# Explainable Generative AI (GenXAI): A Survey, Conceptualization, and Research Agenda

**Authors:** Johannes Schneider  
**arXiv:** [2404.09554](https://arxiv.org/abs/2404.09554)  
**Date:** 2024-04-15  
**Summary page:** https://www.summarizepaper.com/en/paper/2404.09554v1/

## Key points

- Generative AI (GenAI) has revolutionized artificial intelligence by shifting focus to generating solutions
- Importance of explainability in AI (XAI) in the era of GenAI, presenting unique challenges
- Novel desiderata for explanations: verifiability, interactivity, security, cost considerations
- Conducting a narrative review and drawing insights from existing surveys on XAI and GenAI
- Developing a taxonomy to categorize existing XAI techniques based on key dimensions
- Emphasizing interactive adjustment of complex outputs generated by AI systems for user evaluation and refinement
- Identifying gaps in existing literature related to XAI for GenAI and proposing future research directions

## Summary

Generative AI (GenAI) has revolutionized the field of artificial intelligence by shifting its focus from recognizing patterns to generating solutions across a wide range of tasks. This shift has opened up new opportunities and challenges for explainability in AI (XAI). In this research manuscript, we explore the significance of XAI in the era of GenAI and discuss the unique challenges it presents. We also introduce novel desiderata that explanations should meet, including verifiability, interactivity, security, and cost considerations. To address these issues, we conduct a narrative review and draw upon insights from existing surveys on XAI and GenAI. By synthesizing knowledge from various sources, we aim to provide a comprehensive understanding of the current landscape of XAI for GenAI. Our approach includes developing a taxonomy to categorize existing XAI techniques based on key dimensions that distinguish them. We also highlight the technical backgrounds and motivations for XAI in GenAI. Additionally, we emphasize the need for interactive adjustment of complex outputs generated by AI systems as they become more integrated into society and require user input for evaluation and refinement. Furthermore, we identify gaps in existing literature related to XAI for GenAI and propose a research agenda outlining future directions for investigation. Through addressing these gaps and providing a holistic view of XAI in the context of GenAI, our manuscript aims to serve as a valuable resource for both technical experts and researchers from interdisciplinary fields. Overall, our research contributes to advancing the understanding of explainability in AI systems powered by generative technologies. Through our detailed analysis and taxonomy development, we hope to pave the way for further advancements in XAI for GenAI applications.