Value-Sensitive Delegation in Everyday AI Agent Use: Evidence from OpenClaw
Using Value Sensitive Design, analyzed 73,093 Reddit posts about OpenClaw, highlighting the impact of user-set operational conditions on value fulfillment.
Key Findings
Methodology
The study employed Value Sensitive Design, analyzing 73,093 Reddit posts with LLM assistance to identify 21 human values grouped into six categories, such as Autonomous, Dependable, and Affordable Operation. The focus was on the impact of user-set operational conditions on value fulfillment.
Key Results
- In the Autonomous Operation group, 77.3% of values were met, whereas only 42.8% were met in the Equitable Access group. Resource accounting had the lowest fulfillment rate at 35.3%.
- Values were generally met when users described what the agent delivered in five of six groups, and mostly unmet when users described supervising it in all six groups.
- Value fulfillment clustered around user-set operational conditions rather than the agent's outputs.
Significance
The study underscores the importance of considering not only what an AI agent accomplishes but also the conditions users set around delegation, such as cost, access, and oversight. This finding is crucial for designing and evaluating AI agents, particularly in aligning technology with human values.
Technical Contribution
The research provides an empirical analysis of human value fulfillment in AI agent use, introducing the concept of value-sensitive delegation and emphasizing the significance of user-set operational conditions, challenging traditional task completion evaluation methods.
Novelty
This is the first systematic application of Value Sensitive Design to analyze user experiences with AI agents, revealing the critical impact of user-set operational conditions on value fulfillment, contrasting sharply with traditional task completion evaluation methods.
Limitations
- The study is limited to data from the Reddit platform, which may not represent all user experiences.
- Value fulfillment measurement relies on user self-reports, lacking objective verification.
- Did not deeply explore specific needs differences among diverse user groups.
Future Work
Future research could extend to other social media platforms, verify value fulfillment differences among diverse user groups, and explore how to better support these values in design.
AI Executive Summary
As users increasingly delegate tasks to autonomous AI agents, evaluations typically focus on task completion rather than the values users prioritize. This paper analyzes 73,093 Reddit posts about using OpenClaw through Value Sensitive Design, identifying 21 human values grouped into six categories. The study finds that value fulfillment is often linked to the operational conditions set by users rather than the agent's outputs. In five of six groups, values were generally met when users described what the agent delivered, while in all six groups, values were mostly unmet when users described supervising it. This pattern is conceptualized as value-sensitive delegation, indicating that supporting human values requires attention to the conditions users set around delegation, including cost, access, and oversight. The findings are significant for designing and evaluating AI agents, especially in ensuring technology aligns with human values.
The study employs Value Sensitive Design, analyzing 73,093 Reddit posts with LLM assistance to identify 21 human values grouped into six categories, such as Autonomous, Dependable, and Affordable Operation. The focus was on the impact of user-set operational conditions on value fulfillment. Results show that in the Autonomous Operation group, 77.3% of values were met, whereas only 42.8% were met in the Equitable Access group. Resource accounting had the lowest fulfillment rate at 35.3%.
The study underscores the importance of considering not only what an AI agent accomplishes but also the conditions users set around delegation, such as cost, access, and oversight. This finding is crucial for designing and evaluating AI agents, particularly in aligning technology with human values.
Deep Dive
Abstract
Users increasingly delegate work to autonomous AI agents, yet evaluations typically measure task completion rather than the values users prioritize. Using Value Sensitive Design, we analyzed, with LLM assistance, 73,093 first-person Reddit posts about using OpenClaw, each for its human value, agent aspect, value fulfillment, and user outcome. The 21 values form six value groups, including Autonomous, Dependable, and Affordable Operation, Bounded Reach, Reviewability, and Equitable Access. Relative to each aspect's corpus share, values clustered not at the agent's outputs but at the operating conditions users set around a run. Values were usually met where users described what the agent delivered, in five of six groups, and mostly unmet where users described supervising it, in all six groups. We conceptualize this pattern as value-sensitive delegation. Supporting human values requires attention not only to what an agent accomplishes, but to the conditions users set around delegation, including cost, access, and oversight.