W4294152847
- https://openalex.org/W4295808856
- https://openalex.org/W4296437236
- https://openalex.org/W4296978576
- https://openalex.org/W4298051456
- https://openalex.org/W4298084898
- https://openalex.org/W4298135002
- https://openalex.org/W4298154995
- https://openalex.org/W4298325995
- https://openalex.org/W4299700722
- https://openalex.org/W4299960064
- https://openalex.org/W4300071243
- https://openalex.org/W4300416216
- https://openalex.org/W4301886741
- https://openalex.org/W4302311461
- https://openalex.org/W4303520951
- https://openalex.org/W4307212054
- https://openalex.org/W4307715223
- https://openalex.org/W4308211566
- https://openalex.org/W4309730097
- https://openalex.org/W4311007707
- https://openalex.org/W4312078640
- https://openalex.org/W4312933868
- https://openalex.org/W4315652684
- https://openalex.org/W4321473693
- https://openalex.org/W4321786089