Data-driven psychophysical methods to diversify SIAs and address bias

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence
• Social Robotics

Background:

• Socially Interactive Agents (SIAs) often exhibit White- and Western-centric biases.
• These biases limit SIAs' ability to interpret and express social cues across diverse cultures.
• Current SIAs struggle with effective engagement due to a lack of cultural adaptability.

Purpose of the Study:

• To address limitations in current SIAs' cross-cultural social cue interpretation and expression.
• To propose a data-driven method for creating culturally adaptive and inclusive SIAs.
• To enhance user engagement and trust in SIAs by grounding them in user-specific models.

Main Methods:

• Utilizing the data-driven psychophysical method of reverse correlation.
• Modeling users' perceptual expectations, preferences, and sociocultural norms.
• Integrating user insights into the design of SIA appearance and expressive behavior.

Main Results:

• Demonstrated the potential of reverse correlation for modeling user-specific social expectations.
• Showcased how this method enables SIAs to exhibit psychologically grounded social signals.
• Provided examples of culturally adaptive and ethnically inclusive SIA designs.

Conclusions:

• Reverse correlation offers a viable approach to developing culturally sensitive SIAs.
• Empirically derived user models can improve SIA engagement, trust, and inclusivity.
• This approach contributes to mitigating algorithmic bias and real-world prejudice.