Emotional expression must match the robot's actual state

What it is

Robot affect should be grounded in verifiable system or interaction state — hesitation for low confidence, a calming tone when attention is needed — rather than implying an unsupported inner feeling. This is affective transparency: not a claim that the robot experiences emotion, but a requirement that its expression be a reliable signal to the user.

Why it happens

People read face, voice, and posture as cues to intent and capability and use them to decide what to do next — a "struggling" expression prompts an inference that the task has hit trouble, that help is needed, or that caution is warranted. Expression has to track real state precisely because, once it decouples from state (smiling through a failure, say), the user's judgments based on that expression become systematically wrong — this is not just an authenticity complaint, it is the signal's functional value being destroyed, since the user keeps relying on it even as it starts lying persistently. Conversely, when expression tracks confidence, action, and recoverability, it compresses a complex internal state into something the user can read at a glance, cutting interpretive cost.

Studying it

Studies can manipulate expression–state congruence (matched vs. mismatched) and measure state-recognition accuracy, next-action prediction based on the expression, calibrated trust, recovery speed, and subjective liking. A critical condition to include is one where the expression logic is correct but the underlying state sensor itself is wrong — this separates "the expression design is broken" from "the upstream state estimate is broken" as two distinct failure modes. Subscales of the Godspeed questionnaire (perceived intelligence, perceived safety) can serve as auxiliary measures, but a questionnaire score is not a substitute for measuring actual state-judgment accuracy driven by the expression — mere liking does not establish that the expression carries real informational value.

Where it stops holding

Not every product needs a strict mapping from true state to expressed affect — a purely entertainment-oriented robot's expression may itself be the entertainment content rather than a state signal, which is fine as long as the user understands that; this boundary only applies where users actually use the expression as a basis for decisions. Some expressions also regulate interaction rather than report internal state — a polite smile does not correspond to an internal "I am happy" state — and interpretation of the same expression varies across culture, age, and neurodiversity, so no single expression vocabulary can be assumed universal. No single expression should ever be the sole carrier of safety-critical information.

Applying it

First enumerate the robot's real internal or functional states (normal execution, obstacle encountered, low battery, recognition failure), then design a corresponding expression, a user-facing action, and a disable condition for each — rather than building an appealing expression library first and mapping states onto it afterward. Pair anything consequential with text or functional feedback rather than letting expression carry it alone. Verify by auditing a random sample of run logs for mismatches where state was A but the expressed affect showed B, and by specifically testing contradictory-state, low-confidence, and recovery scenarios to check whether users' interpretations track actual capability.

Related

• Same group: X6.03.2 False emotional expression constitutes deception · X6.03.3 Long-term companionship raises attachment ethics
• Nearby: X6.01 Social responses to robots · X6.07 Ethical limits of robot anthropomorphism
• Search terms: affective transparency · Godspeed questionnaire · anthropomorphism · human-robot interaction