Abstract
As robots become increasingly common in a wide variety of domains—from military and scientific applications to entertainment and home use—there is an increasing need to define and assess the trust humans have when interacting with robots. In human interaction with robots and automation, previous work has discovered that humans often have a tendency to either overuse automation, especially in cases of high workload, or underuse automation, both of which can make negative outcomes more likely. Furthermore, this is not limited to naive users, but experienced ones as well. Robotics brings a new dimension to previous work in trust in automation, as they are envisioned by many to work as teammates with their operators in increasingly complex tasks. In this chapter, our goal is to highlight previous work in trust in automation and human-robot interaction and draw conclusions and recommendations based on the existing literature. We believe that, while significant progress has been made in recent years, especially in quantifying and modeling trust, there are still several places where more investigation is needed.
Introduction: From Tools to Teammates
As robots transition from industrial tools to autonomous teammates in high-stakes fields like search-and-rescue, medicine, and the military, the success of these missions increasingly depends on the human-robot relationship. Unlike traditional automation, modern robotics requires a nuanced understanding of trust. The core challenge facing the field is the "maladaptive interaction" problem, where trust is not properly calibrated to the robot's actual capabilities.
The Problem of Miscalibration
Trust calibration is the process of ensuring a human’s trust level matches the robot’s functional reliability. When trust is misaligned, two primary issues arise:
- Disuse (Under-trust): Users may ignore or bypass a robot's capabilities, wasting time and resources by performing tasks manually that the robot could handle more effectively.
- Misuse (Over-trust): Also known as automation bias, this occurs when users follow a robot blindly, ignoring red flags or system failures, which can lead to catastrophic accidents.
Frameworks for Understanding and Measuring Trust
To address these challenges, researchers utilize several key frameworks and metrics:
The Three-Dimensional Model
Following the work of Lee and See (2004), trust is often categorized into three dimensions:
- Purpose: The user’s understanding of why the system was created.
- Process: The user’s understanding of how the system operates.
- Performance: The system’s actual reliability and competence.
Quantifying Trust
Measuring trust requires both psychometric and computational tools. The Trust Perception Scale-HRI, developed through data from 580 participants, uses factor analysis to break trust down into capability, behavior, task, and appearance. Additionally, cross-cultural studies involving over 2,400 participants across the US, Taiwan, and Turkey have highlighted how cultural backgrounds influence a user's propensity to trust automated systems.
On the technical side, "Inverse Trust" models are being developed to allow robots to mathematically estimate a human's trust level in real-time, allowing the machine to adapt its behavior to improve the team's performance.
Key Drivers of Trust in HRI
Reliability and Timing
System reliability remains the strongest predictor of trust. However, the timing of errors is critical; a "primacy-recency" bias exists where errors occurring at the very beginning or the very end of a task damage trust more significantly than those occurring in the middle.
The Anthropomorphism Trap
Humans naturally attribute intent and personality to robots that look or act human. While anthropomorphism can make a robot more likable, it often leads to unwarranted trust. Users may assume a robot is more capable than it actually is simply because it possesses human-like features, such as arms or fluent speech.
The Simulation Gap
Physical presence and perceived risk drastically alter trust behavior. In fire evacuation studies, participants in real-world experiments followed "defective" robots even when they knew the robots were wrong, whereas participants in virtual simulations were much more skeptical. This suggests that trust in HRI is deeply tied to the physical environment and the stakes of the interaction.
Limitations and Future Directions
Despite significant progress in modeling trust, several gaps remain in the current literature:
- Unified Definitions: The field lacks a universally accepted definition of trust, making it difficult to compare results across different studies.
- Longitudinal Research: Most HRI studies are short-term interactions. There is a critical need for longitudinal studies to observe how trust evolves over weeks or months of daily use.
- Bidirectional Trust: Future research should move beyond how humans trust robots and begin exploring how robots can assess the reliability and state of their human partners.
By addressing these gaps, we can move closer to a future where humans and robots operate as truly effective, well-calibrated teams.
Summary updated