The Impact of Algorithmic Transparency on User Experience and Physiological Responses

Academic Background

With the rapid development of Affective Computing technology, Affect-aware Task Adaptation systems have gradually become a research hotspot. These systems recognize users’ psychological states through various measurements (e.g., physiological signals, facial expressions) and adjust computer-based tasks accordingly to optimize user experience. For instance, a system can dynamically adjust game difficulty based on the user’s emotions or modify task complexity according to cognitive load. While previous studies have shown that improving the accuracy of psychological state recognition and task adaptation can significantly enhance user experience, the impact of Algorithmic Transparency on user experience remains underexplored. Algorithmic transparency refers to the degree to which users understand the computer’s decision-making process. Higher transparency may make users more tolerant of system errors and even encourage them to adjust their behavior to compensate for systematic mistakes. However, this theory has yet to be experimentally validated in the field of affective computing.

To address this gap, a research team from University of Cincinnati and University of Wyoming designed a study to explore how algorithmic transparency influences user experience and physiological responses. This study not only focuses on subjective experiences such as interest, stress, and perceived competence but also evaluates objective reactions through physiological metrics like respiration rate, skin conductance, and heart rate.

Source of the Paper

The research was conducted by Mohammad Sohorab Hossain, Joshua D. Clapp, and Vesna D. Novak and published in IEEE Transactions on Affective Computing in 2025. The study was funded by the National Science Foundation (NSF).

Research Process and Results

1. Research Design

The research team developed an affect-aware task adaptation system with four levels of algorithmic transparency: None, Low, Medium, and High. A total of 93 students and staff members from University of Cincinnati were recruited for the experiment. Participants were randomly assigned to low, medium, or high transparency groups, with 31 participants in each group.

The experiment consisted of two phases: - Phase 1: All participants first experienced task adaptation under no transparency for 16 minutes. - Phase 2: Participants then experienced task adaptation under one of the three transparency conditions (low, medium, or high) for another 16 minutes.

2. Task and Transparency Design

The study used OpenMATB (an open-source implementation of NASA’s Multi-Attribute Task Battery) as the experimental task. The task included three subtasks: system monitoring, tracking, and communication. Task difficulty was divided into 10 levels, and the system adjusted difficulty dynamically based on users’ physiological responses (respiration rate). However, to control experimental conditions, the system did not actually adapt based on physiological signals but followed predefined rules to simulate 66.7% adaptation accuracy.

The four levels of transparency were designed as follows: - No Transparency: Users had no knowledge of how the system adjusted difficulty. - Low Transparency: The system displayed users’ difficulty adjustment requests and actual adjustment results. - Medium Transparency: The system not only showed the adjustment results but also explained the reasons (e.g., “Respiration rate is low, so difficulty is increased”). - High Transparency: The system provided detailed explanations, including specific numerical ranges of respiration rates.

3. Data Collection and Analysis

After each phase, participants completed the Intrinsic Motivation Inventory (IMI) and the NASA Task Load Index (TLX) to assess their subjective experiences. Additionally, the research team recorded physiological data, including respiration rate, skin conductance, and heart rate.

Data analysis employed a mixed 3 (transparency group) × 2 (trial phase) analysis of variance (ANOVA) to evaluate the impact of transparency on user experience and physiological responses.

4. Key Results

Subjective Experience

• Interest/Enjoyment and Perceived Competence were significantly higher in the medium and high transparency groups compared to the low transparency group. This indicates that providing algorithmic transparency information can significantly enhance subjective user experience.
• NASA TLX scores and Effort/Importance showed no significant differences across transparency groups, suggesting limited influence of transparency on perceived task load.

Physiological Responses

• Several physiological indicators (e.g., respiration rate, heart rate) decreased significantly in Phase 2, likely due to increased familiarity with the task.
• The long-term impact of transparency levels on physiological responses was not significant, indicating that transparency contributes relatively little to the user’s overall mental state.

Respiration Response to Erroneous Adjustments

The study found that when the system incorrectly adjusted difficulty, changes in respiration rate were significantly greater in the high transparency group than in the low transparency group. This suggests that users may attempt to influence system decisions by adjusting their breathing after understanding system errors.

Conclusions and Implications

The study demonstrates that algorithmic transparency can significantly improve subjective user experience, particularly in terms of interest/enjoyment and perceived competence. Although the long-term impact of transparency on physiological responses is limited, it may trigger active compensatory behaviors in specific contexts (e.g., during system errors). This finding provides important insights for the design of affective computing systems: by providing simple transparency information, developers can significantly enhance user experience without requiring extensive hardware or software improvements.

Research Highlights

1. Filling a Research Gap: The first systematic study on the impact of algorithmic transparency on user experience in affect-aware task adaptation systems.
2. Multi-dimensional Evaluation: Combining subjective experience and physiological response data for a comprehensive assessment of transparency’s effects.
3. Practical Application Value: The results indicate that providing transparency information is a low-cost, high-benefit design strategy applicable to areas such as educational games and adaptive automation.

Future Research Directions

The research team recommends that future studies further optimize experimental designs, for example: - Collecting user ratings immediately after each phase to reduce recall bias. - Designing more incentive-driven scenarios to motivate users to actively compensate for system errors. - Increasing the frequency and predictability of difficulty adjustments to help users better identify and compensate for system errors.

This study lays the foundation for research on algorithmic transparency in affective computing and provides important references for future system design.