Reported trust varies with graded value alignment in AI-attributed economic-environmental choices

Area of Science:

• Human-computer interaction
• Artificial intelligence ethics
• Decision science

Background:

• Artificial intelligence (AI) systems are increasingly involved in decisions with ethical implications.
• Understanding how users perceive and trust AI based on value alignment is crucial.
• Existing theories suggest value similarity fosters trust, but the impact of graded alignment needs further investigation.

Purpose of the Study:

• To investigate how objectively measured, graded value alignment influences user trust in AI decision-making.
• To examine the relationship between perceived value alignment and self-reported trust in AI.
• To explore the effect of decision orientation (economy vs. environment) on trust.

Main Methods:

• A controlled economic-environmental trade-off task was conducted with 250 participants.
• Participants' subjective equivalence points were calibrated using matching and titration tasks.
• Participants observed AI choices varying in objective value alignment and decision orientation, then reported trust.

Main Results:

• Self-reported trust in the AI advisor increased monotonically as AI choices became more aligned with participants' calibrated trade-off points.
• Perceived value alignment also increased with objective alignment and was strongly correlated with trust.
• A slight, less consistent trust advantage was observed for environment-leaning AI choices.

Conclusions:

• Objective value alignment serves as a graded cue influencing user trust in AI decision support.
• Perceived value alignment is closely linked to trust, highlighting its importance in human-AI interaction.
• The study provides empirical evidence for the graded nature of trust formation in AI based on value congruence.