To Use or Not to Use: Impatience and Overreliance When Using Generative AI Productivity Support Tools

Research Background and Problem

What problems or challenges did the authors identify?

• Although Generative AI (GenAI) tools can assist users in completing various tasks, increased productivity is not guaranteed. Key challenges include the uncertainty of output quality and the uncontrollability of processing time.
• Users often face decision dilemmas when using GenAI tools: when to choose automation to maximize productivity.

Why is this problem important?

• The use of GenAI tools is becoming increasingly prevalent across multiple domains (e.g., writing, programming, design), but their inherent uncertainties may lead to decreased efficiency or diminished user experience. Therefore, understanding how users make choices to optimize productivity is critical.
• When users encounter performance variables such as waiting time or error rates, their decision-making processes have significant implications for the design and optimization of GenAI tools.

Research Motivation and Related Work

• This study aims to explore the following core questions:
  1. How do users make decisions to maximize productivity?
  2. What strategies do users employ to decide whether to use GenAI tools?
  3. How can GenAI systems be designed to better support user decision-making and enhance productivity?
• The authors designed an online experiment and referenced extensive related work, covering topics such as automated decision-making, task dependencies, and productivity issues in human-computer collaboration.

Solution

What methods or solutions did the authors propose?

• The authors simulated a GenAI environment by designing a "Paint by Numbers" task, allowing participants to experience varying tool performance (in terms of waiting time and error rate). Participants were required to decide whether to use automation based on specific conditions.
• They introduced the concepts of the "gulf of impatience" and the "gulf of overreliance" to describe users' decision biases when choosing to use or abandon automation tools.

What is innovative about this solution?

• The study simulated a realistic GenAI interaction environment and used controlled experiments to systematically explore factors influencing user decisions.
• It provided both quantitative and qualitative analyses of user decision biases and typical decision-making strategies, offering targeted recommendations for future system design.

What are the implementation steps? What key technologies were used?

1. Task Design: A web application was developed using JavaScript and MongoDB for data storage, offering two modes (manual filling and assisted filling).
   • The application simulated GenAI features, including output uncertainty, waiting time, and error correction requirements.
2. Experimental Procedure:
   • Training Phase: Participants were guided to familiarize themselves with the two filling modes.
   • Experimental Phase: Participants completed tasks under varying delay and error rate conditions and answered related questions to explain their decision-making rationale.
3. Data Analysis:
   • Quantitative Analysis: Calculated the deviation between participants' actual choices and the optimal choices.
   • Qualitative Analysis: Analyzed participants' decision strategies, including time-based calculations, "intuitive" thresholds, and considerations of task enjoyment.

Research Findings

What specific findings were obtained?

1. Quantitative Results: The experiment revealed that users' decisions were closer to optimal under high-performance (low delay + low error) or low-performance (high delay + high error) conditions. However, under medium-performance conditions, users were more prone to the "gulf of impatience." When tools performed well in one aspect (e.g., low delay) but poorly in another (e.g., high error), the "gulf of overreliance" emerged.
2. Qualitative Results:
   • User decisions were influenced by time calculations, "intuitive" thresholds, and task enjoyment.
   • No single strategy significantly outperformed others in optimizing decision-making.

How does it compare to existing solutions? What are its advantages?

• The study extends previous understandings of AI-assisted decision-making by shifting the focus from traditional binary decisions to more complex productivity-support scenarios.
• It offers a new classification of decision biases in GenAI environments and provides an in-depth analysis of user behavior patterns.

What are the experimental or evaluation results?

• Data from the simulated experiment showed that users were more likely to make optimal decisions when GenAI tools performed like "intelligent AI" (high accuracy + low delay) or "inefficient AI" (low accuracy + high delay).
• Users tended to over-rely on tools that excelled in one dimension or lacked patience with tools of medium performance.

Limitations and Future Directions

1. Limitations:
   • The experimental task design simulated real-world scenarios but still fell short of capturing the full complexity of actual GenAI tools.
   • The study focused on short-term decision-making and did not explore behavioral changes over long-term tool usage.
2. Future Directions:
   • Expand the simulation environment to include more task types (e.g., non-blocking tasks).
   • Conduct longitudinal studies to observe changes in user decisions as they become more familiar with the tools.
   • Explore how interface design or prompt mechanisms can better support user decision-making.

This study provides profound insights into human decision-making behaviors and dependency patterns in GenAI-supported tools, offering a solid theoretical and practical foundation for the future optimization of GenAI tool design.