[论文解读] Ironies of Generative AI: Understanding and mitigating productivity loss in human-AI interactions

Auste Simkute, Lev Tankelevitch+4 more|arXiv (Cornell University)|Feb 17, 2024

Ethics and Social Impacts of AI被引用 7

一句话总结

该论文在生成式AI应用中综合了四个生产力损失驱动因素——从生产到评估的转变、无效的工作流重组、干扰以及任务复杂性两极化——以人因学研究和生成式AI研究为基础，并提出缓解这些因素的设计方向。

ABSTRACT

Generative AI (GenAI) systems offer opportunities to increase user productivity in many tasks, such as programming and writing. However, while they boost productivity in some studies, many others show that users are working ineffectively with GenAI systems and losing productivity. Despite the apparent novelty of these usability challenges, these 'ironies of automation' have been observed for over three decades in Human Factors research on the introduction of automation in domains such as aviation, automated driving, and intelligence. We draw on this extensive research alongside recent GenAI user studies to outline four key reasons for productivity loss with GenAI systems: a shift in users' roles from production to evaluation, unhelpful restructuring of workflows, interruptions, and a tendency for automation to make easy tasks easier and hard tasks harder. We then suggest how Human Factors research can also inform GenAI system design to mitigate productivity loss by using approaches such as continuous feedback, system personalization, ecological interface design, task stabilization, and clear task allocation. Thus, we ground developments in GenAI system usability in decades of Human Factors research, ensuring that the design of human-AI interactions in this rapidly moving field learns from history instead of repeating it.

研究动机与目标

• Identify four broad categories of productivity loss in GenAI systems (production-to-evaluation shift, unhelpful workflow restructuring, interruptions, task-complex polarization).
• Ground these challenges in decades of Human Factors research on automation and relate them to GenAI studies across domains.
• Propose design directions informed by Human Factors (continuous feedback, personalization, ecological interface design, task stabilization, clear task allocation).
• Highlight the need for further research on situational awareness and cognitive workload in GenAI use.

提出的方法

• Synthesize insights from Human Factors literature on automation with recent GenAI user studies across domains (coding, writing, design, healthcare).
• Map observed GenAI usability challenges to four productivity-loss categories.
• Extract design implications from established HF principles (feedback, flexibility) to guide GenAI system design.

实验结果

研究问题

• RQ1 What are the main productivity-loss mechanisms observed in GenAI-human interactions?
• RQ2 How do these mechanisms relate to traditional ‘ironies of automation’ in HF research?
• RQ3 What HF-inspired design directions can mitigate productivity loss in GenAI systems?
• RQ4 In which domains do these challenges manifest beyond programming (e.g., writing, design, healthcare)?

主要发现

• Four key productivity-loss categories are identified: production-to-evaluation shift, unhelpful workflow restructuring, task interruptions, and task-complexity polarization.
• GenAI’s high output capacity and opacity contribute to reduced situational awareness and increased monitoring workload.
• Prompting and output adaptation introduce new tasks that can disrupt task sequences and raise cognitive load.
• Automation can lead to complacency and over-reliance, reducing robustness and increasing error risk.
• HF-inspired design directions include continuous feedback, system personalization, ecological interface design, task stabilization, and clear task allocation.
• The paper argues for leveraging decades of HF research to improve GenAI usability and productivity across domains.