Trust as a Function of Three Factors
Trust in AI systems can be modeled as:
T = α₁E + α₂P + α₃(1 − U)
Where:
- E = Explainability (can the system justify its outputs?)
- P = Performance history (has the system been reliable?)
- U = Uncertainty (how confident is the system?)
- α₁ + α₂ + α₃ = 1 (weights sum to 1)
Greater transparency, reliable past performance, and low uncertainty contribute to higher trust. This suggests three distinct intervention points for trust calibration in human-AI interfaces:
- Improve explainability mechanisms
- Surface performance track records
- Make uncertainty visible