Abstract
It is known that when interacting with explainable autonomous systems, user characteristics are important in determining the most appropriate explanation, but understanding which user characteristics are most relevant to consider is not simple. This paper explores such characteristics and analyses how they affect the perceived usefulness of four types of explanations based on the robot's mental states. These types are belief, goal, hybrid (goal and belief) and baseline explanations. In this study, the explanations were evaluated in the context of a domestic service robot. The user characteristics considered are the perception of the robot's rationality and autonomy, the acceptance of the robot and the user's cognitive tendencies. We found differences in perceived usefulness between explanation types based on user characteristics, with hybrid explanations being the most useful.
Introduction
As autonomous robots increasingly transition into domestic environments, the necessity for transparency becomes paramount. For a domestic service robot to be trusted and effective, it must be able to explain its actions—for instance, why it has suddenly begun preparing a meal at an unconventional time. While the field of Explainable AI (XAI) has developed numerous technical frameworks for generating such explanations, the human-centric side of the equation remains under-explored. This research addresses the critical gap between technical explanation generation and the psychological reception of those explanations, investigating how individual user characteristics—such as personality, cognitive style, and attitudes toward technology—influence the perceived usefulness of robot explanations.
Methodology
The study utilized a quantitative online survey involving 468 participants. Participants were presented with 28 distinct actions performed by a domestic service robot, each paired with one of four explanation types derived from the robot's internal states:
- Belief: Focuses on the robot's knowledge or environmental perception (e.g., "It was close to a usual mealtime").
- Goal: Focuses on the intended outcome or objective (e.g., "To feed the user").
- Hybrid: A combination of both belief and goal states.
- Baseline: A control condition that merely restates the action without providing reasoning.
To generate these explanations, the researchers employed a Belief-Desire-Intention (BDI) goal tree hierarchy, which mirrors human "folk psychology" to make the robot's reasoning more intuitive to laypeople.
User characteristics were measured using validated psychological scales, including the Need for Cognition (measuring preference for effortful thinking) and the Analysis-Holism Scale (measuring whether individuals focus on isolated objects or broader contextual interactions).
Results and Discussion
The Hybrid Advantage
The primary finding of the study is a clear preference for multi-faceted explanations. Approximately 64% of users rated hybrid explanations as the most useful. By providing both the "what I know" (belief) and the "what I want" (goal), the robot allows the user to construct a more comprehensive mental model of its decision-making process. Hybrid explanations achieved the highest median usefulness score (42), significantly outperforming the baseline (20).
The Influence of User Perception and Anxiety
Regression analysis revealed that a user's prior perception of the robot significantly colors their evaluation of its explanations. Users who perceived the robot as rational and autonomous were far more likely to find reason-based explanations (Belief and Goal) useful. Specifically, Perception of Rationality was a strong predictor for the usefulness of Belief ($B = .837, p < .001$) and Goal ($B = .813, p = .001$) explanations.
Conversely, "robot anxiety" acted as a significant barrier. Users with high levels of anxiety toward robots rated all reason-based explanations as less useful (e.g., $\Delta$Hybrid $B = -1.433, p = .018$). This suggests that for anxious users, additional information may be perceived as overwhelming, highlighting a need for simplified communication strategies for this demographic.
Cognitive Style and Causality
The study also found that cognitive style—specifically how individuals attribute causality—plays a role. "Interactionists," or those who naturally consider the interplay between an actor and their environment, found hybrid explanations particularly beneficial ($B = 1.482, p = .017$). This indicates that XAI systems may be more effective if they are tailored to match the user's inherent cognitive tendencies.
Conclusion
This research demonstrates that there is no universal "best" explanation for robot behavior. While hybrid explanations are generally preferred, their effectiveness is mediated by the user's psychological profile. Future work should address the limitations of this study, notably its online format and UK-centric sample. Replicating these findings in physical human-robot interaction settings and across diverse cultures will be essential for developing truly personalized and effective explainable autonomous systems.
Summary updated