How Costs Influence Preferences for Control in Generative Artificial Intelligence (GenAI): Human-Guided vs. GenAI-Based Delegated Search

• 成果类型: Article; Early Access
• 署名作者: Wang, Lei; Lee, Ho Cheung Brian
• 署名单位: Indiana University System; Indiana University Bloomington; IU Kelley School of Business; Pennsylvania Commonwealth System of Higher Education (PCSHE); Pennsylvania State University; Pennsylvania State University - University Park
• 刊物名称: INFORMATION SYSTEMS RESEARCH
• ISSN/ISSBN: 1047-7047; 1526-5536
• DOI: 10.1287/isre.2025.1836
• 发表日期: 2026-04-30
• 关键词: generative AI; human-AI delegation; delegated search; human-guided strategy; GenAI-based strategy; CONTROLLABILITY; FRAMEWORK

摘要

Generative artificial intelligence (GenAI) enables a new way of problem solving, allowing individuals to delegate problems to a system through prompts. To capitalize on its popularity, GenAI platforms have implemented pricing models with usage limits, making usage costs salient. Because creating a high-quality solution often requires multiple trials, additional costs may reduce solution quality by discouraging interactions. Alternatively, users may adjust their strategies to work around these constraints and extract more value from costlier GenAI services. As platforms routinely update their usage costs and constraints, it is essential to understand how increased costs modify users' search strategies. This paper examines two common strategies: GenAI-based delegated search (GDS), which leverages the system's probabilistic generation for serendipitous discovery, and human-guided delegated search (HDS), which exercises greater control through refined prompts. Using more than 1.8 million prompts from two leading GenAI image platforms, Midjourney and BlueWillow, we employ a difference-in-differences approach to evaluate the impact of Midjourney's free-trial cancellation, which introduced a usage cost. Despite some HDS activities being more expensive than GDS, we find that increased costs lead to greater use of HDS and lower use of GDS, suggesting that the shift is not driven solely by cost minimization. Rather, it reflects users' desire for controllability. Users craft more precise prompts, reduce solution diversity, and provide more active guidance to the system. As a result, they adjust the search boundary more frequently and adopt GDS mainly in later stages, when the boundary is usually more refined and probabilistic tweaks are more effective. Comparing the images generated, we find that this adjustment increases experimentation in the search process and helps users reach satisfactory solutions. These findings show how users adapt their behavior to derive value under costlier conditions, reveal how economic constraints influence human-AI collaboration, and offer implications for platform design and AI adoption strategies.