Understanding Diversity in Human-AI Data: What Cognitive Style Disaggregation Reveals

Artificial Intelligence (AI) is becoming more pervasive through all levels of society, trying to help us be more productive. Research like Amershi et al.'s 18 guidelines for human-AI interaction aim to provide high-level design advice, yet little remains known about how people react to Applications or Violations of the guidelines. This leaves a gap for designers of human-AI systems applying such guidelines, where AI-powered systems might be working better for certain sets of users than for others, inadvertently introducing inclusiveness issues. To address this, we performed a secondary analysis of 1,016 participants across 16 experiments, disaggregating their data by their 5 cognitive problem-solving styles from the Gender-inclusiveness Magnifier (GenderMag) method and illustrate different situations that participants found themselves in. We found that across all 5 cogniive style spectra, although there were instances where applying the guidelines closed inclusiveness issues, there were also stubborn inclusiveness issues and inadvertent introductions of inclusiveness issues. Lastly, we found that participants' cognitive styles not only clustered by their gender, but they also clustered across different age groups.