# AI and Trust How trust develops between humans and AI systems. The core challenge is trust calibration: neither over-trusting (blind acceptance of AI output) nor under-trusting (ignoring AI despite good output). **Over-trust** leads to errors propagating silently. When people accept AI output without verification, [[AI Hallucination]] becomes invisible. [[AI Sycophancy]] compounds this; the AI tells you what you want to hear, and you believe it because you trust it. Over-trust is the default failure mode for people impressed by AI capabilities. **Under-trust** wastes the investment. When people ignore or second-guess every AI output, they spend more time verifying than they save. Under-trust is the default failure mode for people burned by early AI failures or those who have not updated their mental model of current capabilities. Trust builds through four mechanisms: 1. **Transparency**: AI explains its reasoning. When you can see the chain of thought, you can evaluate the logic, not just the conclusion. 2. **Track record**: Consistent quality over time. Trust is earned through repeated accurate performance in your specific use cases. 3. **Verifiability**: You can check the output against sources. Trust requires the ability to verify, even if you do not verify every time. 4. **Reversibility**: Mistakes can be undone. Trust is easier to grant when errors are recoverable. This connects to [[Human-in-the-Loop]] as a safety mechanism. AI and Trust is connected to but different from [[AI Safety]]. Safety is about preventing catastrophic outcomes. Trust is about the working relationship between human and AI. [[AI Evaluation]] provides the metrics for calibrating trust. [[Cognitive debt]] describes what accumulates when trust is miscalibrated in either direction. ## References ## Related - [[AI Safety]] - [[AI Hallucination]] - [[AI Sycophancy]] - [[Human-in-the-Loop]] - [[Cognitive debt]] - [[AI Evaluation]] - [[Human-AI Collaboration Patterns]]