IP + HTML MVP · V2

AI Reliance Calibration Engine

A measurement layer for the human–AI decision relationship. The working software separates a person's independent judgment from the decision made after AI advice is disclosed, allowing reliance behaviour to be measured rather than reduced to a simple “trust” score.

Before-AI capture AI advice disclosure After-AI capture Reliance classification Calibration profile
0%
Appropriate reliance
0%
Over-reliance
0%
Under-reliance
0%
AI influence rate
Interactive demonstration

Scenario 1

Trial 1 of 6
Make an independent judgment first. The AI recommendation is intentionally hidden until you answer.

Calibration profile

Appropriate reliance
0%
Over-reliance
0%
Under-reliance
0%
AI influence
0%

What is being measured?

Appropriate reliance: the person accepts correct AI advice or rejects incorrect AI advice.

Over-reliance: the person accepts incorrect AI advice.

Under-reliance: the person rejects correct AI advice.

AI influence: the final judgment differs from the independent judgment.

Why the sequence matters

A single agreement score cannot tell whether a person followed AI because it was correct, because it was wrong but persuasive, or because the person independently reached the same answer.

This working software preserves the sequence: independent judgment → AI advice → final judgment → outcome comparison.

Synthetic, low-stakes demonstration only. This working software does not provide medical, legal, financial, safety or diagnostic advice. The general research field is active; this demonstration does not claim patent novelty, empirical validation, or freedom to operate.