Population preferences for AI system features across eight different decision-making contexts

Area of Science:

• Human-Computer Interaction
• Artificial Intelligence Ethics
• Public Opinion Research

Background:

• Deep learning-based artificial intelligence (AI) systems are increasingly used as decision-support tools.
• Existing research on public preferences for AI, particularly in medicine, highlights the importance of features like explainability, accuracy, and human oversight.
• These 'protective' features are crucial for safeguarding user interests, but their context-specific relevance remains under-explored beyond medical applications.

Purpose of the Study:

• To investigate public preferences for five specific protective features of AI systems and their implementation.
• To examine how these preferences vary across eight diverse use cases in public and commercial sectors.
• To understand the influence of decision impact significance on the perceived importance of AI system features.

Main Methods:

• A cross-sectional survey study was conducted.
• The study involved the adult Danish population.
• Participants' preferences for five protective AI features were assessed across eight different use cases, from medical diagnostics to parking ticket issuance.

Main Results:

• All five protective features were consistently deemed important across all eight evaluated contexts.
• The perceived importance of these features slightly decreased when the implications of the AI-driven decision were less significant to respondents.
• This indicates a general demand for trustworthy AI features regardless of the application domain.

Conclusions:

• Public preferences for protective AI features are robust and extend beyond the medical domain.
• Contextual factors, particularly the significance of decision outcomes, influence the perceived importance of these features.
• Findings underscore the need for developing AI systems that incorporate user-centric protective features across all sectors.