What you want from a trustworthy AI depends on whether it's working alone or with people.

Trusting artificial intelligence is complex because what people need from a system changes based on the situation. Yet, we know little about how everyday people view these needs. Researchers surveyed nonexpert participants to find what makes AI trustworthy in social versus nonsocial situations, comparing their answers to human trust. They discovered that protecting private data is always vital for AI, no matter the context. Beyond privacy, context matters. In nonsocial settings, people trust AI if its decisions are clear and easy to interpret. In social settings, they care more about collaboration, good communication, a positive public image, and an attractive look. Human trust showed similar shifts, with one key difference: people always demand collaboration and communication skills to trust humans across every context, whereas for AI, those social skills only matter in social settings.

Why it matters

As AI moves into homes, hospitals, and public spaces, people's trust depends on what the AI is doing. This study shows that in task-oriented settings, people care most about transparency and explainability, but in social settings, they want the AI to be collaborative and have a good appearance. Designers can now tailor AI to match context-specific trust expectations, which could improve adoption and acceptance.

Questions answered