An Activity System-based Perspective of Generative AI: Challenges and Research Directions

Abstract

With its remarkable ability to generate content, generative artificial intelligence (GAI) has been recognized as a milestone in the development of artificial general intelligence. To understand the challenges, potential impact, and implications associated with GAI, we adopt a socio-technical perspective to analyze them. First, we identify the key characteristics of GAI, which include content generation, generalization ability, and reinforcement learning based on human feedback. Next, we address technological, ethical, societal, economic, regulatory, and governance challenges. Finally, we deploy activity theory to explore research directions in GAI. Research questions that warrant further investigation include how GAI may impact the future of work, how GAI can collaborate effectively with humans, and how we can improve the transparency of GAI models as well as mitigate biases and misinformation in GAI to achieve ethical and responsible GAI.