Ethical Concerns of Generative AI and Mitigation Strategies: A Systematic Mapping Study
It addresses ethical challenges in deploying LLMs, which is crucial for developers and policymakers, but it is incremental as it synthesizes existing research without introducing new methods or solutions.
This paper systematically reviewed 39 studies to identify and categorize ethical concerns of generative AI, particularly LLMs, and their mitigation strategies, finding that these concerns are multi-dimensional and context-dependent, with significant challenges remaining in implementation, especially in high-stake domains like healthcare and public governance.
[Context] Generative AI technologies, particularly Large Language Models (LLMs), have transformed numerous domains by enhancing convenience and efficiency in information retrieval, content generation, and decision-making processes. However, deploying LLMs also presents diverse ethical challenges, and their mitigation strategies remain complex and domain-dependent. [Objective] This paper aims to identify and categorize the key ethical concerns associated with using LLMs, examine existing mitigation strategies, and assess the outstanding challenges in implementing these strategies across various domains. [Method] We conducted a systematic mapping study, reviewing 39 studies that discuss ethical concerns and mitigation strategies related to LLMs. We analyzed these ethical concerns using five ethical dimensions that we extracted based on various existing guidelines, frameworks, and an analysis of the mitigation strategies and implementation challenges. [Results] Our findings reveal that ethical concerns in LLMs are multi-dimensional and context-dependent. While proposed mitigation strategies address some of these concerns, significant challenges still remain. [Conclusion] Our results highlight that ethical issues often hinder the practical implementation of the mitigation strategies, particularly in high-stake areas like healthcare and public governance; existing frameworks often lack adaptability, failing to accommodate evolving societal expectations and diverse contexts.