Agentic data science with the Data Agent Kit
cloud.google.com·8h ago
AI Research
arXiv CS.AI·Aug 4·5 min read
Trust and Its Betrayal under Three Representational Strategies
TL;DR
Trust is analyzed as a subjunctive epistemic state, revealing how betrayal affects trust representations differently. Three models—lexicographic assumption, ordinal closeness, and strong belief—show varied impacts of betrayal on trust dynamics.
✦ Why It Matters
Evaluate your trust models in AI systems to better handle potential betrayals and improve reliability.
Key Takeaways
- Understand trust as a subjunctive state to improve AI interactions.
- Differentiate between trust representations to assess betrayal impacts.
- Implement trust models that account for varying betrayal effects.
- Evaluate trustworthiness assumptions in AI systems for better outcomes.
How It Works
The study develops three distinct models of trust: the lexicographic assumption focuses on hierarchical trust levels, ordinal closeness measures the proximity of trust relationships, and strong belief maintains a coherent belief structure despite betrayal. Each model provides a unique lens through which to analyze how trust is built and eroded.