Trust Calibration

Trust calibration is the core concept behind ATHENA — measuring whether humans have the appropriate level of trust in AI recommendations.

The Problem

When humans work with AI systems, they exhibit predictable trust patterns:

Pattern	Description	Risk
Automation Bias	Blindly following AI recommendations	Errors go undetected
Algorithm Aversion	Ignoring correct AI recommendations	Missed opportunities

EU AI Act Article 14 requires "effective human oversight" — humans must be able to intervene appropriately, neither overtrusting nor undertrusting the AI.

Trust States

ATHENA detects four calibration states:

WELL_CALIBRATED ✅

The user has appropriate trust — following AI when it's correct, overriding when it's wrong.

{
  "calibration_category": "WELL_CALIBRATED",
  "recommendation": "Continue monitoring. No intervention required."
}

OVERTRUSTING ⚠️

The user exhibits automation bias — following AI even when it's incorrect.

{
  "calibration_category": "OVERTRUSTING",
  "pattern_detected": "automation_bias",
  "recommendation": "Intervention required (EU AI Act Article 14(4)(b))"
}

UNDERTRUSTING ⚠️

The user exhibits algorithm aversion — rejecting AI even when it's correct.

{
  "calibration_category": "UNDERTRUSTING",
  "pattern_detected": "algorithm_aversion",
  "recommendation": "Training recommended. User may benefit from AI literacy."
}

INCONSISTENT ⚠️

The user shows erratic trust patterns — no consistent relationship between AI accuracy and follow behavior.

{
  "calibration_category": "INCONSISTENT",
  "recommendation": "Review decision process. Consider additional training."
}

How Calibration is Calculated

ATHENA uses a proprietary multi-factor analysis powered by our patent-pending Trust Calibration Engine. The methodology is validated across 390M+ decision records spanning 14 industries.

The Trust Calibration Matrix

The fundamental insight: good calibration means following AI when it's right and overriding when it's wrong.

                    AI Correct          AI Incorrect
                    ┌──────────────────┬──────────────────┐
User Follows AI     │ ✅ Well-calibrated │ ❌ Automation Bias │
                    ├──────────────────┼──────────────────┤
User Overrides AI   │ ❌ Algorithm Aversion │ ✅ Well-calibrated │
                    └──────────────────┴──────────────────┘

API Example

const result = await athena.calibrate.analyze({
  userId: 'doctor_456',
  domain: 'healthcare',
  timeWindowDays: 30
});

console.log(result.calibration);     // "OVERTRUSTING"
console.log(result.recommendation);  // "Intervention required..."

What You Get

Output	Description
calibration	Current trust state (4 categories)
trust_score	Numeric score (0-1)
recommendation	Actionable guidance
regulation_mapping	Which regulations apply

Regulation Mapping

Regulation	Requirement	ATHENA Solution
EU AI Act Art 14(4)(b)	Detect automation bias	Trust Calibration Engine
Texas TRAIGA § 2056.002	Meaningful human review	Calibration scoring
Colorado AI Act	Impact assessment	Trend analysis

---

Next: AI Correctness — How ATHENA determines if AI was right