When humans and large language models collaborate, problem-finding illuminates

This study explores the role of Large Language Models (LLMs) in problem finding (PF) for ill-structured, wicked, and multi-stakeholder problems—an essential yet underexamined aspect of organizational innovation. While prior research has examined artificial intelligence (AI) in problem-solving (PS) and, more recently, the contributions of LLMs, their role in PF remains largely unexplored. Given that PF lays the foundation for effective PS, overlooking it can result in missed opportunities and inefficient resource allocation, ultimately hindering the innovation process. Drawing on a cognitive-behavioral perspective rooted in Simon’s foundational work, this study identifies the key activities and cognitive skills essential for PF and examines how human-LLM collaboration can enhance this process. While humans possess innate PF abilities, cognitive limitations such as bounded rationality, satisficing, and uncertainty avoidance constrain their effectiveness. LLMs, with their advanced reasoning and dataprocessing capabilities, can help overcome these constraints by expanding the search space, generating alternative problem framings, and stimulating creativity. However, their inherent limitations, including biases, hallucinations, and challenges in handling less structured problems, necessitate a structured approach to human-LLM collaboration. To address this, we propose a framework that defines this interaction and illustrates its application through case studies in product development and social innovation. Our findings have significant implications for organizations, emphasizing the need for structured implementation, workforce training, and AI governance. We conclude with research propositions to guide future investigations into humans-LLMs collaboration in PF, positioning it as a critical driver of innovation in the era of Generative AI.