Anthropomorphic expression must not exaggerate understanding

What it is

Truthful anthropomorphic communication keeps first-person, emotional, and understanding claims within actual perception, reasoning, memory, and action. “I understand how you feel” carries a stronger pragmatic commitment than “I detected slower speech”; fluent language does not establish the former capability.

Why it happens

People infer common ground and mental state from conversational conduct. Confirmation, empathy, and memory claims trigger those inferences even when the system has only keywords or short context. Users then issue more implicit commands, disclose more, or monitor less, turning linguistic overstatement into reliance risk.

Studying it

Certain, anthropomorphic, and mechanism-grounded wording can be varied at fixed capability. Scenario-specific predictions of understanding, memory, and responsibility, plus disclosure, command complexity, and monitoring, are more diagnostic than a generic trust score. Longitudinal work tests whether failures correct the model.

Where it stops holding

Friendly natural language is not inherently deceptive; the issue is whether its reasonable implication is delivered. Dense mechanical disclaimers can impair coordination and be ignored. Pragmatic force varies by culture and setting, warranting conservative wording for high stakes or vulnerable users.

Applying it

• Map every first-person claim to testable sensing, memory, or decision competence; remove unsupported understanding and emotion claims.
• Before consequential delegation, state what information was and was not obtained.
• Ask first-time users what implicit information the robot can handle after hearing a response; measure overestimation, disclosure, and reduced monitoring.

Related

• Same group: X3.03.2 Pretending to understand is worse than admitting failure · X3.03.3 Uncertainty should be expressed when present
• Adjacent: X1.05 Choosing a degree of anthropomorphism · X6.07 Ethical limits of robot anthropomorphism
• Search terms: truthful anthropomorphism · capability transparency · calibrated language