Bias-Aware Systems: Exploring Indicators for the Occurrences of Cognitive Biases when Facing Different Opinions

Document Title

Bias-Aware Systems: Exploring Indicators for the Occurrences of Cognitive Biases when Facing Different Opinions

Document Information

• Subject Area: Human-Computer Interaction and Cognitive Psychology, Cognitive Biases and Their Detection
• Keywords: Bias-aware systems, cognitive bias, cognitive perception systems, fNIRS, eye tracking, electrodermal activity (EDA)

Research Background and Issues

• Problems and Challenges:

  1. Cognitive biases significantly impact information evaluation and decision-making abilities, yet individuals are often unaware of their presence.
  2. Phenomena like information cocoons and filter bubbles, driven by biases, facilitate the spread of misinformation.
  3. Reliably quantifying cognitive biases through behavioral and physiological signals is challenging, while existing self-report methods may be influenced by bias and memory limitations.

• Significance:

  • The presence of biases can affect individual opinion formation and lead to erroneous decisions, such as attitudes toward vaccines or climate change.
  • Understanding the mechanisms and quantification of cognitive biases is fundamental for designing effective interventions to reduce the impact of misinformation and political polarization.

• Research Motivation and Related Work:

  • Previous studies have shown that information selection behaviors (e.g., selective exposure) are driven by biases, but experimental results are often contradictory (e.g., differences in processing time).
  • Physiological signals (e.g., fNIRS, EDA) offer a potential, objective method to study biases, but research on their specific manifestations and quantification is limited.

Proposed Solution

• Research Methods:

  • Introduce the concept of bias-aware systems, which detect cognitive biases through behavioral and physiological data.
  • Design two experiments using statements congruent or dissenting with participants' viewpoints, recording interaction data, eye-tracking data, and physiological signals (e.g., fNIRS and EDA).

• Innovations:

  1. Utilize multimodal data (interaction behavior + physiological data) to explore the physiological and behavioral manifestations of cognitive biases.
  2. Integrate diverse technologies (fNIRS, EDA, and eye tracking) to quantify biases during information processing.
  3. Investigate the moderating effects of interest and familiarity on biases.

• Experiments and Implementation:

  • The first experiment exposed participants to content congruent or dissenting with their stance through text and image stimuli.
  • The second experiment expanded the topic range, using only text stimuli and introducing inter-stimulus intervals (ISI) to reduce interference with physiological signals.
  • Collected metrics included self-reports (agreement with content, sharing intention), behavioral data (time spent), and physiological signals (e.g., brain oxygenation levels, electrodermal activity).

Research Findings

• Specific Findings:

  1. Behavioral Performance:
     • Participants spent more time on dissenting information but reported lower cognitive effort.
     • This phenomenon may indicate that conflicting information increases cognitive load but lacks deep processing.
  2. Physiological Signals:
     • For participants with lower interest, significant changes in brain oxygenation levels (detected via fNIRS) were observed when exposed to dissenting information.
     • Trends in EDA data suggested higher physiological arousal, but due to statistical insignificance, further validation is required.

• Advantages Over Existing Solutions:

  • Proposed the concept of bias-aware systems and combined multimodal detection methods using behavioral and physiological signals.
  • Enhanced experimental design for cognitive bias detection by incorporating variables such as interest and familiarity.

• Limitations and Future Directions:

  1. Limitations:
     • The sample primarily consisted of participants with progressive viewpoints, with insufficient representation of conservative participants.
     • Most stimuli were sourced from global websites, potentially lacking validity in the Australian context.
     • Self-report results may be influenced by biases and other factors.
  2. Future Directions:
     • Increase participant diversity, particularly balancing different political stances.
     • Explore more complex psychological or cognitive indicators (e.g., multi-region fNIRS data).
     • Investigate practical applications of bias-aware systems, especially in preventing misinformation dissemination.